DECUServe Journal May 1995
Brian McMahon, Info-VAX Refugee
The DECUServe Journal
=====================
May, 1995
From the Editor's Keyboard
==========================
Welcome to the post-symposium edition of the DECUServe Journal.
Your faithful editors are still digging their way out from under
massive piles of information and accreted e-mail. Isn't it
wonderful to be away from the office for a week? Isn't it wretched
to have to catch up when you get back?
Of course, DECUServe never sleeps. There are always more worthy
articles waiting to be published. And so, on we go.
Incidentally, as we never tire of pointing out, we're always
delighted to hear what you think. We welcome comments, suggestions,
gripes, Pulitzer nominations, et cetera. You can reach us by
sending e-mail to DECUServe-Jo...@DECUS.Org, or by any of
the several ways listed in the "Contact Information" section at the
end of each issue.
Table of Contents
=================
CONTENTS
From the Editor's Keyboard . . . . . . . . . . . . . 1
Table of Contents . . . . . . . . . . . . . . . . . 1
2-Node Cluster Quorum . . . . . . . . . . . . . . . 2
Clearing the Typeahead Buffer . . . . . . . . . . . 9
DSNlink Contract Renewal Problems . . . . . . . . 12
Passive FTP . . . . . . . . . . . . . . . . . . . 15
TeamLinks MIME Support . . . . . . . . . . . . . . 21
IP Subnetting Help . . . . . . . . . . . . . . . . 23
UART Accelerator . . . . . . . . . . . . . . . . . 29
Assorted Tidbits . . . . . . . . . . . . . . . . . 32
About the DECUServe Journal . . . . . . . . . . . 36
Contact Information . . . . . . . . . . . . . . . 36
The DECUServe Journal May, 1995 Page 2
2-Node Cluster Quorum
2-Node Cluster Quorum
=====================
Here's a guaranteed way to start a rhubarb in a VMS-oriented
technical discussion -- bring up quorum settings in a two-node
cluster. This stream also includes a discussion of the merits of
Digital's Distributed File System.
Participants: Gus Altobello, Kevin Angley, John Briggs, Rob Brooks,
David Campen, Dale Coy, Bob Forrest, Bob Hassinger, Larry Kilgallen,
Frank Nagy, Beth Romine, Ray Whitmer, Glenn Zorn.
Note 2025.28, 30-MAR-1995
Romine: What about quorum disk
------------------------------
I'm considering creating a cluster of 1 Vax 4300 and 1 Alpha 3400. The
AXP does not have a DSSI. Will it be possible to have a quorum disk in
this situation? Or will I need to add DSSI to the 3400?
Note 2025.29, 30-MAR-1995
Brooks: well . . .
------------------
Perhaps I'm wrong, but the quorum concept exists to prevent partitioned
clusters, right? If you don't have any disks that are available
to both systems *when one system is down* then you can't run the risk
of trashing the file system. In other words, since you don't have
any CI (obviously), DSSI, or (it appears) SCSI disks that are
dual-ported, you should be able to set EXPECTED_VOTES to 1 and
be done, since node A can only write to node B's local disks
if node B is up (and vice versa).
If you happen to be shadowing, ignore the above!
I'm sure that if I'm wrong, the wrath will rain down quickly!
Note 2025.30, 30-MAR-1995
Coy: Sometimes
--------------
> Perhaps I'm wrong, but the quorum concept exists to prevent partitioned
> clusters, right?
The cluster quorum concept, yes. The quorum disk, no. The quorum disk
exists to provide another "voting member" to clusters that would
otherwise have an _even_ number of votes.
> If you happen to be shadowing, ignore the above!
Shadowing usually has no bearing on a clustering question. The only
place that it really matters is that you "can't" shadow a quorum disk.
The DECUServe Journal May, 1995 Page 3
2-Node Cluster Quorum
But it doesn't seem to be relevant to the question asked.
> dual-ported, you should be able to set EXPECTED_VOTES to 1 and
> be done, since node A can only write to node B's local disks
Well, pick one answer: (yes, no, it depends, you should be able but you
can't...).
> I'm considering creating a cluster of 1 Vax 4300 and 1 Alpha 3400. The
Beth, what are you trying to accomplish by creating such a cluster?
> AXP does not have a DSSI. Will it be possible to have a quorum disk in
To be USEFUL as a quorum disk, the disk must be directly accessible from
more than one node.
Under normal (non-failure) conditions, a "reasonable" cluster must have
an odd number of total votes present (1,3,5,7...).
With only 2 CPUs (and no quorum disk), you can achieve all cluster
functionality _but_ your CPU availability will be reduced (if one
crashes, the other hangs).
OR you do other things -- it all depends on "why" you are considering
clustering.
Note 2025.31, 30-MAR-1995
Angley: A six node cluster seems pretty reasonable to me
--------------------------------------------------------
Not sure I follow why an even number of votes greater than two is
"unreasonable" or begs for a quorum disk. The only thing I can think
that you mean is that quorum for an even number is no less than for the
next higher number, but I don't think that's unreasonable - just math.
Note 2025.32, 30-MAR-1995
Coy: OK, it's reasonable
-------------------------
OK, you said it much better than I did. :-)
[Note that I didn't say "begs for a quorum disk".]
It IS just math. A cluster with 4 votes can stand one failure, but not
two. A cluster with 3 votes can stand one failure, but not two (same
as above). A cluster with 4 votes has more things that _can_ fail,
therefore it _will_ have more failures (and reduced availability).
A cluster with 5 or 6 votes can stand two failures... etc.
Perhaps I should have said that a cluster with an even number of votes
The DECUServe Journal May, 1995 Page 4
2-Node Cluster Quorum
has a lower availability than a cluster with one more vote (an odd
number). Perhaps I shouldn't have used the word "unreasonable".
In other words, a quorum disk exists to provide higher availability of
the cluster, than that which would exist without the quorum disk.
That's true regardless of the number of nodes (there's nothing magic
about 2 or 3 nodes), but is most _obvious_ when you look at two versus
three nodes.
Example: If I have a 2-vote cluster,
o Adding 1 vote increases the availability.
o Adding another vote DEcreases availability (but not as low
as the original 2-vote cluster).
o Adding another vote increases the availability...
Of course, the objective of clustering is not _always_ to increase
availability. Or there may be multiple objectives with different
degrees of importance.
Note 2025.33, 31-MAR-1995
Brooks: I'll try again . . .
----------------------------
My point, which wasn't made too well, is that I think a partitioned
cluster can't cause any problems in a configuration where all
the disks are local to the systems (no CI, DSSI or SCSI [in VMS V6.2]).
If that's the case, then then you run without a quorum disk AND set
EXPECTED_VOTES to 1 AND have the cluster survive the loss of one node.
> Shadowing usually has no bearing on a clustering question. The only
> place that it really matters is that you "can't" shadow a quorum disk.
> But it doesn't seem to be relevant to the question asked.
Well, consider a two-node cluster each with its own system disk
and one other local disk. If you form a shadowset from the two
non-system disks AND the cluster partitioned such that each node
mounted the local disk into a shadowset without the other node's
local disk, you might have some merging issues. Again this
is predicated on the notion that you could safely run this
configuration with no quorum disk and EXPECTED_VOTES = 1.
Note 2025.34, 31-MAR-1995
Zorn: EXPECTED_VOTES only used on boot
--------------------------------------
The above will not work. EXPECTED_VOTES is only used when a node
joins/forms a cluster and not to determine if the cluster will stay
alive when a node leaves. There is no way to have a cluster dynamically
stay alive with two nodes having a vote of one each and no quorum disk.
It sounds like waiting for V6.2 and SCSI clustering might be the best,
most inexpensive, answer since that should allow for a quorum disk.
The DECUServe Journal May, 1995 Page 5
2-Node Cluster Quorum
Note 2025.35, 31-MAR-1995
Coy: Except...
--------------
> My point, which wasn't made too well, is that I think a partitioned
> cluster can't cause any problems in a configuration where all
> the disks are local to the systems (no CI, DSSI or SCSI [in VMS V6.2]).
