Out of stream resources
Gary Quiring
user license) running Informix 7.31UC5 to go over 70+ users without
running out of streams.
It's a DELL 6300 PowerEdge, 4 way XEON, AMI Raid, 2gig of RAM, 3COM
905 ether.
I tried increasing NSTREAM, NSTRPAGES with no improvement. I looked
at several TA's including 107566 but can't find any solutions. I
keeping increasing the numbers in stune, but nothing fixes it, 70+
users and it starts choking.
/var/adm/syslog:
Aug 6 20:45:59 action1 inetd[302]: accept: (for telnet) Out of stream
resources
/etc/conf/cf.d/stune:
NODE "action1"
GPGSLO 2000
GPGSHI 6000
PLOWBUFS 1024
NSTRPAGES 6256
NAIOPROC 50
NAIOREQ 400
NAIOBUF 400
NAIOHBUF 100
NAIOREQPP 400
NAIOLOCKTBL 50
MAX_PROC 10000
MAXUMEM 1048576
NCALL 512
NCLIST 2079
NHINODE 1024
NSTREAM 37440
NSTREVENT 13248
NUMSP 384
NUMTIM 1680
NUMTRW 1680
SDSKOUT 64
SEMMAP 8192
SEMMNI 8192
SEMMAP 8192
SEMMNI 8192
SEMMNS 8192
SEMMSL 150
SEMMNU 100
SHMMAX 2147483647
TTHOG 8192
EVQUEUES 468
EVDEVS 476
NSPTTYS 460
SECLUID 0
SECSTOPIO 0
SECCLEARID 0
MAXACPUS 9
NMUXLINK 320
MAXUP 900
NOFILES 2048
netstat -m
config alloc free total max
fail
stream 37440 282 37158 3647 319
0
queues 1704 718 986 8090 821
0
mblks 2660 2292 368 44049765 2522
0
buffer headers 2746 2631 115 6742074 2713
0
class 1, 64 bytes 320 203 117 21646655 301
0
class 2, 128 bytes 64 0 64 2249646 58
0
class 3, 256 bytes 160 38 122 5535021 158
0
class 4, 512 bytes 32 8 24 101877 27
0
class 5, 1024 bytes 20 0 20 1558669 20
0
class 6, 2048 bytes 818 815 3 74633 818
0
class 7, 4096 bytes 785 782 3 23273 785
0
class 8, 8192 bytes 1 0 1 16 1
0
class 9, 16384 bytes 1 0 1 14 1
0
class 10, 32768 bytes 0 0 0 0 0
0
class 11, 65536 bytes 0 0 0 0 0
0
class 12, 131072 bytes 0 0 0 0 0
0
class 13, 262144 bytes 0 0 0 0 0
0
class 14, 524288 bytes 0 0 0 0 0
0
total configured streams memory: 25024.00KB
streams memory in use: 4943.82KB
maximum streams memory used: 5074.49KB
inet mblk cache: 256 = 0, 2048 = 782, 4096 = 782
networking allocation:
type alloc max fail
socket 101 127 0
rawcb 0 1 0
inpcb 251 287 0
tcpcb 227 263 0
ifnet 4 4 0
route 40 48 0
ifaddr 2 2 0
ipfrag 0 0 0
sockaddr 202 254 0
iovec 0 0 0
moptions 0 0 0
ipmaddr 2 2 0
arpinfo 27 35 0
mbcl 0 0 0
ppp 0 0 0
usock 12 14 0
Bela Lubkin
> I'm not able to get a SCO 5.0.6 (506a patch installed) server (105
> user license) running Informix 7.31UC5 to go over 70+ users without
> running out of streams.
>
> It's a DELL 6300 PowerEdge, 4 way XEON, AMI Raid, 2gig of RAM, 3COM
> 905 ether.
>
> I tried increasing NSTREAM, NSTRPAGES with no improvement. I looked
> at several TA's including 107566 but can't find any solutions. I
> keeping increasing the numbers in stune, but nothing fixes it, 70+
> users and it starts choking.
>
> /var/adm/syslog:
> Aug 6 20:45:59 action1 inetd[302]: accept: (for telnet) Out of stream resources
> netstat -m
> config alloc free total max fail
> stream 37440 282 37158 3647 319 0
> queues 1704 718 986 8090 821 0
> mblks 2660 2292 368 44049765 2522 0
> buffer headers 2746 2631 115 6742074 2713 0
> class 1, 64 bytes 320 203 117 21646655 301 0
> class 2, 128 bytes 64 0 64 2249646 58 0
> class 3, 256 bytes 160 38 122 5535021 158 0
> class 4, 512 bytes 32 8 24 101877 27 0
> class 5, 1024 bytes 20 0 20 1558669 20 0
> class 6, 2048 bytes 818 815 3 74633 818 0
> class 7, 4096 bytes 785 782 3 23273 785 0
> class 8, 8192 bytes 1 0 1 16 1 0
> class 9, 16384 bytes 1 0 1 14 1 0
> class 10, 32768 bytes 0 0 0 0 0 0
> class 11, 65536 bytes 0 0 0 0 0 0
> class 12, 131072 bytes 0 0 0 0 0 0
> class 13, 262144 bytes 0 0 0 0 0 0
> class 14, 524288 bytes 0 0 0 0 0 0
> total configured streams memory: 25024.00KB
> streams memory in use: 4943.82KB
> maximum streams memory used: 5074.49KB
I de-word-wrapped this to make it legible. When posting preformatted
information, try to convince your posting software not to "helpfully"
destroy the text.
