scsi vs ide
Michael Green
market we have IDE hards disks, in the blue corner with the 'professional'
appeal and the big ticket we have SCSI.
And the question is why bother? Why pay 3 times as much per MB and have to
put up with all that installation and naming malarky?
Over to you.
Ben Rosenthal
Michael Green wrote:
Well I am standing by the mud pit where these two will be wrestling shortly.
The ring keeper says that the mud is ready as soon as they are.......
Are you serious?
Ben
Tony Lawrence
>
> Yes its the big one, in the red corner weighing in at 90% or more of the
> market we have IDE hards disks, in the blue corner with the 'professional'
> appeal and the big ticket we have SCSI.
>
> And the question is why bother? Why pay 3 times as much per MB and have to
> put up with all that installation and naming malarky?
Because it's faster. Because you can have more of them.
Because it's faster. Because it's faster. Because it's
faster.
Did I mention that it's faster?
--
Tony Lawrence (to...@aplawrence.com)
SCO/Linux articles, help, book reviews, tests,
job listings and more : http://www.pcunix.com
Bill Campbell
>Yes its the big one, in the red corner weighing in at 90% or more of the
>market we have IDE hards disks, in the blue corner with the 'professional'
>appeal and the big ticket we have SCSI.
>
>And the question is why bother? Why pay 3 times as much per MB and have to
>put up with all that installation and naming malarky?
Because SCSI is significantly faster, even in a workstation Linux
environment, and SCSI is more flexible and stable. My current desktop
machine has a 9GB IDE drive in it, and it's significantly slower than some
other systems we have here with slower CPUs, and nominally slower SCSI hard
drives running the same OS, video card, and applications. I don't have
benchmarks, just months of experience on both machines. The difference is
particularly noticeable when running The Gimp and other programs that may
go into swapping even with 128MB of RAM.
As for the installation and naming, what's difficult about it? I must
admit that I started using SCSI on a Tandy 4000 running SCO Xenix in 1987,
have always used SCSI, and install at least three Caldera OpenLinux systems
on SCSI every week so I'm fairly familiar with it. The SCO 5.0.? systems
I've installed have always gone in very easily as well (with the possible
exception of doing it on a Compaq where their fast-start install was
considerably slower than installing on the run-of-the-mill clone system).
I also drive a car with a 5-speed manual transmission because it allows me
better control over the car than an automatic, and is easier on the
equipment (at 170,000 miles I'm on the original clutch, and the brake pads
have been replaced once -- even making the 917 mile trip from Seattle to
Santa Cruz in 12.25 hours).
Bill
--
INTERNET: bi...@Celestial.COM Bill Campbell; Celestial Systems, Inc.
UUCP: camco!bill PO Box 820; 6641 E. Mercer Way
FAX: (206) 232-9186 Mercer Island, WA 98040-0820; (206) 236-1676
URL: http://www.celestial.com/
The day-to-day travails of the IBM programmer are so amusing to most of
us who are fortunate enough never to have been one -- like watching
Charlie Chaplin trying to cook a shoe.
Ken Wolff
>Yes its the big one, in the red corner weighing in at 90% or more of the
>market we have IDE hards disks, in the blue corner with the 'professional'
>appeal and the big ticket we have SCSI.
>
>And the question is why bother? Why pay 3 times as much per MB and have to
>put up with all that installation and naming malarky?
>
90% of what market? The home market, the desktop market or the server
market? Based on # of units installed or total storage space?
All depends on your needs.
From the home and desktop end, IDE should do just fine. But from the
server end, no chance I'd ever put an IDE drive on our servers.
Does your server need any type of hardware RAID? If so, too bad. I'm not
aware of any IDE RAID controller.
Do you need Hot-Swap drives? Again, I'm don't think IDE can handle that
either.
What if you need more than a couple of drives on a system. Maximum # of
IDE drvices is 4. Maximum number of SCSI is limited only by your
controllers. Currently we have 13 SCSI hard drives, 1 SCSI CD-ROM and 5
SCSI tape drives on our SCO server.
Again, depends on your need.
Just like picking a car. If all you're doing is driving a couple of miles
to and from work, hey, get an inexpensive model. But if your going to be
pulling your 35" boat, you'd better plan on something with weight and
horsepower.