That is entirely correct. Without question for a 2-node cluster, and
with only "theoretical-and-pathological" question for a cluster of more
than 2 active nodes (which might in _theory_ partition non-uniformly)
> If that's the case, then then you run without a quorum disk AND set
> EXPECTED_VOTES to 1
That is certainly possible.
> AND have the cluster survive the loss of one node.
Yes (sorta) and No. If each of the two nodes has a vote, then (as
pointed out in the previous reply) then cluster will not survive the
loss of either node. Also, if there is only one copy of the operating
system, the cluster cannot survive without whichever node owns that
disk.
If Node A has a vote and Node B does _not_ have a vote, then the
cluster will survive the loss of Node B, but will not survive the loss
of Node A. This is "supported" if Node B is a satellite node.
There are also unsupported alternatives, etc.
This whole discussion cannot be resolved until there is a clear
understanding of WHY! That is, there are several possible reasons for
wanting to cluster.
For instance, if one requirement is "no cost", and if it is acceptable
to have slightly reduced availability, and if the major thing that you
want is "easy system maintenance", then the master/slave solution is
quite OK.
> Well, consider a two-node cluster each with its own system disk
> and one other local disk. If you form a shadowset from the two
> non-system disks AND the cluster partitioned such that each node
> mounted the local disk into a shadowset without the other node's
> local disk, you might have some merging issues.
Absolutely. Which is why all "solutions" that might allow this are
"unsupported". Again, there's nothing inherently wrong with that -- it
all depends on your objectives.
The DECUServe Journal May, 1995 Page 6
2-Node Cluster Quorum
Note 2025.36, 2-APR-1995
Forrest: clusters designed for single security domain
-----------------------------------------------------
> My point, which wasn't made too well, is that I think a partitioned
> cluster can't cause any problems in a configuration where all
> the disks are local to the systems (no CI, DSSI or SCSI [in VMS V6.2]).
I guess you assume to have two copies of the SYSUAF and
VMSMAIL_PROFILE (one for each node). In which case password changes
and last login times on one node wouldn't be reflected on the other node,
and the mail count and any mail profile changes wouldn't be reflected on
the other node.
Note 2025.37, 3-APR-1995
Briggs: Can of worms
------------------------
And when you add/modify/remove a user/identifier in one file, you have to
remember to add/modify/remove the user/identifier in the other file as
well.
And when the user messes with their mail settings on one node, he's got
to log onto the other node and repeat the process.
And, if you want the queue database to stay up when the cluster partitions,
then you have to put a separate queue manager database on each node.
Under 5.5-2 and earlier, that means that the nodes cannot see each other's
queues. I don't know whether 6.x allows multiple cooperating queue
managers accessing independent databases -- it seems unlikely.
And, as has been hashed out elsewhere, the fact that you're not sharing
disks or doing host-based shadowing is _NOT_ a guarantee that you
are not doing other things that break when the cluster is partitioned.
It may be an adequate assurance, however, depending on what applications
you are running.
And, the only way to merge a partitioned cluster is to reboot half of it.
If you try to merge without rebooting, you'll get a CLUEXIT bugcheck.
Note 2025.38, 3-APR-1995
Hassinger:
-----------
Somewhere around here about six months ago more-or-less I asked a
similar question. In my case I only wanted to share disks in my
contemplated two node cluster - a VAX 3800 and an Alpha 3000. The two
systems would be operated as logically separate resources otherwise -
different UAFs, etc. I suggested that with no common path to the disks
(the VAX disks are all on its DSSI bus and the Alpha disks are all on
its SCSI bus), with the sharing being done via Ethernet, logic seemed
to me to suggest there "should" be some way to have either member be
able to survive when the other went down.
The DECUServe Journal May, 1995 Page 7
2-Node Cluster Quorum
There was a lot of discussion about why that could not be done. In the
main stream applications of clustering with common paths to disks it is
true. But eventually John Covert came to the discussion, got
interested, and looked into it. A while back he posted that he had
found a way to make it work in the special case I was talking about and
I am hoping it will make it into a future release of VMS.
Note 2025.39, 3-APR-1995
Coy: As usual, it depends...
----------------------------
Exactly!
Also, there are a few "unsupported" ways of configuring a cluster.
All of this is the reason I asked "why is clustering being considered?"
Without knowing the objectives, it's hard to evaluate potential
solutions.
For instance, if the primary objective is "reduced system management
because of a single copy of VMS", then one set of factors applies. If
the primary objective is "convenient file transfer", then another set
applies, etc.
Note 2025.40, 3-APR-1995
Altobello: (two node cluster) <> (two nodes sharing disks)
----------------------------------------------------------
If you only want to share disks, and not
- provide a single system management domain,
- share queues,
- provide higher availability to users on either system
or any one of a number of other reasons for clustering, then you may
want to look into the Distributed File Services software that DEC
sells.
If all you want to do is make the B-disks available to the A-system
users, and vice-versa, then get the software that does it and doesn't
complicate your lives with all that other stuff (like considering a
cluster a "single entity" and not something you'd just want to be
taking apart at random times).
This, of course, would cost license money and so may be out of the
question. But if you *ever* want to run differing versions of O/S or
layered products on the two nodes, then you'd better closely reexamine
your ideas about clustering these systems.
The DECUServe Journal May, 1995 Page 8
2-Node Cluster Quorum
Note 2025.41, 3-APR-1995
Campen: DFS doesn't need a cluster?
-----------------------------------
So DFS does not require a cluster? I had read the SPD once but wasn't
clear about this.
Note 2025.42, 3-APR-1995
Kilgallen: No cluster required
------------------------------
It does not require a cluster. If it did, there would be no content to DFS.
One thing you give up with DFS that you have with clusters is the ability to
share a file which is opened write access.
Compared to FAL (normal DECnet access), DFS is _very_fast_, because the
implementation does weird and wonderful things in kernel mode such as
requeueing IRPs back and forth between Disk and Network drivers.
Note 2025.43, 4-APR-1995
Nagy: DFS, reaching out and touching some disk
----------------------------------------------
As Larry said... Fermilab has on-site disk volumes that are served
world-wide using DFS - the Sun never sets on these volumes as they
are mounted in Japan, India and Italy. Even *I* am not crazy enough
to suggest setting up a cluster that "distributed".
Note 2025.44, 4-APR-1995
Whitmer: DFS seems less reliable than a cluster, for the record.
----------------------------------------------------------------
If a connection to a DFS node experiences a glitch, all open files on
the node become invalid. This can be interesting if you are in the
middle of writing the file. It can also be interesting for
applications (such as WordPerfect) which may open temporary files on a
DFS node, keep them open for an extended period of time, and expect to
write to them later. We spent days trying to work this one out with
DEC, who finally stated that instead of trying to reestablish the
connections to open files on the lost node they terminate the
connections.
I do not argue that DFS is wrong in terminating the connections
(although I did at the time...) or that it is worse than what a cluster
might do in certain cases (although I suspect it is), but I do know
that we received numerous support calls on this. DEC may have even
done something in the mean while to correct this.
The DECUServe Journal May, 1995 Page 9
2-Node Cluster Quorum
Note 2025.45, 5-APR-1995
Nagy: Major DFS problem fixed, root cause in VMS V6.1
-----------------------------------------------------
We have just recently gotten a fix for a nasty DFS problem (DFS V2.0
under OpenVMS VAX V6.1 - not clear if the Alpha had the same problem)
in which DFS would not re-open the network links. Normally, DFS is
supposed to re-open the network links (meanwhile stalling I/O with
MountVerify) but in this case we would get FILALRACC errors trying
to use DFS. We could fix things by doing a DFS$CONTROL STOP COMM
and then restarting but this would terminate all open file accesses.