There are no STREAMS failures showing. None in the parts I cut out,
either.
accept() can return ENOSR (Out of stream resources) for some failures
other than actual STREAMS allocation failures. For instance, there are
paths through it where a regular kernel memory allocation (kmem_alloc())
failure would return ENOSR. Is the rest of the system low on memory?
>Bela<
Helene D.
Here is an old tread from comp.databases.informix that covered this
issue for SCO 5.0.2. Hope it's helpful.
-Helene D.
-----------------
> From: Den Miller (dmi...@linuxlab.ru)
> Subject: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/10
>
>
> Hi, all!
>
> This SCO box was running IDS 7.30UC2 and 4GL with no problem. Now I've
> installed another box for IDS
> to increase performance. And now 4GL connects to IDS via tlitcp, not
> shared memory.
>
> The following problem arises:
> When the number of connected users grows up to 100, they began to get
> the following message:
>
> 4GL run-time error number -25574
> cannot open file 'netsrv.iem'.
>
> At the same time the following record is made in /var/adm/syslog when
> users connect:
>
> inetd[]: accept: (for telnet) Out of stream resources
>
> IDS release notes tell to tune NSTREAM, NQUEUE, NUMTIM, NUMTRW.
> SCO docs reference one more parameter: NMUXLINK.
> I've tried to increase all of these, but unsuccessfully.
> 'netstat -m' shows no fails, so I don't know what else parameter I can
> touch.
>
> Can anybody help me?
>
>
>
> From: Christina Southam (christin...@dla-law.co.uk)
> Subject: Re: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/11
>
>
> Den
>
> I recently had a problem on a sco box running informix where once 110
> connections where reached any additional connection attempts from client pcs
> would hang...unlike you however I had no error messages anywhere.
>
> My problem was fixed by increasing the NOFILES SCO kernel param which was
> set to 110.
>
> Christina Southam
>
> Den Miller wrote in message <387A2022...@linuxlab.ru>...
> >Hi, all!
> >
> >This SCO box was running IDS 7.30UC2 and 4GL with no problem. Now I've
> >installed another box for IDS
> >to increase performance. And now 4GL connects to IDS via tlitcp, not
> >shared memory.
> >
> >The following problem arises:
> >When the number of connected users grows up to 100, they began to get
> >the following message:
> >
> >4GL run-time error number -25574
> >cannot open file 'netsrv.iem'.
> >
> >At the same time the following record is made in /var/adm/syslog when
> >users connect:
> >
> >inetd[]: accept: (for telnet) Out of stream resources
> >
> >IDS release notes tell to tune NSTREAM, NQUEUE, NUMTIM, NUMTRW.
> >SCO docs reference one more parameter: NMUXLINK.
> >I've tried to increase all of these, but unsuccessfully.
> >'netstat -m' shows no fails, so I don't know what else parameter I can
> >touch.
> >
> >Can anybody help me?
> >
>
>
>
>
> From: Den Miller (dmi...@linuxlab.ru)
> Subject: Re: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/11
>
>
> Christina,
>
> Unfortunately, it didn't help (I set NOFILES to 220).
>
>
> Christina Southam wrote:
>
> > Den
> >
> > I recently had a problem on a sco box running informix where once 110
> > connections where reached any additional connection attempts from client pcs
> > would hang...unlike you however I had no error messages anywhere.
> >
> > My problem was fixed by increasing the NOFILES SCO kernel param which was
> > set to 110.
> >
> > Christina Southam
> >
>
>
>
> From: richard palmer (pal...@tor.maves.ca)
> Subject: Re: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/11
>
>
>
> Did you try upping the limit on NPROCS yet ? Common problem
> on SCO boxes. Hope this helps.
>
> have a nice day.
>
> Den Miller wrote:
>
> > Christina,
> >
> > Unfortunately, it didn't help (I set NOFILES to 220).
> >
> > Christina Southam wrote:
> >
> > > Den
> > >
> > > I recently had a problem on a sco box running informix where once 110
> > > connections where reached any additional connection attempts from client pcs
> > > would hang...unlike you however I had no error messages anywhere.
> > >
> > > My problem was fixed by increasing the NOFILES SCO kernel param which was
> > > set to 110.
> > >
> > > Christina Southam
> > >
>
>
>
> From: Den Miller (dmi...@linuxlab.ru)
> Subject: Re: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/12
>
>
> Richard,
>
> 5.0.2 has no NPROCS. It has NPROC which is stated redundant. What does it mean?
> If it must set size of process table, then there is MAX_PROC set to 0 (ie grows
> dynamically).