Ken
Jerry Heyman
"Michael Green" <michael...@lineone.net> writes:
|> Yes its the big one, in the red corner weighing in at 90% or more of the
|> market we have IDE hards disks, in the blue corner with the 'professional'
|> appeal and the big ticket we have SCSI.
|>
|> And the question is why bother? Why pay 3 times as much per MB and have to
|> put up with all that installation and naming malarky?
Depends on how many devices you want to have. IDE is limited to four (4),
whereas SCSI is limited to 7 per card. On my machine I have a mix of:
2 IDE disk drives
1 SCSI disk drive
1 SCSI Jaz drive
1 SCSI cdrom
1 SCSI tape drive
Yes, disk drives are cheaper IDE, but flexibility of adding more
devices goes to SCSI
|> Over to you.
jerry
--
Jerry Heyman 919.224.1442 | Tivoli Systems |"Software is the
Build Infrastructure Architect | 3901 S Miami Blvd | difference between
Jerry....@tivoli.com | RTP, NC 27709 | hardware and reality"
http://www.cs.stedwards.edu/u/faculty/heyman
Bill Vermillion
>Yes its the big one, in the red corner weighing in at 90% or more
>of the market we have IDE hards disks, in the blue corner with the
>'professional' appeal and the big ticket we have SCSI.
>And the question is why bother? Why pay 3 times as much per MB and
>have to put up with all that installation and naming malarky?
You bother because you are running a multi-tasking OS to which SCSI
is quite suited. You should note that NT also recommends SCSI for
the same reasons.
In a synchronous OS ( writes performed while everything else waits
- as in many MS products ) everything waits while the disk gets
written. But since it is primarily a single user system this
doesn't impact things as greatly. Only the latest IDE drives use
DMA. Many use PIO - which means the CPU has to do all the work of
moving the data from the system to the hard drive.
Cheap SCSI adaptors also use PIO - which is why the lowest level of
cheap adaptors recommend for Unix systems were the 154x series from
Adapted which also use DMA. THe 152x, 151x and 150x series are PIO
and thus rob performance from the system.
There are a lot of other reasons that could be enumerated by you'd
be better served by looking at comp.periphs.scsi - as it address
the probelms on a broader plain.
If you are using Unix as a workstation/desktop OS it might not
matter, and definately the IDE's would be cheaper. The 7200 RPM
IDE drives make an MS systems quite snappy. In a multi-user system
you could possibly get enough small delays to impact production
enough to make the cheaper drives more expensive in just a few
weeks if you count the lost production time of the employees.
answer
--
Bill Vermillion bv @ wjv.com
cdlvj
maybe now it does.
Does anyone know if IDE can map badtracks?
Tony Lawrence <to...@aplawrence.com> wrote in message
news:38D28854...@aplawrence.com...
> Michael Green wrote:
> >
> > Yes its the big one, in the red corner weighing in at 90% or more of the
> > market we have IDE hards disks, in the blue corner with the
'professional'
> > appeal and the big ticket we have SCSI.
> >
> > And the question is why bother? Why pay 3 times as much per MB and have
to
> > put up with all that installation and naming malarky?
>
>
Ken Wolff
>In article <01bf9040$62b810c0$b41b063e@travis>,
>|> Yes its the big one, in the red corner weighing in at 90% or more of the
>|> market we have IDE hards disks, in the blue corner with the 'professional'
>|> appeal and the big ticket we have SCSI.
>|>
>|> And the question is why bother? Why pay 3 times as much per MB and have to
>|> put up with all that installation and naming malarky?
>
>whereas SCSI is limited to 7 per card. On my machine I have a mix of:
The limit of 7 per card depends on the card. We have Adaptec's that
support 15. We also have a 3 channel card that supports 15 per channel.
>2 IDE disk drives
>1 SCSI disk drive
>1 SCSI Jaz drive
>1 SCSI cdrom
>1 SCSI tape drive
>
>Yes, disk drives are cheaper IDE, but flexibility of adding more
>devices goes to SCSI
>
>|> Over to you.