After much effort by Digital (thanks to both CSC and DFS Engineering),
we now have a pre-release of the fix in place and DFS in once more
wonderfulness and happiness. Initially this problem was fixed in
the DFS$COM_ACP but the root cause was traced to VMS itself and
a new SECURITY.EXE loadable image provided which fixed the problem
of delivering ASTs to processes at inappropriate times (like when
ASTs were blocked or executing at AST level). Watch for a patch
for this problem to appear soonish on the official patch list.
Clearing the Typeahead Buffer
=============================
Sometimes we wonder if ANY of the DCL-hackery threads in the VMS
conference stay confined to their original topic. We're
exaggerating, of course, but there does seem to be a fair amount of
"topic roving" (one step beyond topic drift) in those notes. The
following discussion begins with the way to clear typeahead from a
command procedure, then continues to discuss RECALL/ERASE and SET
TERMINAL/INQUIRE.
Participants: John Briggs, Jim Campobello, Rick Carter, Keith
Chadwick, George Cornelius, John Covert, Terry Kennedy, John Osudar,
Gerardo Razumney.
Note 2510.0, 5-APR-1995
Campobello: Clearing the type-ahead buffer?
-------------------------------------------
What is the best way in DCL to clear the type-ahead buffer from
a command procedure?
Recently an operator hit "N" <Return> twice while executing a backup
procedure. He was answering "no" to a prompt, but the proc of course
took his second "N" as a "no" answer to the subsequent prompt. There
was no harm done in this case because the person realized something
was wrong and re-ran it. However it made me think about clearing the
type-ahead buffer between important prompts.
I can use RECALL/ERASE to clear the recall buffer, but I couldn't find
a specific command to clear type-ahead. SET TERM/INQUIRE does the trick,
The DECUServe Journal May, 1995 Page 10
Clearing the Typeahead Buffer
but it's a little slow because of the system-terminal communication.
Is there a faster way? Thanks.
Note 2510.1, 5-APR-1995
Chadwick: What about SET TERMINAL/NOTYPE_AHEAD ?
------------------------------------------------
Note 2510.2, 5-APR-1995
Razumney: Read with timeout
----------------------------
Try
$ READ/TIMEOUT=0/ERROR=CONTINUE SYS$COMMAND X
$CONTINUE:
Note 2510.3, 5-APR-1995
Osudar: ...unless more than one line was typed
-----------------------------------------------
That's fine, as long as he never types ahead more than one line.
Otherwise you have to put the command in a loop, e.g.:
$eatinput:
$ read/timeout=0/prompt=""/error=ateitall sys$command junk
$ goto eatinput
$ateitall:
btw the /prompt="" is also necessary unless you want to see "Data:"
each time the READ is executed.
Note 2510.4, 6-APR-1995
Carter: Oh, no you can't (digression)
-------------------------------------
> I can use RECALL/ERASE to clear the recall buffer, but I couldn't find
> a specific command to clear type-ahead. SET TERM/INQUIRE does the trick,
Not in a command procedure. I wanted that real bad for some simple
security stuff I was doing in a .COM, and DEC told me that it's
hard-coded into the code to check and be a no-op in a command
procedure. I argued with two DECCies (feel free to post obligatory
replies about how insecure doing security in a .COM procedure is; but
it was still a nifty quick-and-dirty to do some cleanup on your
terminal -- I just wanted a clearscreen/clear-recall-buffer .COM I
could invoke with a DEFINE/KEY keystroke) but to no avail -- they said
they wouldn't put in a request to change it and it should be that way,
with no explanation.
The DECUServe Journal May, 1995 Page 11
Clearing the Typeahead Buffer
Note 2510.5, 6-APR-1995
Briggs: what?
-------------
What's a no-op? The SET TERMINAL /INQUIRE or the side-effect of
clearing the typeahead buffer?
$ SET TERMINAL /INQUIRE works just fine in command procedures. Lots
of people (foolishly or not) put it in their LOGIN.COM.
I just tried and found that $ SET TERM /INQUIRE retains its ability
to zap the typeahead buffer even inside a command procedure. VMS 5.5-2.
Test procedure follows:
$ WRITE SYS$OUTPUT "Waiting..."
$ WAIT 0:0:15
$ WRITE SYS$OUTPUT "Inquiring..."
$ SET TERMINAL /INQUIRE
$ READ SYS$COMMAND Q /PROMPT="Reading..."
$ WRITE SYS$OUTPUT Q
If you put your terminal in display controls mode, you can see the
inquiry escape sequences. But you can't pre-respond to them, not even
if you know the correct escape sequences to use. And any other pre-typed
input doesn't survive to the READ, not even if you stick in lots of
terminators.
One moment while I try the advanced test...
The advanced test demonstrates that $ SET TERMINAL /INQUIRE retains its
ability to zap the typeahead buffer even inside of a batch job. In
that environment, you generally have to specify the terminal explicitly
since SYS$COMMAND doesn't often point to a terminal.
Note 2510.6, 6-APR-1995
Carter: Nope, the $ RECALL/ERASE
--------------------------------
Note 2510.7, 7-APR-1995
Briggs: Ahh. OK -- that makes more sense
-----------------------------------------
Note 2510.8, 16-APR-1995
Covert: Pet peeve #237.4 as a 53-line terminal user
---------------------------------------------------
Well, if you're going to put a SET TERM/INQUIRE in a command procedure,
please get the values of /page and /width before executing the command,
and then include them as /page='savedpage' and /width='savedwidth' on
the inquire command.
The DECUServe Journal May, 1995 Page 12
Clearing the Typeahead Buffer
Note 2510.9, 17-APR-1995
Kennedy: Eh?
------------
I thought VMS V6.something fixed this? At least, there's some makes-no-sense-
for-anything-else stuff in the inquire string (save cursor, set row 255, set
column 255, report cursor position, restore cursor) which drove Kermit nuts
until we figured out what DEC was doing. I hope VMS uses it for something 8-)
Note 2510.10, 2-MAY-1995
Cornelius: Yes, the 255,255 is useful
--------------------------------------
It's probably mentioned elsewhere in this conference, but the
standard trick is
$ SET TERM /INQUIRE -
/PAGE='f$getdvi(dev,"TT_PAGE")'/WIDTH='f$getdvi(dev,"DEVBUFSIZ")'
What it does is inquire as to the terminal type but causes the
SET TERM to not send an escape sequence to resize the window.
> I thought VMS V6.something fixed this? At least, there's some makes-no-sense-
>for-anything-else stuff in the inquire string (save cursor, set row 255, set
>column 255, report cursor position, restore cursor) which drove Kermit nuts
>until we figured out what DEC was doing. I hope VMS uses it for something 8-)
I have used the 255,255 technique for several years.
The point is that if you have LAT or Telnet into a VMS system and
have a nonstandard-sized (non-24x80) terminal window, there is no
way for the terminal driver to know what the actual size is. For
emulations of VT300 series and later, there's an inquiry sequence
to ask the terminal for its page size info, but if you don't want to
rely on all users having set their terminal emulations to one of the
newer terminal types, a good solution is to attempt to position to
255,255 and then read actual cursor position.
Works well the commercial terminal emulators I use, and with DECterms,
but I did notice that ProComm (a few versions back - one of the
shareware versions) was confused by it.
DSNlink Contract Renewal Problems
=================================
Many people seem to be foggy about exactly what the Business
Practices Service Group (and its associated DECUServe Notes
conference) does all day...until they run into the "right" kind of
The DECUServe Journal May, 1995 Page 13
DSNlink Contract Renewal Problems
problem. The next few notes illustrate some of the traffic on a
recurring theme in BUSINESS_PRACTICES, namely DSNlink renewal
problems.
Participants: Ta Fuh Chiam, Keith Hare, Terry Kennedy, Bill Mayhew,
Pat Scopelliti.
Note 481.0, 27-Apr-1995
Hare: Still More DSNlink Problems
---------------------------------
Our multi-year maintenance contract rolled over on April 1, and ever since,
we have had one hassle after another trying to get DSN access straightened
out.
A significant portion of our layered product support got dropped from DSN
access. I have had multiple rounds of discovering individual missing
pieces and attempting to get them added. Sometimes it has worked.
