Linux Software RAID5 Performance

Michael Prinkey mikeprinkey at hotmail.com
Sun Mar 31 11:33:59 PST 2002


Some time ago, a thread discussed the relative performance and stability 
merits of different RAID solutions.  At that time, I gave some results for 
640-GB arrays that I had build using EIDE drives and Software RAID5.  I just 
recently constructed and installed a 1.0-TB array and had some performance 
numbers to share for it as well.  They are interesting for two reasons:  
First, the filesystem in use is ext3, rather than ext2.  Second, the read 
performance is significantly better (almost 2x) than that of the 640-GB 
units.

The system uses 11 120-GB Maxtor 5400-RPM drives, two Promise Ultra66 
controllers, a P4 1.6-GHz CPU, an Intel 850 motherboard, and 512 MB ECC 
RDRAM.  Drives are configured in RAID5 (9 data, 1 parity, 1 hot spare).  
Four drives are on each Promise controller.  Three are on the on-board EIDE 
controller (UDMA100).  A small boot drive is also on the on-board 
controller.  I had intended to use Ultra100 TX2 controllers, but the latest 
EIDE driver updates with TX2 support are not making it into the latest 
kernels (I'm using 2.4.18), so I opted for the older, slower controllers 
rather than patching.  So, I am both cautious and lazy.  8)

Again, performance (see below) is remarkably good, especially considering 
all of the strikes against this configuration:  EIDE instead of SCSI, UDMA66 
instead of 100/133, 5400-RPM instead of 7200-RPM, and master/slave drives on 
each port instead of a single drive per port.  With some hdparm tuning (-c 3 
-u 1), the read performance went from 83 MB/sec to 93 MB/sec.  Write 
performance remained essentially unchanged by tuning at 26 MB/sec.  For 
comparison, the 640-GB arrays gave read performance of about 56 MB/sec, 
write performance of 28.5 MB/sec.

Had I more time, I would have tested ext2 vs ext3 to ascertain how much that 
change effected performance.  Likewise, I was considering the use of a raid1 
array as the ext3 journal device to perhaps improve write performance.  Any 
thoughts?

Regards,

Mike Prinkey
Aeolus Research, Inc.

----------------------

[root at tera /root]# df; mount; cat /proc/mdstat; cat bonnie10.log
Filesystem           1k-blocks      Used Available Use% Mounted on
/dev/hda6             38764268   2601128  34193976   8% /
/dev/hda1               101089      4965     90905   6% /boot
/dev/md0             1063591944  58195936 1005396008   6% /raid
raid640:/raid/home   630296592 284066148 346230444  46% /mnt/tmp
/dev/hda6 on / type ext2 (rw)
none on /proc type proc (rw)
/dev/hda1 on /boot type ext2 (rw)
none on /dev/pts type devpts (rw,gid=5,mode=620)
/dev/md0 on /raid type ext3 (rw)
automount(pid580) on /misc type autofs 
(rw,fd=5,pgrp=580,minproto=2,maxproto=3)
raid640:/raid/home on /mnt/tmp type nfs (rw,addr=192.168.0.123)
Personalities : [raid5]
read_ahead 1024 sectors
md0 : active raid5 hdl1[10] hdk1[9] hdj1[8] hdi1[7] hdh1[6] hdg1[5] hdf1[4] 
hde1[3] hdd1[2] hdc1[1] hdb1[0]
      1080546624 blocks level 5, 32k chunk, algorithm 2 [10/10] [UUUUUUUUUU]

unused devices: <none>
Bonnie 1.2: File '/raid/Bonnie.1027', size: 1048576000, volumes: 10
Writing with putc()...         done:  14810 kB/s  88.9 %CPU
Rewriting...                   done:  22288 kB/s  13.4 %CPU
Writing intelligently...       done:  26438 kB/s  21.7 %CPU
Reading with getc()...         done:  17112 kB/s  97.9 %CPU
Reading intelligently...       done:  93332 kB/s  32.2 %CPU
Seek numbers calculated on first volume only
Seeker 1...Seeker 2...Seeker 3...start 'em...done...done...done...
              ---Sequential Output (nosync)--- ---Sequential Input-- --Rnd 
Seek-
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --04k 
(03)-
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU   /sec 
%CPU
raid05 10*1000 14810 88.9 26438 21.7 22288 13.4 17112 97.9 93332 32.2  206.3 
  2.1


_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp.




More information about the Beowulf mailing list