>
>jerry
>--
>Jerry Heyman 919.224.1442 | Tivoli Systems |"Software is the
>Build Infrastructure Architect | 3901 S Miami Blvd | difference between
>Jerry....@tivoli.com | RTP, NC 27709 | hardware and reality"
> http://www.cs.stedwards.edu/u/faculty/heyman
--------------------------------------------------------------
Ken Wolff
Phone: 616-957-4949 Ext: 111
FAX: 616-957-1614
--------------------------------------------------------------
Warren Young
>
> Depends on how many devices you want to have. IDE is limited to four (4),
> whereas SCSI is limited to 7 per card.
Not quite. IDE is limited to two disks per channel. Depending on the
system, you can have as many as four channels, though a limit of two
channels is more common.
The '7' limit is obsolete, too, with SCSI-3. (The 68-pin stuff.)
SCSI-3 allows 15 devices per bus.
--
= Warren -- See the *ix pages at http://www.cyberport.com/~tangent/ix/
=
= ICBM Address: 36.8274040 N, 108.0204086 W, alt. 1714m
Warren Young
>
> Because SCSI is significantly faster, even in a workstation Linux
> environment, and SCSI is more flexible and stable. My current desktop
> machine has a 9GB IDE drive in it, and it's significantly slower than some
> other systems we have here with slower CPUs, and nominally slower SCSI hard
> drives running the same OS, video card, and applications. I don't have
> benchmarks, just months of experience on both machines.
I _do_ have benchmarks. My home Linux box has a newish 18 GB UDMA/33
EIDE hard drive (the drive can do UDMA/66, but the Linux drivers aren't
there yet) and a pair of older 4 GB SCSI-3 (40 MB/s) disks in a software
RAID-0 array. Nominally, these should be head-to-head, roughly, and the
benchmarks bear that out. The RAID-0 of course does much better at
reading, but overall, they do about equally.
However:
1. The IDE disk takes more CPU to hit the same performance numbers the
SCSI disks manage.
2. I've done several huge transfers between the two filesystems. It
seems that most of the time, as it's swapping back and forth between the
drives, that it spends more time accessing the IDE side. You can watch
the LEDs on the front of the case, and the IDE LED will stay lit for a
couple seconds, and then the SCSI one will blink a bit, then back to the
IDE LED. Not conclusive, I know, but I've seen it time and time again.
3. Software RAID-0 on IDE would be madness, because of IDE's
poor-to-nonexistent multitasking and scatter/gather capabilities.
SCSI's the only way to fly for that.
Warren Young
>
> I would also add mapping of bad tracks, something that ide never used to do,
> maybe now it does.
> Does anyone know if IDE can map badtracks?
Bad sectors, yes. Not entire bad tracks, but I can't imagine why you'd
want to ignore an entire track just because a few sectors in it are bad.
If you go to www.grc.com and read about the modern versions of Spinrite,
you see where they talk a lot about all the contortions they have to go
through to get the hard disk to _not_ do its own bad block mapping,
because that confuses Spinrite.
Modern IDE will automatcially translate a request for a bad sector to
the location where it moved the data, without the host controller even
noticing. Spinrite can disable this functionality on some drives, but
others (WDC, cough, cough) won't play along.
Warren Young
>
> Does your server need any type of hardware RAID? If so, too bad. I'm not
> aware of any IDE RAID controller.
There are a few, but they're FM (F***ing Magic) and really new
technology. They're mainly meant for RAID-0 or RAID-1, not RAID-5 and
other interesting setups.
They tend to cost about $60, making them about on the same par as
Adaptec's AAA-13x line of cheap SCSI RAID controllers, or perhaps
Mylex's Flashpoint RAID. Just icky all the way around.... I'd rather
do software RAID: at least then you can throw some intelligence at the
problem. (Automatic failover, drive reconstruction, etc.)
> What if you need more than a couple of drives on a system. Maximum # of
> IDE drvices is 4.
See my other messages in this thread. This info is not correct. Yes,
SCSI is better, but IDE's not as bad as some are saying in this thread.
Bill Vermillion
Jerry Heyman <Jerry....@tivoli.com> wrote:
>In article <01bf9040$62b810c0$b41b063e@travis>,
> "Michael Green" <michael...@lineone.net> writes:
>|> Yes its the big one, in the red corner weighing in at 90% or
>|> more of the market we have IDE hards disks, in the blue corner
>|> with the 'professional' appeal and the big ticket we have SCSI.