Sometimes other things have gotten dropped in the process.
I have had to go through our maintenance contract and compare it to our DSN
supported product list and our DSN valid addresses.
I have been working through our local Digital contracts person, our
contract administrator in Chicago, and a couple of discussions with people
in the support center.
Nobody seems to be able to take our contract and automatically figure out
what DSN access we are supposed to have, and get it set up correctly. It
sounds like the contracts person in Chicago ships an All-in-1 message to
someone in DSN land listing the supported products. DSN land then updates
its supported product list and DSN valid addresses and sends back a
response. The problems can occur in Chicago, between Chicago and DSN land,
and even inside DSN land. I had one product (fortran) that for a while was
on our supported products list, but not on our DSN valid addresses list.
In the process, I also discovered that we had (prior to April 1) had DSN
access for a three products that were not on our contract but should have
been. And when I added those products to our contract, two of them made
it onto our DSN valid addresses list and the third one didn't.
I am getting very frustrated because I discover missing addresses when I
have a problem and attempt to ship a message.
Is there anyplace I can go and check a part number on my contract to see
what type of DSN access I am supposed to get?
Shouldn't someone within Digital be able to take our contract and set up
appropriate DSN access without all of this hassle?
The DECUServe Journal May, 1995 Page 14
DSNlink Contract Renewal Problems
Note 481.1, 27-Apr-1995
Chiam:
-------
Would you please post or send me e-mail with your phone number and
address so that I can get someone from Digital to contact you about
this problem?
Note 481.2, 28-Apr-1995
Kennedy: And another
--------------------
While you're in there... 8-)
When we renewed our contract we canceled all software support except for
DECserver Network Access software. It is listed as SSS (System Support
Service) on our contract, which allegedly means it's an operating system
(which it is) as opposed to SNS (another copy of the same thing) or LPS
(a layered product).
Digital claims that we are not entitled to DSNlink or DSIN access for this
system product, and that we will have to contact the CSC by phone and that
they will have to FedEx us patch tapes because of this.
This is far more inconvenient for me, and is going to cost Digital real
money, as we have 4 incidents open on this software now and will be getting
new images on a regular basis.
What is really infuriating is that they claim that we would need to have
support for VMS in order to have DSNlink or DSIN access, even though it is
unrelated to the product I want support for. This sounds like an "illegal
tying" to me.
Note 481.3, 28-Apr-1995
Chiam:
-------
Terry, could you please give me the name of the Digital folks with whom
you work on this issue? This just does not sound right. I had an access
number before and it definitely did not have any operating system
associated with it, just a few layered products.
Note 481.4, 28-Apr-1995
Mayhew:
--------
By the way, Keith's description of the problem... i.e. "the way things
work now" in Digital... is essentially correct. An incredible amount
of stuff happens through manual processes among disjoint legacy
systems. (This is a problem with which many of us who live in the real
world can identify, I think.)
The DECUServe Journal May, 1995 Page 15
DSNlink Contract Renewal Problems
It's about time for us (DECUS) to get an update from Digital on their
progress on solving this mess. Chiam, do you have the original
timeframe estimate handy?
Note 481.5, 28-Apr-1995
Hare: Immediate problem resolved, for the moment
------------------------------------------------
After I posted the .0 note, I shipped a copy of it to dsn%mod. I got a
response that the manager of the "entitlements group" will be contacting
our contracts administrator.
Also, when I checked our valid addresses this morning, it looks like
everything is turned on, including some things that I've never heard of.
This solves the immediate issue, for the moment. Until we change something
else.
The question that still remains is how can I look at my contract and figure
out what DSN support I am supposed to have?
Note 481.6, 28-Apr-1995
Scopelliti: How to tell what's about to happen
----------------------------------------------
FWIW you can send mail to DSN%SERVICE-CONTRACTS to see what DSNlink
thinks you have registered and when it expires.
Passive FTP
===========
For long stretches at a time, activity in the SECURITY conference is
confined to regular automatic posting of the ACM's RISKS digest and
the semi-regular posting of CERT sendmail advisories. (Cheap shot,
we admit it.) Discussion topics more often than not are related to
network security. The following extracts are of a recent discussion
of "passive FTP" as a security measure.
Participants: John Briggs, Geoff Bryant, John Covert, Linwood
Ferguson, Hunter Goatley, Terry Kennedy, Pat Scopelliti, Harrison
Spain, Brian Tillman.
Note 307.0, 20-APR-1995
Spain: Passive FTP required for security at my site
---------------------------------------------------
I have a product that claims to work on all systems but is apparently
designed to work on unix systems only.
The DECUServe Journal May, 1995 Page 16
Passive FTP
The product is called "Passive FTP" and is used to help assure security
at my site :-}.
If I try to compile the product, the compile fails. It is looking for
#include <sys/param.h>.
Does anyone know of a "Passive FTP" for VAX/VMS & AXP/VMS?
Thank you very much for any help you can offer :-).
Note 307.1, 20-APR-1995
Kennedy: MultiNet, of course 8-)
--------------------------------
I FTP'd to a MultiNet system (VMS V5.5-2) from a Unix box and said:
quote pasv
and I got the expected "227 Entering passive mode; use PORT a,b,c,d,x,y
so it seems MultiNet can do what you want. What you want it *for* is less
clear, though.
Note 307.2, 21-APR-1995
Spain: Passive FTP
------------------
Well at least I have one machine where I can do this :-). Our internet
service provider gave us a zip file with the source for a small program
that implements passive ftp on most systems.
I was hoping there would be an equivalent in the public domain since
I'm running Wolly and UCX mostly :-).
Note 307.3, 21-APR-1995
Spain: Using Passive FTP for security
-------------------------------------
Could you explain how to use Multinet in this way? I have Multinet
3.3(14) running on a VAX/VMS V6.1 system.
Apparently using this passive FTP allows you to keep your firewall
restricted while still allowing outgoing FTP.
When you FTP out, you use a single socket but FTP requires a series of
sockets to be opened up on the return and this has to be controlled if
you don't want to open up a huge security hole in your firewall.
Passive FTP (I think) allows you to restrict the range for these
sockets.
The DECUServe Journal May, 1995 Page 17
Passive FTP
Note 307.4, 21-APR-1995
Goatley: Coming to MGFTP
------------------------
MadGoat FTP should support passive mode sometime this summer....
Note 307.5, 21-APR-1995
Tillman: Wollongong and UCX
---------------------------
On a Pathway (Wollongong) system, find param.h in
TWG$TCP:[NETDIST.INCLUDE.SYS]. When you compile C source that is
TCP-aware, for Wollongong you should define SYS to point to this
directory and SYS$LIBRARY:
$ define sys twg$tcp:[netdist.include.sys],sys$library:
For UCX, most programs I've seen that compile on multiple TCP/IPs,
there is an
#ifndef VMS
#include <sys/param.h>
#endif
construct, meaning that UCX probably doesn't need this include.
Note 307.6, 21-APR-1995
Spain:
-------
Thank you! I'll try the compile with this include commented out.
I "discovered" that "quote pasv" works on my HP-UX systems, and my VMS
UCX systems! Now all I have to figure out is how to use passive FTP.
It seems a mystery to most folks :-).
Note 307.7, 21-APR-1995
Spain:
-------
I tried removing the #include <sys/param.h> line and it just complained
about another file (located in the posix structure). I defined SYS to
point to that directory and it complained about an "arpa/ftp.h" file
which is not on my system :-).
I think my best bet is to use the passive FTP built into UCX. Once I
can use it on UCX, I can then let everyone know the way to use it on
HP-UX and we are back in business :-).
The DECUServe Journal May, 1995 Page 18
Passive FTP
Note 307.8, 22-APR-1995
Scopelliti: Quick how-to for passive FTP
----------------------------------------
OK.. Let's see if I got this right..
Passive FTP allows me to FTP from my Internal network to my firewall
system and thence out to an external FTP site.
