questions about your distribution of Linux

Seth Mos seth at arosa.nl
Thu Oct 18 08:08:01 CDT 2001


At 03:16 18-10-2001 +0200, Chris Chabot wrote:
>Seth, on your 6x72Gb 10Krpm/raid10, try kernel 2.4.11+, for me this made
>a _HUGE_ difference in performance.
>
>the read spead on my 2550's (4x10Krpm 18gb drives, perc 3/si, 128Mb
>cache) went up from ~ 20Mb/Sec to 60-70Mb/Sec.. a _HUGE_ performance
>fix..

I have a AMI megaraid controller. Write speeds went up from 34 to 52 but 
the read speeds went up only 2 MB/s

Here is the bonnie++ output.

2.4.13-pre3-xfs
Version  1.02       ------Sequential Output------ --Sequential Input- --Random-
                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
coltex-2             4G 15120  95 32377  19 18147  14 12335  82 52344  17 
558.5   3
                     ------Sequential Create------ --------Random 
Create--------
                     -Create-- --Read--- -Delete-- -Create-- --Read--- 
-Delete--
               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
%CP  /sec %CP
                  16  4563  66 +++++ +++  4962  58  4529  63 +++++ 
+++  3129  44

2.4.8-mdk-xfs
Version  1.02       ------Sequential Output------ --Sequential Input- --Random-
                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
coltex-2             4G 15920  99 30269  26 15825  12 10242  69 34218  15 
544.6   2
                     ------Sequential Create------ --------Random 
Create--------
                     -Create-- --Read--- -Delete-- -Create-- --Read--- 
-Delete--
               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec 
%CP  /sec %CP
                  16  4496  63 +++++ +++  4979  58  4506  66 +++++ 
+++  3130  41


>the change logs between 2.4.10 (still slow) and 2.4.11 do show some
>block device changes, my theory is that somewhere in that they speeded
>this up majorly..

Or the driver fits the kernel better. I am not seeing this drastic jump in 
speed. Software raid is still a heck of alot faster.

>i'm hoping that adaptec's new drivers, when released (they have been
>re-coding those 'linux instead of nt style' for a while now, so lets
>hope they are not vapor ware).

I believe it when I see it. In the mean time I will use the AMI megaraid 
cards. I can't wait for adaptec to build me the drivers and test them 
thoroughly.

>those are also the drivers that are targeted for kernel inclusion, the
>current ones are considered to crappy ;-)

They are. The drivers are ported from NT to Solaris to Linux. Bad idea.

>Anyways, try that latest & greatest kernel, i'm hoping you will be as
>pleasently suprised as me ;-)

It's a bit faster and it hangs the box very fast as well. The new VM 
inserted into 2.4.10+ is abolute crap with HIGHMEM and SMP on the PE 2500. 
I see deadlocks, oopses, IO stalls, and hard crashes.
With the older pre 2.4.10 kernels everything is fine and the box does not 
die with more then 4 decent IO streams.

A few reasons I am with the 2.4.8 Mandrake kernel from 8.1:

This kernel is relatively stable and well tested.
It is newer.
It is 40% faster then the 2.4.2 kernel.
It includes drivers for the bcm5700 standard.
It uses the old pre 2.4.10 VM.
It includes XFS support.

Cheers

--
Seth
Every program has two purposes one for which
it was written and another for which it wasn't
I use the last kind.




More information about the Linux-PowerEdge mailing list