>
>
> richard palmer wrote:
>
> > Did you try upping the limit on NPROCS yet ? Common problem
> > on SCO boxes. Hope this helps.
> >
> > have a nice day.
>
>
>
> From: Michael Hansen (mic...@scot.dsvsamson.dk)
> Subject: Re: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/14
>
>
> Hi,
>
> I had the same problems with my 5.0.4 and IDS 7.31.
>
> 1) Streams
>
> Even though netstat -m dosen't show any problems, i was able
> to solve this problem by changing the STREAMS ressources.
>
> When the streams problem started on my machine
> (2 or 3 times during work hours), there was only
> 2 out of 4mb streams in use, and max used was around 2.5 mb.
>
> I changed 3 or 4 streams parameters, but i think that the
> NSTRPAGES in mtune was the winner. I changed it from 1000 to 4000.
> and after unix_link and boot, i had 16 mb of total stream, and
> now it using around 3-4 mb of streams, during a working day.
> I havent seen the out of strems problem since then.
>
> 2) connect problem
>
> I had a simular problem when more than 100 4gl programs ran at the
> same time.
>
> if the number informix user processes, is geting near the maximum
> number of Open files pr unix proccess, you the see problem -25574
> and proberly -27001 to.
>
> check "onstat -u | wc -l" and "ulimit -a" (if use are using bash),
> or look at NOFILES in mtune.
> SCO default is 110 open files pr process, i changed it to 500 and
> here after no problems.
>
> Regards
> Michael Hansen
> mha...@dsvsamson.dk
>
> In article <387A2022...@linuxlab.ru>, Den Miller wrote:
> >Hi, all!
> >
> >This SCO box was running IDS 7.30UC2 and 4GL with no problem. Now I've
> >installed another box for IDS
> >to increase performance. And now 4GL connects to IDS via tlitcp, not
> >shared memory.
> >
> >The following problem arises:
> >When the number of connected users grows up to 100, they began to get
> >the following message:
> >
> >4GL run-time error number -25574
> >cannot open file 'netsrv.iem'.
> >
> >At the same time the following record is made in /var/adm/syslog when
> >users connect:
> >
> >inetd[]: accept: (for telnet) Out of stream resources
> >
> >IDS release notes tell to tune NSTREAM, NQUEUE, NUMTIM, NUMTRW.
> >SCO docs reference one more parameter: NMUXLINK.
> >I've tried to increase all of these, but unsuccessfully.
> >'netstat -m' shows no fails, so I don't know what else parameter I can
> >touch.
> >
> >Can anybody help me?
> >
>
>
> From: Den Miller (dmi...@linuxlab.ru)
> Subject: Re: Tuning SCO 5.0.2
> Newsgroups: comp.databases.informix
> Date: 2000/01/14
>
>
> Thank you, Michael!
>
> Unfortunately, these parameters didn't help.
> BUT! The problem was solved after increasing 'TCP Connections' in
> 'netconfig' up to 1024 (was 256).
>
> Thank you again.
>
> Michael Hansen wrote:
>
> > Hi,
> >
> > I had the same problems with my 5.0.4 and IDS 7.31.
> >
> > 1) Streams
> >
> > Even though netstat -m dosen't show any problems, i was able
> > to solve this problem by changing the STREAMS ressources.
> >
> > When the streams problem started on my machine
> > (2 or 3 times during work hours), there was only
> > 2 out of 4mb streams in use, and max used was around 2.5 mb.
> >
> > I changed 3 or 4 streams parameters, but i think that the
> > NSTRPAGES in mtune was the winner. I changed it from 1000 to 4000.
> > and after unix_link and boot, i had 16 mb of total stream, and
> > now it using around 3-4 mb of streams, during a working day.
> > I havent seen the out of strems problem since then.
> >
> > 2) connect problem
> >
> > I had a simular problem when more than 100 4gl programs ran at the
> > same time.
> >
> > if the number informix user processes, is geting near the maximum
> > number of Open files pr unix proccess, you the see problem -25574
> > and proberly -27001 to.
> >
> > check "onstat -u | wc -l" and "ulimit -a" (if use are using bash),
> > or look at NOFILES in mtune.
> > SCO default is 110 open files pr process, i changed it to 500 and
> > here after no problems.
> >
> > Regards
> > Michael Hansen
> > mha...@dsvsamson.dk
Gary Quiring
WOW!!! Bela where have you been??
Actually the problem turned out to be the kernal tuning for a 100 user license. The sdevice file only had 4 socket lines and 1 tcp
line.
The problem was fixed by running netconfig and changing TCP connections to 1024.
(Sorry about the post, I changed my line length to 132)
* SCCS IDENTIFICATION
socket Y 256 0 0 0 0 0 0 0
socket Y 256 0 0 0 0 0 0 0
socket Y 256 0 0 0 0 0 0 0
socket Y 256 0 0 0 0 0 0 0
And only 1 tcp line:
* SCCS IDENTIFICATION
tcp Y 256 0 0 0 0 0 0 0
Gary Q