Now, here's the question: Once I do the quote pasv command and get the
PORT.... back, what do I do to access the external node? I tried OPEN,
CONNECT, PORT and all were rejected as inappropriate commands.
Note 307.9, 24-APR-1995
Spain: Getting further but still stuck (need Passive FTP Client for VMS
-----------------------------------------------------------------------
I appears that you need a client (FTP) that can talk "Passive FTP" to a
server that supports the QUOTE PASV command.
Many TCP/IP packages support Passive FTP (just type QUOTE PASV after
connection to see if yours does) *but* getting a client for VMS is a
problem :-).
BTW, UCX V3.2 "supports" Passive FTP but it is broken at the moment.
The fix is in ECO 1 for UCX V3.3.
Does anyone know of a Passive FTP client for VMS?
Note 307.10, 25-APR-1995
Spain: Update on Passive FTP (TGV works!)
-----------------------------------------
An update on this process :-). I called TGV (if anyone would know
about this, they would!).
The provided a VMSINSTAL'able kit that implements Passive FTP Client!
It worked perfectly :-).
The theory was correct. You need a client FTP that supports passive.
In this case, you "toggle" passive by typing PASSIVE at the FTP prompt
before connecting to the FTP server.
Note 307.11, 25-APR-1995
Covert: I use Netscape to do my FTP retrievals
----------------------------------------------
What is the relationship (if any) between "passive FTP" and the proxy
servers supported by Mosaic/Lynx/Netscape to allow http, gopher, and
ftp access (usually retrieve only) through a firewall?
The DECUServe Journal May, 1995 Page 19
Passive FTP
Note 307.12, 25-APR-1995
Briggs: Here's why you'd use passive FTP
----------------------------------------
None really.
Proxy gateways are one way of allowing controlled access through a firewall
that would otherwise prevent access entirely. All access is funneled
through the proxy gateway which can then enforce some kind of policy
and provide some kind of audit trail.
Passive FTP is a way through a firewall that is set up to allow for
outbound-only access.
When you connect to an FTP server from your FTP client, there is a
command connection that is opened up. This connection is initiated by
the client. Firewalls that allow outbound connection initiations won't
be bothered by this. This part is the same under both ordinary and
passive FTP.
But when you transfer a file (or request a directory listing), a data
connection is opened up. With ordinary FTP, the client allocates a
TCP port number and tells the server that port number. The server then
initiates the data connection. This is an inbound connection and
is likely to be forbidden by your firewall.
With passive FTP, the client requests a port number from the server.
Data connections are initiated by the client instead of by the server.
This keeps your firewall happy.
Note 307.13, 25-APR-1995
Bryant: ftp client pasv support coming to TCPware
-------------------------------------------------
The upcoming version of TCPware from Process Software (currently in
beta test) will have passive support in the ftp client.
Standard Disclaimer: I work for Process Software.
Note 307.14, 25-APR-1995
Scopelliti: Still passively lost in the haze
--------------------------------------------
Still confused...
I'm sitting on my PC using Reflection Network Series (RNS). I'd like
to open an FTP connection to eisner.decus.org and pass through my
firewall system INET (running Multinet V3.3B).
RNS allows me to specify a passthrough server (along with its
username/password), as well as a Style (either SITE servername or
username@servername).
The DECUServe Journal May, 1995 Page 20
Passive FTP
Connection never seems to make it through the firewall. Messages seem
to be coming directly from the firewall (e.g. you're already logged
in).
Is there anything on the firewall system that needs to be configured?
Note 307.15, 26-APR-1995
Briggs: FTP proxy gateways
--------------------------
> I'm sitting on my PC using Reflection Network Series (RNS). I'd like
> to open an FTP connection to eisner.decus.org and pass through my
> firewall system INET (running Multinet V3.3B).
That is, you have an FTP proxy gateway that you wish to navigate through.
> RNS allows me to specify a passthrough server (along with its
> username/password), as well as a Style (either SITE servername or
> username@servername).
These being two mechanisms that can be used for your FTP client to specify
to the proxy gateway the actual name of the host to which you wish to
connect. Some gateways use one mechanism. Some use the other.
> Connection never seems to make it through the firewall. Messages seem
> to be coming directly from the firewall (e.g. you're already logged
> in).
>
> Is there anything on the firewall system that needs to be configured?
Shouldn't be. But you need documentation. What commands does your FTP
proxy gateway expect to tell it where to go? Typically, you can enter
these commands interactively. You don't need any fancy client. Just
do it yourself. Then, once you understand the dialogue, you can try
a fancy client.
For your information, an FTP proxy gateway looks to the client like an
ordinary FTP server. But it has to accept some extended syntax to let
your client tell this pseudo-server the name of the real FTP server that
you want to use. With this information in hand, the proxy gateway connects
the real server on your behalf. Your FTP commands are relayed through
the gateway to the real server. The data connections are relayed as well.
The gateway plays games with the PORT command so that everything works out.
I certainly apologize if all of the above was already obvious to you.
The DECUServe Journal May, 1995 Page 21
Passive FTP
Note 307.16, 26-APR-1995
Scopelliti: No problemo!
------------------------
> I certainly apologize if all of the above was already obvious to you.
One can never be insulted or annoyed (or whatever other word is
appropriate) when one is learning. Actually, you confirmed how I
imagined it would function.
I just haven't been able to find that subset in Multinet 3.3B
Note 307.17, 28-APR-1995
Ferguson: RNS ok with TIS firewall
----------------------------------
> I'm sitting on my PC using Reflection Network Series (RNS). I'd like
> to open an FTP connection to eisner.decus.org and pass through my
> firewall system INET (running Multinet V3.3B).
I use RNS's FTP all the time through a TIS firewall.
I did have one problem with the production release. It _always_
attempts to use the ABORT command to tell if the destination (beyond
the FW) respects the command. We found one site (ftp.ksc.nasa.gov,
home of winvn) which responded to that by logging you off, leaving you
with a sort of strange state
WRQ said this had to do with going through a firewall, and gave us an
alpha version of their next FTP client which fixed it. I never thought
it was the firewall's problem, but it did fix it for us. They said it
now does not issue the ABORT test when going through a firewall.
It is possible that the same command is having a somewhat different
effect on you? You might ask them about testing this newer version.
You might also post the response in the bottom window that you get when
running, and I'll compare it to mine and see if there is a clue there.
TeamLinks MIME Support
======================
Picking up on an older thread concerning TeamLinks and mail servers
in the OFFICE_AUTOMATION conference, a subscriber recently asked
about support for the MIME standard. After the initial question was
answered, the discussion turned to those ever-vexing TLAs
(Three-Letter Acronyms). Note that a few remain unanswered...do YOU
know what they are?
The DECUServe Journal May, 1995 Page 22
TeamLinks MIME Support
Participants: Bruce Bowler, Roger Bruner, Graham Pye, Joe Whatley.
Note 160.3, 24-APR-1995
Whatley: TeamLinks and MIME
---------------------------
One of the criteria we have in developing our Mail strategy for the
next several years is that our primary mail system must be
MIME-compliant or MIME-enabled.
Our hope is that TeamLinks will have this flexibility, but I haven't
been able to find anything here that indicates that is in the works.
Does anyone have any information on MIME and TeamLinks?
Note 160.4, 2-MAY-1995
Pye: Yes (Soon...)
------------------
I've asked our architect, and this is what he says:
As you know, ALL-IN-1 V3.2 will be able to use either MAILbus 400 or
continue to use Message Router for mail. MAILbus 400 will have a
SMTP/MIME gateway available, so TeamLinks users will be able to take
advantage of it (transparently).
In a PFR (Possible Future Release :-) ) we will be looking at direct
connection to Internet messaging backbones using SMTP/MIME interfaces.
Note 160.5, 2-MAY-1995
Bruner: Needed: Acronym interpretation
---------------------------------------
[Graham, sometimes I lose sight of the meaning of what are probably
really obvious acronyms. What is 'HTH' please?]