>|> And the question is why bother? Why pay 3 times as much per
>|> MB and have to put up with all that installation and naming
>|> malarky?
>Depends on how many devices you want to have. IDE is limited to
>four (4), whereas SCSI is limited to 7 per card.
Not true - the 7 per card.
A single channel narrow SCSI adapater is limited to 7HDs
(providing you don't have a LUN adapter installed).
A single channel wide SCSI adapter will let you have 15 HD's.
Something like the DPT or Adaptec cards with three channels will
give you up to 45 drives. I heard once that someone had a card
with more than 3 channels but I have never had that verified.
MTNBIKE151
SCO Open Server 5.05 on a 40 Gig Maxtor 7200rpm UDMA66 HD. All I can say is
HOLY JEEBUS it is fast and at only $258.00 bucks for the drive it is a steal.
Just my 2 cents
Chris
Panikos Panayi
>(providing you don't have a LUN adapter installed).
>
>A single channel wide SCSI adapter will let you have 15 HD's.
>
>Something like the DPT or Adaptec cards with three channels will
>give you up to 45 drives. I heard once that someone had a card
>with more than 3 channels but I have never had that verified.
>
>
>--
>Bill Vermillion bv @ wjv.com
DELL sell an Adaptec (PERC 2) quad channel card with 128Mb RAM.
Unfortunately no drivers for any flavour of SCO at present time (bought one
and then had to send it back) only NetWare and NT4.
Mike Kenyon
I have worked on a number of IDE machines with SCO Openserver 5.0.4. Both have
>8GB primary master drives. One refused point blank to have a big drive on the primary slave, the other on the secondary master! I presume something must be different in the chipsets. I have no problem setting up SCSI on a couple of other boxes.
I have had so much trouble with IDE under SCO, I would not recommend it unless
you definately have a low load on the system and are only going to have one
drive.
As an aside, I added a second SCSI drive to my Linux system at home (Advansys
930 SCSI adaptor), move /usr onto it and performance seemed to go up by about
20-30%. X-Windows and Netscape certainly loaded quicker. I've made the
investment now, and I'm going to stick with it.
For home use, hunting down second-hand SCSI drives gives you the performance
without the cost, and who needs 40Gb of space at home anyway? :)
--
Mike Kenyon <mke...@promtek.com> Software Engineer for Promtek Ltd
Warren Young
>
> There is no doubt that SCSI in great. But IDE is coming along nicly. I just put
> SCO Open Server 5.05 on a 40 Gig Maxtor 7200rpm UDMA66 HD. All I can say is
> HOLY JEEBUS it is fast and at only $258.00 bucks for the drive it is a steal.
Yes, that's fine if you stick with one drive: multitasking doesn't
matter then. Just add a second drive to the bus, though, and watch
performance drop.
Then move the system to a pair of equivalent SCSI drives. Performance
returns. Joy, joy, hapiness in the server room. :)
And heaven help you should want to add a _third_ drive without eating
another IRQ.
Using IDE in a Unix server is, in a word, shortsighted.
And then let's talk about Maxtor....<wince> Gimme WDC instead any day,
and only that under duress.
cdlvj
I agree and have had the problems with IDE, What you save in price is by far
lost in the manhours making IDE work.
Mike Kenyon <mke...@promtek.com> wrote in message
news:38D624A1...@promtek.com...
> MTNBIKE151 wrote:
> > There is no doubt that SCSI in great. But IDE is coming along nicly. I
just put
> > SCO Open Server 5.05 on a 40 Gig Maxtor 7200rpm UDMA66 HD. All I can say
is
> > HOLY JEEBUS it is fast and at only $258.00 bucks for the drive it is a
steal.
>
Warren Young
>
> For home use, hunting down second-hand SCSI drives gives you the performance
> without the cost, and who needs 40Gb of space at home anyway? :)
I do. Gotta have my MP3 jukebox, accessible from the other hosts on the
LAN. :)
`Course, the 18GB MP3 drive is IDE. This is a fine use for IDE:
low-load, high-capacity. I have a pair of SCSI-3 drives for the main
filesystems. (I think they're set up as / and /usr, at the moment.)
