Skip to contentSkip to conversation

Write throughput in >= 2.4.10

DISCUSSION Zlatko Calusic 2 participants 3 messages Oct 18, 2001

lore
Share & tools
lore
#0 lore raw

It looks like recent kernels have some serious trouble during simple writing of files. Throughput is cut to half.

2.4.12-ac3 (Riel VM):

procs memory swap io system cpu

 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 0  4  1      0   2548   7652 369104   0   0     0 21992  291   592   0  23  77
 0  4  1      0   2556   7652 369104   0   0     0 22500  280   175   0   3  97
 0  4  1      0   3064   7652 368588   0   0     8 19644  278   202   0   4  96

...

2.4.13-pre4 (Andrea/Linus VM):

procs memory swap io system cpu

 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 1  1  1  30312   2884   2304 384076   0  28     0  7784  199   241   0  12  88
 0  1  1  30316   2068   2256 385772   0 132     4  7900  186   188   1   8  91
 0  1  0  30316   3960   2232 384140   0   0     8  6744  179   204   0   4  96

...

'bo' column is the one to check out... I copied just 3 lines, but they are all alike. 2.4.13-pre ends up with 11MB/sec, where -ac kernels are over 20MB/sec (during sequential writing of big files - ext2 of course).

Also it looks like pre4 swaps when it is not necessary to do so. With 380MB in page cache I don't expect any swap traffic at all.

Signature · 1 line
Zlatko
#1 lore raw
Quoted 19 lines · Zlatko Calusic

It looks like recent kernels have some serious trouble during simple writing of files. Throughput is cut to half.

2.4.12-ac3 (Riel VM):

procs memory swap io system cpu

 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 0  4  1      0   2548   7652 369104   0   0     0 21992  291   592   0  23  77
 0  4  1      0   2556   7652 369104   0   0     0 22500  280   175   0   3  97
 0  4  1      0   3064   7652 368588   0   0     8 19644  278   202   0   4  96

...

2.4.13-pre4 (Andrea/Linus VM):

procs memory swap io system cpu

 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 1  1  1  30312   2884   2304 384076   0  28     0  7784  199   241   0  12  88
 0  1  1  30316   2068   2256 385772   0 132     4  7900  186   188   1   8  91
 0  1  0  30316   3960   2232 384140   0   0     8  6744  179   204   0   4  96
Quoted 8 lines

...

'bo' column is the one to check out... I copied just 3 lines, but they are all alike. 2.4.13-pre ends up with 11MB/sec, where -ac kernels are over 20MB/sec (during sequential writing of big files - ext2 of course).

Also it looks like pre4 swaps when it is not necessary to do so. With 380MB in page cache I don't expect any swap traffic at all.

Zlatko,

Could you please try 2.4.12-pre3 instead pre4 ?

Linus has made some changes and I want to see if those are partly responsible for the problems you're seeing.

Thanks

#2 lore raw
Quoted 5 lines · Marcelo Tosatti
Quoted 3 lines · Zlatko Calusic

It looks like recent kernels have some serious trouble during simple writing of files. Throughput is cut to half.

[snip]

Quoted 8 lines

Zlatko,

Could you please try 2.4.12-pre3 instead pre4 ?

Linus has made some changes and I want to see if those are partly responsible for the problems you're seeing.

pre3 behaves exactly the same, so pre4 is not the culprit.

Signature · 1 line
Zlatko