Note 160.6, 2-MAY-1995
Whatley: Pretty good news
-------------------------
Graham,
Thank you for the quick reply and relatively good news. Knowing that
we can at least plan on a MAILbus 400 implementation helps eliminate
some of the unknowns in our planning.
The DECUServe Journal May, 1995 Page 23
TeamLinks MIME Support
Note 160.7, 2-MAY-1995
Bowler: TLA
-----------
> [Graham, sometimes I lose sight of the meaning of what are probably
> really obvious acronyms. What is 'HTH' please?]
Hope This Helps
AIS (Acronym Interpretation Services)
Note 160.8, 2-MAY-1995
Bruner: TH (That helped!)
-------------------------
Thanks, Bruce! Now that I know, I will probably opt to confuse folks
here at work by using it because it's a good one.
Note 160.9, 4-MAY-1995
Pye: Step forward the AIS!
--------------------------
Do you also know about :
RTFM, WAG, SWAG, FWIW, YMMV, TTWIW, WFM, IM(H)O...
Note 160.10, 4-MAY-1995
Bowler: 75%
-----------
Read The Fine Manual
Wild A__ Guess
Silly WAG
For What It's Worth
Your Mileage May Vary
?
?
In My (Humble) Opinion
Note 160.11, 4-MAY-1995
Bruner: FWIW, I'm going to ponder ;9
------------------------------------
re ;9:
Graham, right off the bat, you have me with most of those, but I'm
printing a copy to think about before I give up that easily (at least
one of them was familiar: IMHO; oh, and also FWIW).
The DECUServe Journal May, 1995 Page 24
IP Subnetting Help
IP Subnetting Help
==================
Next, from the INTERNETWORKING conference, we bring you this
elucidation of IP subnet masks. As one participant so nicely put
it, "valuable and gentle" ... the DECUServe way.
Participants: Kevin Angley, John Briggs, Matt Holdrege, Bret
Wortman.
Note 326.0, 10-Mar-1995
Wortman: Weird ftp/telnet connectivity problem.
-----------------------------------------------
I have a customer configured something like this:
AS/400 (xxx.yyy.11.n)
|
Bridge
|
Bridge
|
Bridge
|
VAX 4705A (xxx.yyy.15.3)
The AS/400 can only telnet to me if it changes its .11 to a .15 first.
Where should I be looking for trouble?
Note 326.1, 10-Mar-1995
Angley: What is xxx?
--------------------
If the bridges are really bridges, then they don't have a clue about
IP, so they are not likely to be culprits here. So you effectively have
AS/400 (xxx.yyy.11.n)
|
VAX 4705A (xxx.yyy.15.3)
Now .. if xxx is less than 192 (and you aren't subnetting) then these
nodes are on the same network, and you do indeed have a problem.
But if xxx is 192 or greater, or you subnet the class A or B, then
somebody somewhere needs to route between these nodes on different
networks.
The DECUServe Journal May, 1995 Page 25
IP Subnetting Help
Note 326.2, 10-Mar-1995
Holdrege: what are the masks?
-----------------------------
Note 326.3, 13-Mar-1995
Wortman: *This* one I know for sure....
---------------------------------------
The masks I know: 255.255.255.0 on the VAX with certainty, they've
told me the same holds true on the AS/400.
Note 326.4, 13-Mar-1995
Angley: Are the bridges bridges, or brouters, or routers?
---------------------------------------------------------
OK, so the nodes you cited in .0 are on two different networks.
Somebody needs to be routing between them.
Note 326.5, 13-Mar-1995
Briggs: Subnets and routers
----------------------------
> AS/400 (xxx.yyy.11.n)
> |
> VAX 4705A (xxx.yyy.15.3)
> The masks I know: 255.255.255.0 on the VAX with certainty, they've
> told me the same holds true on the AS/400.
You need either to have a router to move the packets between subnet 11 and
subnet 15 or you need to change the network mask to something like
255.255.248.0 to allow both 11 and 15 to be within the same subnet.
Before you change the subnet mask, be sure you know what you're doing.
Stop reading now if you already know about IP, ARP and subnet masks...
The purpose of a subnet mask is to divide an IP address into two halves:
a network number and a host number within the network. Two hosts are on
the same network if their network numbers match.
When a host has to transmit an IP packet, there are two possibilities:
o The target IP address has the same network number as (some interface
of) the sender.
o The target IP address is on some other network number.
If the target is on the same network then the packet can be transmitted
directly. The manner in packets are directly transmitted is network
The DECUServe Journal May, 1995 Page 26
IP Subnetting Help
dependent. On Ethernet, it is done with the assistance of the ARP
protocol:
Sender broadcasts an ARP request: "who has IP address xxx.yyy.15.3"
Target sends ARP response: "MAC adddress for xxx.yyy.15.3 is
dd:dd:dd:dd:dd:dd"
Sender sends IP packet to Ethernet address dd:dd:dd:dd:dd:dd
The sender maintains an "ARP cache" so that it doesn't have to
re-verify the Ethernet address for the target for each new packet.
If the target is on a different network then the packet must be
transmitted through a router:
Sender looks up the target network number in its routing table
Often this table consists of a single default entry: "route everything
via x.y.z.w"
Sender sends the packet directly to the router. On Ethernet, this
is done with the assistance of ARP as above.
When you've got two different IP network numbers on the same Ethernet,
things still work the same way. The sender sees that the network numbers
are different and transmits through a router. If you've got a router
that can deal with having two different network numbers on the same
cable and have configured it accordingly, both sender and target can
shoot their packets toward the router. The router will then retransmit
them appropriately.
Note 326.6, 13-Mar-1995
Wortman: Got it, I think....
----------------------------
Okay -- this seems to verify what I only had a foggy grasp of: That
the subnet mask of 255.255.255.0 was probably forcing each machine to
see only the last octet of addresses and assume the rest was the same
as its own.
So our choices are three:
1. Change the subnet mask to allow .11 and .15 to see each other.
2. Stick in a router and educate it about the two subnets.
3. Change the .15 subnet to .11 also.
This is largely up to the client, but from a technical POV, I'd say #2
or #3 are preferable. I know *I*'m not swift enough to set a subnet
mask without a wee more education, and since the client missed this
concept entirely, I'm not sure they'll want to tackle it either.
Thanks for the input, valuable (and gentle) as always....
The DECUServe Journal May, 1995 Page 27
IP Subnetting Help
Note 326.7, 13-Mar-1995
Briggs:
--------
> Okay -- this seems to verify what I only had a foggy grasp of: That
> the subnet mask of 255.255.255.0 was probably forcing each machine to
> see only the last octet of addresses and assume the rest was the same
> as its own.
Reasonably close. Rather than assuming that the rest of the address
was the same as its own, I'm sure that both machines were aware of the
mismatch in network numbers.
If a default router was not specified when the machines were configured
(and not learned via a routing protocol) then the machines will refuse
to transmit. This should result in an error message like "no route to
host".
If a default router is specified but the router has no knowledge of
the target subnet, you should also get "no route to host". (The
router is responding with an ICMP "net unreachable" message which
is then reported to the user as "no route to host").
If you're getting "host unreachable" or "target does not respond" then
you've probably got something more going on (like a mis-specified,
misconfigured or nonexistent router).
Note 326.8, 13-Mar-1995
Wortman: <LIGHTBULB.MPG>
------------------------
Thanks for the clarification -- I'm going to log on and check a few
things out today.
Note 326.9, 13-Mar-1995
Wortman: More info
------------------
New datum:
The first octet of both addresses is 192 (I'll withhold the second in
deference to my client's anonymity). They're claiming this means that
subnetting and masks don't really apply to them.
I told them that installing a router was the best solution anyway, to
which they asked "where?" and I hedged again, trying to push
topological decisions back on their internal networks folks. If
they're insistent, what kind of guidelines can I provide and/or use?
And what are some good resources for learning this stuff?
The DECUServe Journal May, 1995 Page 28
IP Subnetting Help
Note 326.10, 13-Mar-1995
Angley:
--------
> The first octet of both addresses is 192
Then it is a Class C address, and if there is no subnetting going on,
then the first three octets specify the network and the last octet
specifies the host interface.
