lkml.org 
[lkml]   [2004]   [Oct]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectIO performance problems with 2.6.9
From
Date
Hi,

I was testing Linux performance on 80 drives disk array today
(8 hardware RAID volumes stripped by Linux RAID0 256K stripe)

I've initially spotted this by very poor performance with DBT2
benchmark, while later isolated test to "SysBench"
http://sysbench.sourceforge.net/, running random read/write tests
on single 40G file.

The system is dual Opteron 2.4 with 8G of memory.

ReiserFS file system (data=writeback,noatime,notail) was used.

Results I'm getting in IOs/Sec with 256 concurrent threads are:

READ WRITE
Buffered IO 16K 6185 2496
O_DIRECT 16K 5169 471
Buffered IO 4K 7091 4758
O_DIRECT 4K 7676 571


Note I had only 1Gb FC connection so 16K reads saturated about 5xxx/sec


For Comparison I also did the run with SUSE SLES Kernel (2.6.5-7.97SMP)


READ MIXED (R/W)
Buffered IO 16K 2375 2520
O_DIRECT 16K 5391 1052
Buffered IO 4K 9576 5736
O_DIRECT 4K 8063 1041

Sorry I did not do Exactly the same run due to lack of time accessing
hardware.

The issues with these results are:

1) SuSE kernel does "Buffered" reads in 4K reads instead of 16K reads.
This can be seen in "iostat -x". This seems to be fixed in 2.6.9

2) 2.6.9 does writes in 4K requests instead of 16K. This also can be
seen in "iostat -x"

3) O_DIRECT writes are simply broken :)
One can see in vmstat only 1 thread will be in "b" if this mode is used
and only one outstanding request will be submitted to SCSI controller at
the same time. Interesting enough if I use 128 files totaling the same
size O_DIRECT shows good performance. I thought inode locking problem
was removed in 2.6 :(

4) Unrelated but still unfortunate. 2.6.9 kernel seems to behave weird
if swap is disabled. I was running with 8G of memory allocating 6G for
MySQL buffers which left about 1.5G for kernel, 1G of which was used
for file cache. During IO intensive run, using buffered IO I got
"kswapd" running like a crazy taking 90% of CPU time on one of CPUs.
What for is it running if there is no swap files enabled ?



--
Peter Zaitsev, Senior Support Engineer
MySQL AB, www.mysql.com



-
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 14:07    [W:0.035 / U:0.124 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site