new kernel = megaraid performance cut in half
Hello,
I have a dell server that has a perc 3 DCL (megaraid) controller in it running redhat 7.2. I was using the default kernel but I decided to upgrade to 2.4.18 to get the new VM code and for some quota fixes. Anyway, I compiled the kernel and the machine boots fine but my disk speed cut in half.
Is there anything I can do to resolve the speed issue so that I can continue using the 2.4.18 kernel? This box is serving mail so 21MB sec on 3x10000rpm scsi3 disks in a raid 5 array is not acceptable.
Thanks for the help,
schu
[root@mail_test linux]# uname -a
Linux mail_test 2.4.18 #1 SMP Mon May 13 18:39:45 AKDT 2002 i686 unknown [root@mail_test linux]# hdparm -t /dev/sda
/dev/sda:
Timing buffered disk reads: 64 MB in 3.04 seconds = 21.05 MB/sec
Where on the default redhat kernel I get:
[root@mail_test root]# uname -a
Linux mail_test 2.4.7-10smp #1 SMP Thu Sep 6 17:09:31 EDT 2001 i686 unknown [root@mail_test root]# hdparm -t /dev/sda
/dev/sda:
Timing buffered disk reads: 64 MB in 1.71 seconds = 37.43 MB/sec
Quoted 4 lines
running redhat 7.2. I was using the default kernel but I decided to upgrade to 2.4.18 to get the new VM code and for some quota fixes. Anyway, I compiled the kernel and the machine boots fine but my disk speed cut in half.
Do you have large amounts of memory ? The Red Hat kernel includes patches that vastly up performance of DMA capable controllers when more than 1Gb of RAM is in use.
Alan
Alan,
I am running exactly 1g of ram in the machine. Is there a way to apply those patches manually or should I get the kernel-2.4.18-4.i686.rpm from here: http://rhn.redhat.com/errata/RHBA-2002-085.html.
Thanks for the quick response.
schu
Quoted 8 lines · Alan Cox
Quoted 1 line
running redhat 7.2. I was using the default kernel but I decided to upgrade to 2.4.18 to get the new VM code and for some quota fixes. Anyway, I compiled the kernel and the machine boots fine but my disk speed cut in half.
Do you have large amounts of memory ? The Red Hat kernel includes patches that vastly up performance of DMA capable controllers when more than 1Gb of RAM is in use.
Alan
Quoted 3 lines
I am running exactly 1g of ram in the machine. Is there a way to apply those patches manually or should I get the kernel-2.4.18-4.i686.rpm from
here: http://rhn.redhat.com/errata/RHBA-2002-085.html.
Your best bet is probably to grab the RH errata kernel. If you get the source rpm that also includes the patches for some controllers to do I/O from all of memory.
Hopefully that restores the speed
Alan,
FYI,
[root@mail_test root]# uname -a
Linux mail_test 2.4.18-4smp #1 SMP Thu May 2 18:32:34 EDT 2002 i686 unknown [root@mail_test root]# hdparm -t /dev/sda
/dev/sda:
Timing buffered disk reads: 64 MB in 1.22 seconds = 52.46 MB/sec
[root@mail_test root]# hdparm -t /dev/sda
So the redhat 2.4.18 kernel reads disk almost 2.5 times better on my system.... sounds like something that would be handy in a stable kernel release.
Thanks for you help,
schu
Quoted 8 lines · Alan Cox
Quoted 1 line
I am running exactly 1g of ram in the machine. Is there a way to apply those patches manually or should I get the kernel-2.4.18-4.i686.rpm from here: http://rhn.redhat.com/errata/RHBA-2002-085.html.
Your best bet is probably to grab the RH errata kernel. If you get the source rpm that also includes the patches for some controllers to do I/O from all of memory.
Hopefully that restores the speed
Quoted 6 lines
Timing buffered disk reads: 64 MB in 1.22 seconds = 52.46 MB/sec
[root@mail_test root]# hdparm -t/dev/sdaSo the redhat 2.4.18 kernel reads disk almost 2.5 times better on my system.... sounds like something that would be handy in a stable kernel release.
Agreed, however thats something that really has to be post 2.4.19 so it gets a lot of testing when merged with the base tree to make sure it has no dependancies on other changes
Quoted 3 lines · Matthew Schumacher
Is there anything I can do to resolve the speed issue so that I can continue using the 2.4.18 kernel? This box is serving mail so 21MB sec on 3x10000rpm scsi3 disks in a raid 5 array is not acceptable.
I am running 7.3 (upgraded from 7.2) and I feel that the 2.4.18 kernel with the new VM code is actually inferior to that of 2.4.9-31. I have lots of memory (1.2 gig) and lots of disk access, and I see more swap and worse interactive performance on 2.4.18 (redhat supplied kernel) as compared to the redhat supplied 2.4.9-31 for 7.2.
Signature · 1 line
Mikael Abrahamsson email: [email protected]
Quoted 11 lines · Mikael Abrahamsson
Quoted 1 line · Matthew Schumacher
Is there anything I can do to resolve the speed issue so that I can continue using the 2.4.18 kernel? This box is serving mail so 21MB sec on 3x10000rpm scsi3 disks in a raid 5 array is not acceptable.
I am running 7.3 (upgraded from 7.2) and I feel that the 2.4.18 kernel
with the new VM code is actually inferior to that of 2.4.9-31. I have lots
of memory (1.2 gig) and lots of disk access, and I see more swap and worse
interactive performance on 2.4.18 (redhat supplied kernel) as compared to the redhat supplied 2.4.9-31 for 7.2.
Hmm, the reason why I upgraded to 2.4.18 in the first place was to fix a bunch of problems with quota. I wonder if 2.4.9-31 will work with the quotas correctly, I was on 2.4.7.
schu
Quoted 9 lines
Quoted 5 lines
I am running 7.3 (upgraded from 7.2) and I feel that the 2.4.18 kernel with the new VM code is actually inferior to that of 2.4.9-31. I have lots of memory (1.2 gig) and lots of disk access, and I see more swap and worse interactive performance on 2.4.18 (redhat supplied kernel) as compared to the redhat supplied 2.4.9-31 for 7.2.
Hmm, the reason why I upgraded to 2.4.18 in the first place was to fix a bunch of problems with quota. I wonder if 2.4.9-31 will work with the quotas correctly, I was on 2.4.7.
Quite a few quota race fixes are after 2.4.9. Its also going to depend entirely on your workload what works best.
Quoted 10 lines · Alan Cox
Quoted 6 lines
Quoted 4 lines
I am running 7.3 (upgraded from 7.2) and I feel that the 2.4.18 kernel
with the new VM code is actually inferior to that of 2.4.9-31. I have lots
of memory (1.2 gig) and lots of disk access, and I see more swap and worse
interactive performance on 2.4.18 (redhat supplied kernel) as compared to the redhat supplied 2.4.9-31 for 7.2.Hmm, the reason why I upgraded to 2.4.18 in the first place was to fix a bunch of problems with quota. I wonder if 2.4.9-31 will work with the quotas correctly, I was on 2.4.7.
Quite a few quota race fixes are after 2.4.9. Its also going to depend entirely on your workload what works best.
Hmm, sounds like I am going to go with 2.4.18 on redhat 7.3 then. I really don't want any quota trouble.
schu