Date: Mon, 12 Feb 2001 01:27:11 -0600 From: Mike Meyer <mwm@mired.org> To: "Ted Mittelstaedt" <tedm@toybox.placo.com> Cc: "Mike Meyer" <mwm@mired.org>, <questions@freebsd.org> Subject: RE: Problems installing 4.x on large disks Message-ID: <14983.36943.315670.474001@guru.mired.org> In-Reply-To: <000901c094bc$f7385300$1401a8c0@tedm.placo.com> References: <14983.29323.218278.757276@guru.mired.org> <000901c094bc$f7385300$1401a8c0@tedm.placo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Ted Mittelstaedt <tedm@toybox.placo.com> types:
> > Ted Mittelstaedt <tedm@toybox.placo.com> types:
> > > OK Mike,
> > > Since your obviously dying for a comparison, here's the
> > > output of bonnie run on one of the spools on our news server.
> > > This is on a stripe set of 3 9GB SCSI disks that has been
> > > built with the command:
> > Actually, I've done my comparison. I'm asking you to check your
> > beliefs against reality. Which your test doesn't do. We both agreed
> Whoah, here! I didn't post those results with the goal of getting into
> "This proves my point is right and yours is wrong" type of debate.
Whoops. Sorry about that.
> I figured that since you have obviously got some systems with the most
> modern IDE drives and busmastering controllers running that you might
Bad assumption. As I already said, I tend to buy the smallest drives I
can find (meaning old), because I simply don't need 40 gig of
space.
> post your results, and as you pointed out we both agreed that a striped
> SCSI set is _supposed_ to be the fastest. If you would post the best
> results you can get off cheap IDE disk servers then not only we but all
> the readers of the list could see just _how much_ different the output
> is. I mean, after all if the fastest IDE/UDMA server you can deploy is
> only 10% slower than my posted results of striped SCSI, then that pretty
> well puts the nail in the coffin of my "gobs of performance" argument
> doesen't it?
Except my IDE drive isn't 10% slower; it's more like 3x faster. Here's
your result:
File './Bonnie.12717', size: 268435456
Writing with putc()...done
Rewriting...done
Writing intelligently...done
Reading with getc()...done
Reading intelligently...done
Seeker 1...Seeker 2...Seeker 3...start 'em...done...done...done...
-------Sequential Output-------- ---Sequential
Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per
Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec
%CPU
256 3820 97.0 8054 88.4 4753 59.5 6434 94.3 9524 27.7 195.3
11.0
#
Here's my IDE test results from the Maxtor I mentioned.
File '/wdtmp/x/Bonnie.515', size: 104857600
Writing with putc()...done
Rewriting...done
Writing intelligently...done
Reading with getc()...done
Reading intelligently...done
Seeker 1...Seeker 2...Seeker 3...start 'em...done...done...done...
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
wd 100 16841 88.5 15186 26.1 16610 38.2 21409 100.0 133543 100.0 3004.7 4
8.6
And I'll be the first to say that this comparison is pretty much
meaningless. The two systems we did this on were *totally* different,
and all the advantages are in the IDE drive favor (except CCD, of
course). His system is running FreeBSD 2.2.8 vs. 3.x (probably 3.3) on
mine; I had freshly formatted file systems, and he was working on a
news spool. I suspect I've got a faster CPU (dual PII/Xeons 400s) and
possibly system bus (100MHz) than his test system did as well, because
a news server doesn't need that kind of horsepower.
> In any case, to do a _true and fair_ comparison would take going out
> and buying 2 drives from the same manufacturer that are identical save
> that one has a SCSI and one has an IDE interface, then do a battery of
Depends on what the question is. If it's "We want to compare SCSI with
IDE", then you're right. Except you need to put the two drives in the
same system, and you need to do something to make sure the two
controllers are in some way comparable.
On the other hand, if the question is "Does an IDE drive have any
place in a Unix workstation", then you ought to use two comparable
drives that people are liable to buy, and ditto for controllers - and
put them in the same box.
I bought a 10Gig IDE drive that was comparable to my nice SCSI system
drive to use as scratch space. I figured I'd do some measurements, so
I would have tests to show people how much better SCSI was than
IDE. They did that, until I turned on UDMA. At which point, the IDE
drive outperformed the SCSI drive. I've since replaced the IDE drive
with a second SCSI drive (that cost nearly twice as much, nearly six
months later :-() to free up the IDE controller's IRQ.
> benchmarking on them with a program like bonnie with different sample
> sizes, multiple sessions of bonnie running, different niceness levels,
> different filesystem frag sizes, etc. I can't do that because I don't
> have the hardware, and I don't have the time to do a good benchmark
> performance test right now. All I can do is post some different results
> from different systems I have and compare them to yours and we can both
> attempt to infer some reasonable guesses from the results.
I didn't do the system interaction testing - I figured the CPU
measurement would provide an indication of that. This will warp the
results for heavily loaded servers, of course. But that's not the kind
of system we're talking about, or that I was interested in.
> > that with multiple disks, SCSI is much faster. Sure, if you want a
> > file server, or a heavily loaded network server, or even a high-end,
> > cost-is-no object workstation, that's the kind of thing you do. But
> > the requirements on those is in no way indicative of a typical unix
> > workstation, which is what your original claim was about:
> Maybe I'm wrong in my definition of a UNIX workstation, but a workstation
> is the type of system that would be specified for a computer for a serious
> user,
> like an engineer, software
> developer, etc. to interface with.
Gee, that sounds like the kinds of boxes I'm thinking about, and
people are discussing here. Dual P-IIIs, or maybe a Gighertz Athelon,
with WNT installed by the IT staff and a nice, big Nokia monitor (ok,
the client I'm thinking of had a special relationship with Nokia),
costing less than any sparcstation, and outperforming most of them?
With a single IDE drive installed in it, of course.
> I don't consider PC's built specifically
> for gaming (ie: highly optimized for video, otherwise made with as cheap
> components as possible) nor do I consider cheap but bland PC's bought for
> the
> office secretary to run Word on as workstations. Unfortunately PC
> marketeers
> are in the process of wrecking the work "workstation" to try to make it to
> mean any PC that is made out of garbage-grade components that they are
> trying to get an extra $200 for. But I figured you knew better.
Well, once you take that "bland PC" and install FreeBSD on it, you've
got a mouse, a GUI, Unix, and a complete software development
environment. Sounds like a Unix workstation to me. It'll also
outperform most of the RISC workstation I've dealt with, even though
they had SCSI. What more do you want?
> You say "cost-is-no object workstation", well that's silly, what do you
> think
> that a "workstation" is? Last I checked, "cost-is-no-object" was an
> integral
> part of the definition of the word "workstation"
It's certainly never been anywhere I've been. Most of them were
business trying to make a profit, and cost was *always* an
object. That's why most Unix workstations (and here I'm talking about
RISC boxes running commercial Unix) I've dealt with only had one disk
instead of striped disks, and so on.
If you want to make that your definition of workstation, then I won't
argue with what you said. Of course that makes your objection pretty
much meaningless to almost everyone running FreeBSD, because they do
worry about cost.
<mike
--
Mike Meyer <mwm@mired.org> http://www.mired.org/home/mwm/
Independent WWW/Perforce/FreeBSD/Unix consultant, email for more information.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?14983.36943.315670.474001>