Skip to contentSkip to conversation

2.6.9: performance issues on Via Epia

DISCUSSION Alistair John Strachan 3 participants 4 messages Oct 19, 2004 – Oct 20, 2004

lore
Share & tools
lore
#0 lore raw

Hi,

I recently upgraded from 2.6.8.1 to 2.6.9 (the release, not -final) on my Via Epia 5000 router. Now when I transfer files from the machine's HD vsftpd can only achieve 3MB/s.

I believe this is some performance problem specifically related to XFS, or something specific to the local VM, because if I transfer from an NFS mounted directory on the same machine, vsftpd easily achieves the 10MB/s I'm used to.

Top shows something typical to this during transfers from the machine's local HD;

Cpu(s): 0.7% us, 9.2% sy, 0.0% ni, 0.3% id, 84.5% wa, 5.3% hi, 0.0% si

Which seems like an awful lot of wait time. Anybody got any suggestions of where to start reverting patches? The amount of difference between 2.6.8.1 and 2.6.9 is quite daunting.

By the way, copying a file locally on the system from the same partition to another directory is far more efficient.

[root] 16:02 [~] time cp /var/cache/swapfile here
/var/cache/swapfile' -> here'

real    0m37.904s
user    0m0.115s
sys     0m13.033s
Signature · 8 lines
Cheers,
Alistair.

personal:   alistair()devzero!co!uk
university: s0348365()sms!ed!ac!uk
student:    CS/AI Undergraduate
contact:    1F2 55 South Clerk Street,
            Edinburgh. EH8 9PP.
#1 lore raw
Quoted 8 lines · Alistair John Strachan

Hi,

I recently upgraded from 2.6.8.1 to 2.6.9 (the release, not -final) on my Via Epia 5000 router. Now when I transfer files from the machine's HD vsftpd can only achieve 3MB/s.

I believe this is some performance problem specifically related to XFS, or something specific to the local VM, because if I transfer from an NFS mounted directory on the same machine, vsftpd easily achieves the 10MB/s I'm used to.

Top shows something typical to this during transfers from the machine's local HD;

Quoted 8 lines

Cpu(s): 0.7% us, 9.2% sy, 0.0% ni, 0.3% id, 84.5% wa, 5.3% hi, 0.0% si

Which seems like an awful lot of wait time. Anybody got any suggestions of where to start reverting patches? The amount of difference between 2.6.8.1 and 2.6.9 is quite daunting.

By the way, copying a file locally on the system from the same partition to another directory is far more efficient.

[root] 16:02 [~] time cp /var/cache/swapfile here
/var/cache/swapfile' -> here'

check ethernet too...

Jeff

#2 lore raw
In reply to Alistair John Strachan
Quoted 9 lines · Alistair John Strachan

Hi,

I recently upgraded from 2.6.8.1 to 2.6.9 (the release, not -final) on my Via Epia 5000 router. Now when I transfer files from the machine's HD vsftpd can only achieve 3MB/s.

I believe this is some performance problem specifically related to XFS, or something specific to the local VM, because if I transfer from an NFS mounted directory on the same machine, vsftpd easily achieves the 10MB/s I'm used to.

Sound like 'DMA off' problem.

Quoted 8 lines

Top shows something typical to this during transfers from the machine's local HD;

Cpu(s): 0.7% us, 9.2% sy, 0.0% ni, 0.3% id, 84.5% wa, 5.3% hi, 0.0% si

Which seems like an awful lot of wait time. Anybody got any suggestions of where to start reverting patches? The amount of difference between 2.6.8.1 and 2.6.9 is quite daunting.

Binary search is converging quickly.

Quoted 9 lines

By the way, copying a file locally on the system from the same partition to another directory is far more efficient.

[root] 16:02 [~] time cp /var/cache/swapfile here
/var/cache/swapfile' -> here'

real    0m37.904s
user    0m0.115s
sys     0m13.033s

size of this file?

#3 lore raw
Quoted 7 lines · you
Quoted 6 lines · Alistair John Strachan

Hi,

I recently upgraded from 2.6.8.1 to 2.6.9 (the release, not -final) on my Via Epia 5000 router. Now when I transfer files from the machine's HD vsftpd can only achieve 3MB/s.

Quoted 6 lines
Quoted 4 lines

I believe this is some performance problem specifically related to XFS, or something specific to the local VM, because if I transfer from an NFS mounted directory on the same machine, vsftpd easily achieves the 10MB/s I'm used to.

Sound like 'DMA off' problem.

Could be, not changed anything since 2.6.8.1 though and dmesg claims dma is enabled on hda, which is the only connected drive

Quoted 4 lines
Quoted 4 lines

Top shows something typical to this during transfers from the machine's local HD;

Quoted 8 lines
Quoted 6 lines

Cpu(s): 0.7% us, 9.2% sy, 0.0% ni, 0.3% id, 84.5% wa, 5.3% hi, 0.0% si

Which seems like an awful lot of wait time. Anybody got any suggestions of where to start reverting patches? The amount of difference between 2.6.8.1 and 2.6.9 is quite daunting.

Binary search is converging quickly.

That's still a lot of work, but thanks.. I'll probably just work through the bk snapshots until it breaks, then binary search on the remaining patches. It'll still take a couple of hours, especially on such a slow machine.

Quoted 4 lines
Quoted 4 lines

By the way, copying a file locally on the system from the same partition to another directory is far more efficient.

Quoted 8 lines
Quoted 6 lines

[root] 16:02 [~] time cp /var/cache/swapfile here
/var/cache/swapfile' -> here'

real    0m37.904s
user    0m0.115s
sys     0m13.033s

size of this file?

512MB, sorry about that..

By the way Jeff, I already said in my original post that copying from a local NFS mount gives me to get the full 10MB/s i.e.

NFS server X -> problem case machine's NFS mount -> some client machine = fine problem case machine's local mount -> some client machine = slow.

My email was just an attempt to stir up any obvious changes that might've taken place. Since the problem seems obscure, I'll probably have to do it properly and manually search through the patches until I find the problem one.

Thanks for your prompt replies.

Signature · 8 lines
Cheers,
Alistair.

personal:   alistair()devzero!co!uk
university: s0348365()sms!ed!ac!uk
student:    CS/AI Undergraduate
contact:    1F2 55 South Clerk Street,
            Edinburgh. EH8 9PP.