Impact of different RAID configurations on performance

I've benchmarked the disk performance of RAID5 and RAID10 containers, each with EXT3 and Reiserfs. Each test run 3 times.

This is to decide on the optimum configuration for a MySQL server, so I'll be doing some actual MySQL benchmarks with the same settings.

The test machine is a Dell Poweredge 1ghz Pentium III Coppermine, 256KB cache, 512MB RAM, with an Adaptec AIC-7899 Ultra 160/m SCSI host adapter. The OS is a standard RedHat Advanced Server 2.1 installtion.

EXT3 was used in "ordered" mode, and Reiserfs with defaults (i.e: tail mode on).

The tests were done with the Bonnie++ 1.03a disk benchmark software.

RAID 5 (3 disks)

		Sequential Output						Sequential Input				Random Seeks			Sequential Create						Random Create
	Size:Chunk Size	Per Char		Block		Rewrite		Per Char		Block		Random Seeks		Num Files	Create		Read		Delete		Create		Read		Delete
		K/sec	% CPU	K/sec	% CPU	K/sec	% CPU	K/sec	% CPU	K/sec	% CPU	/ sec	% CPU		/ sec	% CPU	/ sec	% CPU	/ sec	% CPU	/ sec	% CPU	/ sec	% CPU	/ sec	% CPU
ext3-ordered	1G			24188	22	13102	10			29531	10	602.1	2	5:4096:1024	360	14	+++++	+++	701	2	387	13	+++++	+++	619	7
ext3-ordered	1G			23307	21	13042	10			30294	11	471.6	2	5:4096:1024	374	14	+++++	+++	692	4	385	13	+++++	+++	620	8
ext3-ordered	1G			23135	20	13234	10			37975	12	587.1	2	5:4096:1024	372	13	+++++	+++	692	4	381	15	+++++	+++	622	8
reiserfs-r5	1G			27369	29	12671	10			26328	11	628.4	3	5:4096:1024	318	16	2824	13	408	13	310	17	3724	17	375	13
reiserfs-r5	1G			25629	26	13007	10			28494	11	628.3	3	5:4096:1024	305	15	3846	17	396	13	291	17	3161	12	382	13
reiserfs-r5	1G			26558	27	13053	9			27644	12	633.5	2	5:4096:1024	302	17	3988	18	398	13	297	16	3949	16	375	13

This shows EXT3 with an avg output (write) of 23,543k/sec (13,126k/sec rewrite), and an avg input (read) of 32,600k/sec. ReiserFS with an avg output of 26,519k/sec (12,910k/sec rewrite), and an avg input of 27,488k/sec.

ReiserFS over EXT3 has a performance change for output of +11% (-2% rewrite) and +4% for input.

RAID 10 (4 disks)

		Sequential Output						Sequential Input				Random Seeks			Sequential Create						Random Create
	Size:Chunk Size	Per Char		Block		Rewrite		Per Char		Block		Random Seeks		Num Files	Create		Read		Delete		Create		Read		Delete
		K/sec	% CPU	K/sec	% CPU	K/sec	% CPU	K/sec	% CPU	K/sec	% CPU	/ sec	% CPU		/ sec	% CPU	/ sec	% CPU	/ sec	% CPU	/ sec	% CPU	/ sec	% CPU	/ sec	% CPU
ext3-ordered	1G			21376	19	16212	13			41710	15	695.1	3	5:4096:1024	376	14	+++++	+++	704	5	386	14	+++++	+++	620	8
ext3-ordered	1G			20112	19	16095	13			44497	15	724.0	3	5:4096:1024	389	13	+++++	+++	703	3	384	14	+++++	+++	620	8
ext3-ordered	1G			20427	19	16205	13			42278	16	749.4	3	5:4096:1024	381	12	+++++	+++	484	3	382	15	+++++	+++	619	10
reiserfs-r5	1G			25794	27	9679	7			39052	15	761.1	3	5:4096:1024	313	17	3807	17	397	12	310	17	4085	19	374	13
reiserfs-r5	1G			23995	25	9824	8			47301	21	778.5	4	5:4096:1024	311	16	3818	17	395	12	313	16	4364	19	372	14
reiserfs-r5	1G			23713	25	9862	7			55341	22	744.1	3	5:4096:1024	311	19	3026	12	404	15	312	16	3937	16	373	13

This shows EXT3 with an avg output of 20,638k/sec (16,170k/sec rewrite), and an avg input of 42,828k/sec. ReiserFS with an avg output of 24,500k/sec (9,788k/sec rewrite), and an avg input of 47,231k/sec.

ReiserFS over EXT3 has a performance change for output of +16% (-40% rewrite) and +10% for input.

RAID 5 vs. RAID 10

For EXT3, RAID10 over RAID5 has a performance change for output of -14% (+23% rewrite), and +31% for input. This shows that for EXT3, RAID 10 is a benefit for reading and rewriting data, but is a hinderance to new output.

For ReiserFS, RAID10 over RAID5 has a performance change for output of +8% (-32% rewrite), and +72% for input. This shows that for ReiserFS, RAID 10 is a slight benefit for writing new data, a massive benefit to reading but quite a serious hinderance to rewriting.

Overall, RAID 10 certainly increases throughput when reading from disks, but can affect different filesystems in different ways for writing.

$Date: 2003/09/09 09:34:42 $
(c) Copyright 2003 John Leach <john at johnleach dot co dot uk >