Filip Verborgh
When a SCSI controller has to read from sector eg. 1000, 1560, 7000,1350,
2800, the contoller will do this in ascending or descending order -in this
example- : 1000, 1350, 1560, 2800, 7000 en then place the readings in the
correct order.
An IDE controller does not have this feature and will read from disk in the
order of the requests he gets.
Thats one of the main factors why SCSI is so much faster in multiyser
systems.
Michael Green <michael...@lineone.net> schreef in berichtnieuws
01bf9040$62b810c0$b41b063e@travis...
> Yes its the big one, in the red corner weighing in at 90% or more of the
> market we have IDE hards disks, in the blue corner with the 'professional'
> appeal and the big ticket we have SCSI.
>
> And the question is why bother? Why pay 3 times as much per MB and have to
> put up with all that installation and naming malarky?
>
> Over to you.
Mike Kenyon
> Anyone ever heard of command queueing?
Isn't command queuing only of relevance for more than one drives? The drives
run off with their queues and write the data onto the bus as and when they
find it.
> When a SCSI controller has to read from sector eg. 1000, 1560, 7000,1350,
> 2800, the contoller will do this in ascending or descending order -in this
> example- : 1000, 1350, 1560, 2800, 7000 en then place the readings in the
> correct order.
> An IDE controller does not have this feature and will read from disk in the
> order of the requests he gets.
Isn't that a feature of the O/S disk schedular? (I'm not saying it's not a
feature of the controller.)
Mike Kenyon
> And heaven help you should want to add a _third_ drive without eating
> another IRQ.
Assuming the driver doesn't through a wobbly at the chipset.
> Using IDE in a Unix server is, in a word, shortsighted.
We usually get a handful of disk timeouts every day thanks to IDE.
> And then let's talk about Maxtor....<wince> Gimme WDC instead any day,
> and only that under duress.
So who would people recommend for disks?
Bill Campbell
...
>So who would people recommend for disks?
We have been using IBM SCSI drives almost exclusively for the last three
year or so. We've also had good results with Fujitsu, with very fast
warranty service and a five-year warranty. My experience with other
manufacturers, particularly Seagate and Quantum, hasn't been all that good.
We've used some Western Digital IDE drives on non-critical workstations.
Bill
--
INTERNET: bi...@Celestial.COM Bill Campbell; Celestial Systems, Inc.
UUCP: camco!bill PO Box 820; 6641 E. Mercer Way
FAX: (206) 232-9186 Mercer Island, WA 98040-0820; (206) 236-1676
URL: http://www.celestial.com/
``The children who know how to think for themselves spoil the harmony of
the collective society that is coming, where everyone would be
interdependent.'' 1899 John Dewey, educational philosopher, proponent of
modern public schools.
Warren Young
>
> On Tue, Mar 21, 2000 at 10:18:20AM +0000, Mike Kenyon wrote:
> ...
> >So who would people recommend for disks?
>
> We have been using IBM SCSI drives almost exclusively for the last three
> year or so. We've also had good results with Fujitsu, with very fast
> warranty service and a five-year warranty. My experience with other
> manufacturers, particularly Seagate and Quantum, hasn't been all that good.
>
> We've used some Western Digital IDE drives on non-critical workstations.
Item: Western Digital just got out of the SCSI business, after a whole 2
years (IIRC).
Item: I have two 4.<mumble> GB SCSI-3 disks in my Linux box at home.
They're about the same age, and have been paired as long as I've had
them both, about 2 years. One's an IBM drive, and the other's a Western
Digital. The Western Digital died about 3 months ago, the IBM never.
(Both have 5-year warranties.)
Item: I use this system 20 hours a week tops: that's a 12% real-use duty
cycle. (It's powered-up 100% of the time, however.)
Item: The IBM is the faster of the pair, by a significant margin.
Item: We've got a lot of IDE Western Digital disks here at work that are
doing fine -- no failures at all, over many years. None of these
machines are used heavily, though they all stay on all the time. I carp
about WDC drives, but they're not horrible, at least not the IDE ones.
Item: We have two machines here with a pair of 10,000 RPM Seagate SCSI-3
drives each in them, each pair on a RAID-0 controller. No failures,
after 1 year. These are development machines: a fair amount of big
compiles each day.
