lkml.org 
[lkml]   [2001]   [Oct]   [31]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: The good, the bad & the ugly (or VM, block devices, and SCSI :-)


On Wed, 31 Oct 2001, Stephan von Krawczynski wrote:

> Hello all,
>
> this is a message especially for:
>
> Gerard Roudier <groudier@club-internet.fr>, symbios driver
> Justin T. Gibbs, adaptec driver
> Andrea and Rik, VM
> Linus, the man with the big picture :-)
>
> Everything started with a note from a friend who tried some small all-in-one
> box with a sym53c1010 onboard. He (an all-time adaptec user like me) told me
> the small box feels like flying, compared to his at-home box (with adaptec).
> This made me curious for trying myself. I bought a Tekram DC390 U3W (which is
> in fact a relabeled U3D) with symbios chipset. I simply replaced the controller
> in my box (compiled a kernel with both drivers of course) and gave it a try
> with bonnie. It did not look impressing. Effectively adaptec A29160 and DC390
> had the same read and write speeds, only noticeable was that tekram performed
> twice as many seeks as adaptec. This was reproducable in all bonnie-tests. Hm,
> that should not make the big difference. Anyway I was too lazy to put the
> adaptec back in and continued working (for several days). Today it hit me:
> As Linus said something about testing pre6 I gave it a try and did the usual
> nfs-copy, cd read test. I was pretty astonished to see the tekram perform very
> well under heavy I/O load, here are the numbers:
>
> A29160: symbios:
>
> cd read without nfs-load: cd read without nfs-load
> 2998,9 kB 3619,3 kB
> 3168,2 kB 3611,1 kB
> 2968,4 kB 3620,2 kB
>
> cd read with nfs load: cd read with nfs load
> 1926,2 kB 3408,1 kB
> 2123,4 kB 3395,2 kB
> 2539,4 kB 3605,1 kB
> 2631,9 kB 3605,8 kB
>
> The rest of the hardware involved is completely the same, only the controller
> boards got exchanged. Another thing quite remarkable: during symbios tests the
> network throughput derived from nfs load is _higher_ and looks more stable.
> Whereas during adaptec the whole picture looks like having hiccup. More to say:
> starting an application during the tests results in waiting a bit (some 10-20
> seconds) with tekram, but waiting pretty long (or even forever, "the ugly" ;-)
> while using adaptec. This is particularly interesting for the vm guys since all
> the scene is in high vm load with around 3-5 MB of free mem and a damn lot of
> page cache. So if you try something around vm I can only urge you to perform
> tests that do _no_ I/O at all, because you may be greatly bitten by your
> controller (or its driver).
> Another thing to mention: during the last cd-read tests with tekram setup I
> already have been deeply impressed by the driver, so I decided to stress it
> some more and start applications (like mozilla) in the background. And in the
> end I was even more impressed, because it turned out (you can see in the last
> two figures), that it got even _faster_. Obviously I cannot explain why.
> If anybody wants me to test anything, feel free to ask.
>
> My personal opinion: Justin has work to do.

Agreed here. Justin should write a clean SCSI access method for Linux for
free as he did for FreeBSD. :-)

Just considering the CD read thoughtput differences, we cannot get any
useful information that applies to software driver differences from your
report. Given the very low throughput it involves (about 3 MB/s) compared
to the capabilities of the controllers (160 MB/s), the results should be
explainable by something related to difference in configuration or to some
hardware or kernel weirdness. I cannot believe a single second that the
difference is due to the software drivers.

Thanks, anyway, for your report.

Regards,
Gérard.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:12    [W:0.052 / U:1.976 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site