> subnetting and masks don't really apply to them
Well, with a mask of 255.255.255.0, then they aren't doing any
subnetting so they are right. But they could subnet if they wanted -
but we digress.
> installing a router was the best solution
Before I did that, I would ask why two nodes on the same physical
network are numbered as if they were on two different networks. If
there is no good reason, I'd renumber on of them and you're done.
> And what are some good resources for learning this stuff?
Hunt, Craig: TCP/IP Network Administration. Published by O'Reilly.
ISBN 0-937175-82-X.
Note 326.11, 13-Mar-1995
Wortman: "Can't we all just get along?"
---------------------------------------
> Before I did that, I would ask why two nodes on the same physical
> network are numbered as if they were on two different networks. If
> there is no good reason, I'd renumber on of them and you're done.
Just a hunch, but I think the (primary) reason is political. Namely,
they don't really want this VAX thingy sitting on the same LAN as their
AS/400 and its kin.
It's also a question of ownership of the VAX equipment by the network
admin people who want to keep it logically separate from the rest of
their gear.
In other words -- the separation is logical only. Communications are
very much required between the two, but politics are dictating separate
LANs. Those who have no faith in this project own the .11 subnet and
those who don't want to give the project up own the .15. One wants to
avoid culpability in case of failure while the other doesn't want to
share credit when it works.
Or something like that, I suspect....
The DECUServe Journal May, 1995 Page 29
IP Subnetting Help
Note 326.12, 14-Mar-1995
Briggs: Any existing routers?
-----------------------------
> I told them that installing a router was the best solution anyway, to
> which they asked "where?" and I hedged again, trying to push
> topological decisions back on their internal networks folks. If
> they're insistent, what kind of guidelines can I provide and/or use?
Regardless of the fact that you've got a number of bridges, you've still
got single logical Ethernet. You can plug the router in on whichever
segment is most convenient (use the politically expedient phone closet).
But first, let me ask another question: Does the customer have any
connections from this Ethernet to the Internet? (That is, does he
already have one or more routers installed?)
If so, you may be able to get a no-cost solution by simply using the
existing router.
Note 326.13, 14-Mar-1995
Wortman: Woolgathering
----------------------
How about this -- could the VAX itself be designated router for both
itself and the AS/400? I suspect not if for no other reason that there
doesn't seem to be any way to tell UCX that it has two addresses (and
I'd assume this to be a requirement if it's going to route for both the
.15 and .11 subnets).
Have I missed something really elementary?
Note 326.14, 15-Mar-1995
Briggs: UCX claims to be able to do multiple interfaces per cable
-----------------------------------------------------------------
It looks like that will work. I just went into UCX and the HELP under
SET INTERFACE indicates that you can do exactly that.
Assuming that your primary interface is ZE0:
UCX> UCX SET INTERFACE ZE1 /HOST=x.y.z.15 /NETWORK_MASK=255.255.255.0
Unfortunately, when I tried this on my machine just now, it failed with
a "no device available" message. Possibly you need to do a SET CONF INTERFACE
and restart UCX.
The DECUServe Journal May, 1995 Page 30
UART Accelerator
UART Accelerator
================
Next up, from the PERSONAL_COMPUTING conference, is a treatment of
UART issues that may well be more than you ever wanted to know about
the topic.
Participants: Dale Coy, Bob Early, Matt Holdrege, John Osudar, Pete
Sivia.
Note 698.0, 14-APR-1995
Early: 16550 UART accelerator boards for 115200 bps? help
---------------------------------------------------------
Has anyone used or are intimately familiar with the so-called "UART Accelra
Note 698.1, 14-APR-1995
Holdrege: what are you looking for?
-----------------------------------
Only that you need 16550 UART's to do 115200. The older 8??? series
can't handle higher speeds.
Note 698.2, 16-APR-1995
Osudar: what's the question?
----------------------------
I use a Boca I/O card to talk to my 28.8K modem at 115200 bps. What do
you want to know? (I'm no UART expert but without knowing what your
questions are, it's hard to attempt an answer! :-)
Note 698.3, 28-APR-1995
Early: .. based on reports truue 115200bos is possible ..?
----------------------------------------------------------
More specifically, I have been told that there are accelerator cards
available for pc's; which will run at 115200 bps, without data overrun, for
dos and windows applications. I haven't seen one but it is an interesting
thought.
My curiosity is one to know if anyone else has tried one, on an ordinary pc
(26b, 286 through Pentium), and what there opinion is after trying one.
I can almost believe it possible if using a direct connect with a high speed
processor, but am cynical (skeptical) about through a modem!
The DECUServe Journal May, 1995 Page 31
UART Accelerator
Note 698.4, 28-APR-1995
Coy: Sure, if you don't use a modem
-----------------------------------
> More specifically, I have been told that there are accelerator cards
> available for pc's; which will run at 115200
> bps, without data overrun, for dos and windows applications. I haven't seen
one
> but it is an interesting thought.
That sounds possible to me. Some of my serial ports run at about half
that speed.
> My curiosity is one to know if anyone else has tried one, on an ordinary pc
> (26b, 286 through Pentium), and what there opinion is after trying one.
386DX-33, as indicated above, works fine talking to a modem.
> I can almost believe it possible if using a direct connect with a high speed
> processor, but am cynical (skeptical) about through a modem!
It's not a problem with direct connections. In fact, a very slow 286
would have no problem. It just doesn't take a high speed processor at
all. And it's no trick at all (today) to design a UART that handles
that kind of speed. The 16550 actually uses rather slow (and fairly old)
technology.
Did somebody tell you that there are modems available that will do
115,200 over a normal telephone line? If they did, I hope you were
either very rude to them, or that you bet them a lot of money. There
are no such things.
A UART isn't a modem. It is used in the connection between a PC and a
modem. The fact that I have a serial connection TO a modem that runs
at 56K, does NOT mean that the modem is talking to another modem at
56K. In fact, that particular modem is only capable of 9600 with
another modem. But it talks to the PC at 56K.
Note 698.5, 28-APR-1995
Sivia: Bottleneck between modem and PC at 115K?
------------------------------------------------
I was at a technical talk the day before yesterday and the speaker
pointed out that it's very possible for the V.34 (28.8K) modems with
compression to do 115K between them these days. And if you have COM
ports on the PC or workstation that won't do 115K to the modems then
you have a built-in bottleneck. The fellow was describing SLIP
connections and the problem of com ports being not being able to handle
115K was bottleneck in these high-speed communications links and
apparently well known in the Sun WS community as earlier Suns had
problems getting the WS ports to go past 38.4K. PC users were reported
to be experiencing the same problem with some of the "better" Internet
access providers. NS16550A's on the communications cards with their 16
byte buffers were offered as the solution.
The DECUServe Journal May, 1995 Page 32
UART Accelerator
Note 698.6, 28-APR-1995
Coy: I agree
------------
Pete, I think we're both right.
> pointed out that it's very possible for the V.34 (28.8K) modems with
> compression to do 115K between them these days.
Yes, it's possible, under the conditions you mention, and with good
LUCK on the lines (or with dedicated lines or whatever). It's just not
what I would call "normal", and it certainly has nothing (directly) to
do with what is going on in a 16550 (or with an "accelerator" card).
The issue of 16550s seemed to be the major question here.
Of course, if you _do_ have a situation where the 16550 is not running
as fast as the modem connection, it certainly is a bottleneck.
Note 698.7, 28-APR-1995
Osudar: keep in mind, someone's got to read the data regardless
---------------------------------------------------------------
Let's not forget that the advantage of the 16550 is that it can hold
up to 16 characters in its buffer. But if your application can't get
around to reading the port until after a 17th character comes in,
you're still going to get overruns.