If money and availability were not a problem, I'd probably buy IBM
drives all the time. They're very reliable, and consistently among the
fastest in their class. (IBM has many of the most important patents in
hard disk technology.) Both their IDE and SCSI lines are wonderful.
Warren Young
>
> Filip Verborgh wrote:
> > Anyone ever heard of command queueing?
>
> Isn't command queuing only of relevance for more than one drives? The drives
> run off with their queues and write the data onto the bus as and when they
> find it.
All hard drives have a _data_ cache -- command queueing is like an
_instruction_ cache. Command queueing lets the OS send the disk
subsystem several commands, each to be handled in turn.
Advantages:
1. It causes less bus overhead to send a bunch of commands at once than
it does to send each individually.
2. It can be more efficient for the OS to send several commands at once:
depending on the OS's architecture, multiple individual commands might
cause more work for the OS than sending several at once.
SCSI has a further refinement on this idea: "tagged" command queueing.
Each command is associated with a unique tag, somewhat like the sequence
number in TCP/IP. This lets the host adapter and/or the disk drive
handle the commands in the order that's convenient, rather than in the
order they arrived. (The OS sends the tag with the command, and the
disk subsystem sends the tag back with each request's corresponding
reply; the OS uses the tag to figure out what reply goes with what
request.)
The primary advantage to this is that the hard drive often gets requests
for several nearby data blocks at once, perhaps interspersed with other
commands, and probably not in the most efficient order. With TCQ, the
drive can handle those commands in the most efficient order.
If you couple that feature with scatter/gather I/O, you get even more
efficiency. Scattering is sending several blocks of data to the disk
subsystem at once, each of which goes to a different physical area of
the disk. Gathering is similar: the OS asks for a bunch of data, and
the disk reads it all and sends it back to the OS in a single chunk.
The efficiencies are obvious: the drive can reorder reads and writes to
be efficient, and there's less bus bandwidth wasted on SCSI protocol
overhead.
> > When a SCSI controller has to read from sector eg. 1000, 1560, 7000,1350,
> > 2800, the contoller will do this in ascending or descending order -in this
> > example- : 1000, 1350, 1560, 2800, 7000 en then place the readings in the
> > correct order.
> > An IDE controller does not have this feature and will read from disk in the
> > order of the requests he gets.
>
> Isn't that a feature of the O/S disk schedular? (I'm not saying it's not a
> feature of the controller.)
Actually, the OS's scheduler can make this problem worse, by causing
disk requests to become out of order, even if all the running programs'
data access patterns are _in order_. They may each be in-order, but
chances are excellent that each program's in-order stream is on a
different area of the disk -- as the system gives each process a time
slice, it jumps from stream to stream with a naive disk subsystem.
TCQ helps here: the OS doesn't have to worry about reordering disk
requests. It can just pass the requests straight on to the HBA and let
it sort things out.
At first glance, you might think the OS could implement these features,
so they'd work on all disk subsystems, naive or no. High-end OSes
(beyond Win9x's level, at least) do have some similar efficiency
features, but much is still left to the disk subsystem. Here's why:
1. It's not wise to rely on knowing the physical geometry of the hard
disk, because the hard disk can lie about its geometry.
2. The OS can't predict the internal workings of the hard drive. For
example, what track is the drive's head over? What sector is it over on
the track? Is the head changing tracks? Which direction? (The latter
can affect whether it's more efficient to reorder the data accesses in
ascending or descending order, by sector number.)
3. Hardware RAID. `Nuff said. :)
4. It would cause code bloat and eat CPU cycles. Better to save RAM and
cycles for general purpose use (i.e. The Users) and let the hardware
specialize in doing specific things very fast.
Yves Leclerc
will always be faster than IDE! SCSI can allow multiple concurrent I/O
access of any of the 7-15 drives connected to the SCSI host bus. Whereas
IDE must pause one drive in order to access the second drive and vise-versa.
Therefore, in the long run, SCSI is the host bus of choice for any intense
PC disk usage.
SCSI is now at ULTRA160 speeds!!!
As for the installation and naming malarkey, Unix treats IDE drives as SCSI
with a SCSI Host adapter using the 'wd' drivers. The naming scheme still
remains.
Yves
"Michael Green" <michael...@lineone.net> wrote in message
news:01bf9040$62b810c0$b41b063e@travis...