[Ocfs-users] ocfs seek-performance

[Wim Coekaerts]

shouldnt affect db performance.

On Thu, Mar 17, 2005 at 03:17:28PM +0100, Andreas Piesk wrote:
> hi list,
> 
> 
> i have a little problem with 2-node RAC using OCFS. the application running 
> on this cluster does heavily index-based accesses. the data volumes are SAN 
> volumes connected by fibrechannel.
> 
> the throughput does not exceed 10mb/s, average is 7-8 mb/s. i've used 
> 'iostat -x' and got rkB/s=8000 while %util=100% (device was saturated) from 
> kernel's POV.
> 
> 
> i did some benchmarking with Bonnie (http://www.textuality.com/bonnie/):
> 
> # ./Bonnie -d /tmp/ocfstest/ -s 1024 -m bn1rac001 -o_direct -p 1
> Bonnie 1.4: File '/tmp/ocfstest/Bonnie.7166', size: 1073741824, volumes: 1
> Using O_DIRECT for block based I/O
> Writing with putc()...         done:  10816 kB/s  59.1 %CPU
> Rewriting...                   done:  15607 kB/s   2.2 %CPU
> Writing intelligently...       done:  18442 kB/s   2.1 %CPU
> Reading with getc()...         done:  20908 kB/s  60.6 %CPU
> Reading intelligently...       done:  45169 kB/s   4.4 %CPU
> Seeker 1...start 'em...done...
>               ---Sequential Output (nosync)--- ---Sequential Input-- --Rnd 
>               Seek-
>               -Per Char- -DIOBlock- -DRewrite- -Per Char- -DIOBlock- --04k 
>               (01)-
> Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU   /sec 
> %CPU
> bn1rac 1*1024 10816 59.1 18442  2.1 15607  2.2 20908 60.6 45169  4.4   11.3 
> 0.4
> 
> 
> # ./Bonnie -d /root/ext3test/ -s 1024 -m bn1rac001 -o_direct -p 1
> Bonnie 1.4: File '/tmp/ext3test/Bonnie.7843', size: 1073741824, volumes: 1
> Using O_DIRECT for block based I/O
> Writing with putc()...         done:  28471 kB/s  91.6 %CPU
> Rewriting...                   done:  15014 kB/s   5.1 %CPU
> Writing intelligently...       done:  23810 kB/s   7.0 %CPU
> Reading with getc()...         done:  13150 kB/s  42.8 %CPU
> Reading intelligently...       done:  42893 kB/s   3.6 %CPU
> Seeker 1...start 'em...done...
>               ---Sequential Output (nosync)--- ---Sequential Input-- --Rnd 
>               Seek-
>               -Per Char- -DIOBlock- -DRewrite- -Per Char- -DIOBlock- --04k 
>               (01)-
> Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU   /sec 
> %CPU
> bn1rac 1*1024 28471 91.6 23810  7.0 15014  5.1 13150 42.8 42893  3.6  182.4 
> 1.8
> 
> 
> the first run is on ocfs, the second an ext3. both runs used 'O_DIRECT'. 
> the directory '/tmp' is on a local hard disk (takes the SAN out of the 
> game).
> the overall performance is not too bad. writing on ocfs is somewhat slower 
> but reading is faster.
> the random seeks are worst, 11.3 vs. 182.4 /s. i'm not sure what's the 
> reason. Bonnie uses lseek(2) to perform the seek operation.
> 
> maybe this seek-problem is the reason for the insufficient read performance 
> of 8mb/s on index-bases access? is this type of benchmark valid for ocfs? 
> if yes, can someone could repeat it to verify the results?
> 
> thanks for reading.
> 
> 
> Best,
> 
> -ap
> 
> -- 
> Andreas Piesk
> Systemadministration E-/M-Commerce
> --------------------------------------------------------------
> V?B-ZVD
> Bank f?r Zahlungsverkehrsdienstleistungen GmbH
> Godesberger Allee 88
> 
> 53175 Bonn
> 
> Tel.: +49 (0) 228 9377-535
> Fax:  +49 (0) 228 9377-596
> E-Mail: andreas.piesk at voeb-zvd.de
> Internet: http://www.voeb-zvd.de
> 
> _______________________________________________
> Ocfs-users mailing list
> Ocfs-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs-users