I have a 28.8Kbps modem on my home PC (486DX/33); I bought a new serial
card with a 16550 UART to work with the modem, because my original
serial port wouldn't work reliably above 19.2K. But in order for the
16550 port to do any good, I had to install a driver that would enable
the buffer; and I had to set a threshold in the driver (at which an
interrupt happens, I suppose?) to some reasonable value (8 or 12, I
forget which -- but definitely not 1 or 15!) Anyway -- even with the
16550 and the right driver, when I run PPP and am doing heavy data
transfers (e.g. FTP's of large files), I see data overruns if I'm
running other applications in Windows while the file transfer is in
progress. So there's no guarantee, with the 16550 or without, that you
won't get overruns.
Assorted Tidbits
================
We'll round off the issue with a handful of shorter items. The
first is once again for those of the PC persuasion. The topic is a
problem with larger (700 MB) IDE drives under DOS and Windows NT.
The DECUServe Journal May, 1995 Page 33
Assorted Tidbits
Participants: Kevin Angley, Pat Scopelliti, Alan Striegel.
Note 700.0, 21-APR-1995
Scopelliti: IDE drive > 503 MB ??
---------------------------------
Help a not-quite-newbie (me)
A co-worker has a PC with a 244 MB IDE disk drive (MS-DOS V6.2 and NT
V3.5). He needed more space, so we got a 700 MB IDE drive and a cable
that allows us to connect both drives (one primary, other secondary).
Unfortunately, neither OS sees more than 503 MB on the new disk. He
got a disk utility that allows DOS to see all of the bigger disk, but
it walks over the boot block and prevents NT from ever booting (it goes
right to DOS without ever hitting the dual-boot option).
Any solution to this besides writing off the excess 200MB on the new
disk? Where's AUTOGEN when you need it? ;-}
Note 700.1, 21-APR-1995
Angley: Enhanced IDE
--------------------
Try a different disk controller. One described as an Enhanced IDE
Interface.
Note 700.2, 24-APR-1995
Striegel: You may need new BIOS code
------------------------------------
Some systems, even ones with enhanced IDE interfaces still cannot address
the full size of a large drive. It has more to do with the BIOS.
A number of vendors offer upgrades to the BIOS for the purpose of
addressing larger hard disks. I know Gateway 2000 does not provide them for
free, but has you call another company that sells the Phoenix BIOS for many
different systems. The cost might be $75 or $80.
Note 700.3, 24-APR-1995
Scopelliti: Patch kit fixed it.
--------------------------------
Got the fix! It required installation of Service Pack 2 for WNT V3.5
and it works! Thanks for the help.
Our next "quickie" concerns a small problem that came up with the
license server in PATHWORKS 5.1.
The DECUServe Journal May, 1995 Page 34
Assorted Tidbits
Participants: Petri Backstrom, Bill Bochnik, Rick Bowen, Kevin
Roels.
Note 818.0, 10-APR-1995
Roels: Can't attach to License server w/PW51 TCP/IP
---------------------------------------------------
We have run into a bit of an issue here, and I'm wondering if it's just
me....
We are upgrading a few clients from 4.1 to 5.1 (MS-Windows 3.11). The
server is 4.1. As a 4.1 client running with DECNET or NetBui, we could
attach to the license server just fine. Under 5.1, wanting to use TCP/IP
(the server is in a different subnet than the client) we cannot attach to
the license server anymore. I know it still works because if I enable
NetBui as a protocol, we attach to the license server just fine.
Can anybody help me out?
Note 818.1, 10-APR-1995
Bochnik: netbeui needed
-----------------------
Netbeui or UDP broadcasts are used for the License verification. You
need to allow those to cross the subnets (while increasing traffic :-(
)
Note 818.2, 11-APR-1995
Bowen: File needed on client
----------------------------
we had a similar problem. You need to set up a NBHOST file on the
client with the IP address of the license servers in the different
subnet. (This information came from one of the groups with which I
work so I would suggest looking up the specifics as far as the correct
spelling and format of the file).
Note 818.3, 11-APR-1995
Backstrom:
-----------
If the client is on a different IP subnet as the license
server, you use the INETNAME utility to add a NetBIOS
remote adapter name PWRK$Llicense-server-name to get by
the issue with broadcasts (the NetBIOS name query) not
going across subnets.
This is also documented in either the Installation
and Configuration Guide or the Guide to PATHWORKS
Licensing; check the appendices.
The DECUServe Journal May, 1995 Page 35
Assorted Tidbits
Note 818.4, 12-APR-1995
Roels: Silly me....
-------------------
Petri,
I tried it, and it worked. Turns out the name I was using was
PWRK$LMname, and it should have been PWRK$Lname. I've only tried
it a few times so far, but it seems to work fine. I guess nobody
will ever accuse me of being an accurate typist.
Thanx!
We wrap up this issue with one last SDA tidbit from the VMS
conference. See you next month!
Participants: John Briggs, Simon Maufe, Brian Schenkenberger.
Note 2506.0, 30-MAR-1995
Maufe: how to find processes attached to device/mailbox
-------------------------------------------------------
How about an SDA question? If I I use SDA to chase down a problem, and
find a person is writing to mailbox X, or whatever, how can I tell what
other processes have a channel to this same mailbox?
This is VAX/VMS V6.1..
Note 2506.1, 31-MAR-1995
Briggs: Make that 'SHOW SUMMARY', not 'SHOW PROCESS'
----------------------------------------------------
I think that the only way is to look through the open channels for all
processes on the system.
SDA> SET OUTPUT your-file.out
SDA> SHOW PROCESS
SDA> SET OUTPUT SYS$OUTPUT
SDA> SPAWN
$ EDIT your-file.out
(Change it into a format like the following:)
SHOW PROC /CHANNELS /INDEX=x
SHOW PROC /CHANNELS /INDEX=y
...
$ LOGOUT
SDA> SET OUTPUT your-file.new
SDA> @your-file.out
SDA> SET OUTPUT SYS$OUTPUT
SDA> SPAWN
$ SEARCH your-file.new "MBAnnn" ! (Or use your favorite utility to
! check for matches)
The DECUServe Journal May, 1995 Page 36
Assorted Tidbits
Note 2506.2, 31-MAR-1995
Schenkenberger: Peek inside the mailbox
---------------------------------------
If the mailbox messages haven't been read, you could try looking
and the actual messages for a clue.
SDA> SHOW DEVICE MBAx
SDA> EXAM @UCB;100
And continue to follow the queue of mailbox messages. Keep your I&DS
available for the format of the mailbox messages.
About the DECUServe Journal
===========================
Publication Information
Topic threads in the DEC Notes conferences on DECUServe are selected
for publication on the basis of strong technical content and/or
interest to a wide audience. They are submitted to the editor from
various sources, including DECUServe Moderators, Executive Committee
members, and other volunteers. Suggestions for inclusion are
enthusiastically solicited. Articles selected for publication are
edited on an OpenVMS VAX system in TPU and then formatted with
Digital Standard Runoff.
What's a DECUServe, Anyway?
DECUServe is an on-line conferencing system which is available 24
hours a day, seven days a week. System backups are done on Fridays
from 7:00am to approximately 9:00am Eastern (Daylight/Standard)
Time. Scheduled downtimes for equipment or software maintenance are
announced in advance. The system can be reached worldwide, and
membership is by individual subscription only (no group or company
accounts). The current annual subscription fee is US$75.00.
On-line subscription information is available in the U.S. by
dialing 1-800-521-8950 and logging in with username INFORMATION.
DECUServe and the INFORMATION account can also be reached on the
Internet via telnet connection to decuserve.decus.org or e-mail to
infor...@decuserve.decus.org.
Contact Information
===================
The editors of the DECUServe Journal are Brian and Sherrie McMahon.
They can be reached by any of the following means:
The DECUServe Journal May, 1995 Page 37
Contact Information
mcma...@decuserve.decus.org
mcma...@decuserve.decus.org
mcma...@decus.org
grif...@decus.org
mcm...@ac.grin.edu
grif...@ac.grin.edu
+1 515 269 4901 (normal office hours, U.S. Central time)
+1 515 269 4936 FAX