File:Graid-performance.png - Revision history

Jimbo: add raw data for Linux RAID4

2007-12-29T07:55:29Z

add raw data for Linux RAID4

Jimbo at 06:20, 28 December 2007

2007-12-28T06:20:55Z

Jimbo at 06:15, 28 December 2007

2007-12-28T06:15:16Z

Jimbo: raw data

2007-12-28T04:38:54Z

raw data

Jimbo: add raw data

2007-12-28T04:35:49Z

add raw data

Jimbo at 04:08, 27 December 2007

2007-12-27T04:08:12Z

Jimbo at 02:07, 27 December 2007

2007-12-27T02:07:39Z

Jimbo at 02:04, 27 December 2007

2007-12-27T02:04:21Z

Jimbo at 02:00, 27 December 2007

2007-12-27T02:00:32Z

New page

@@ Line 15: / Line 15: @@
 (using dd if=/dev/random bs=16m count=200).
-For these tests, read-ahead cache value was GREATLY increased from 6.2-RELEASE's default of 8: '''sysctl -w vfs.read_max=128'''.  Linux's read-ahead cache value was similarly greatly increased from 256 to 4096 using '''hdparm /dev/md0 -a4192'''.  I noticed Linux automatically bumped the default 256 value up to 1024 for the RAID5 array, so I retested it and discovered the "sweet spot" for a 5-member Linux RAID5 array to be 8192.
+For these tests, read-ahead cache value was GREATLY increased from 6.2-RELEASE's default of 8: '''sysctl -w vfs.read_max=128'''.  Linux's read-ahead cache value was similarly greatly increased from 256 to 4096 using '''hdparm /dev/md0 -a4192'''.  I noticed Linux automatically bumped the default 256 value up to 1024 for the RAID5 array, so I retested it and discovered the "sweet spot" for a 5-member Linux RAID5 or RAID4 array to be 8192.
 The "-r" option (use parity drive on reads) for graid3 was also tested but typically resulted in slight to severe performance decrease, so was not included on the graph, although its raw data is available below.
@@ Line 21: / Line 21: @@
 Raw data follows:
   Linux RAID5: 3x250/2x500		161	173	163	164	173	* Hdparm -a 8192
   Linux RAID1: 3x250/2x500		68	152	213	251	258	* Hdparm -a 4096

@@ Line 15: / Line 15: @@
 (using dd if=/dev/random bs=16m count=200).
-For these tests, read-ahead cache value was GREATLY increased from 6.2-RELEASE's default of 8: '''sysctl -w vfs.read_max=128'''.  Linux's read-ahead cache value was similarly greatly increased from 256 to 4192 using '''hdparm /dev/md0 -a4192'''.
+For these tests, read-ahead cache value was GREATLY increased from 6.2-RELEASE's default of 8: '''sysctl -w vfs.read_max=128'''.  Linux's read-ahead cache value was similarly greatly increased from 256 to 4096 using '''hdparm /dev/md0 -a4192'''.  I noticed Linux automatically bumped the default 256 value up to 1024 for the RAID5 array, so I retested it and discovered the "sweet spot" for a 5-member Linux RAID5 array to be 8192.
 The "-r" option (use parity drive on reads) for graid3 was also tested but typically resulted in slight to severe performance decrease, so was not included on the graph, although its raw data is available below.

@@ Line 21: / Line 21: @@
 Raw data follows:
   Linux RAID1: 3x250/2x500		68	152	213	251	258	* Hdparm -a 4096
   Nvidia onboard RAID0+1: 3x250/1x500	97	142	160	151	144	* 7s delta on 2-process copy... 66 second delta on 5-process copy

@@ Line 2: / Line 2: @@
   '''System: '''
-     amd64 FreeBSD 6.2-RELEASE
+     FreeBSD OS: amd64 6.2-RELEASE
-    amd64 Ubuntu Server 7.04
+      Linux OS: amd64 Ubuntu Server 7.04
      Athlon X2 5000+, 2GB RAM
 x Western Digital 250GB drives (WDC WD2500JS-22NCB1 10.02E02 SATA-300)
 x Western Digital 500GB drives (WDC WD5000AAKS-00YGA0 12.01C02)
-     Nvidia nForce onboard RAID controller, Promise TX2300 RAID controller
+     Nvidia nForce onboard RAID controller
 Each process shown in the graph is a cp to /dev/null of an individually generated 3200MB random binary file
@@ Line 19: / Line 21: @@
 Raw data follows:
   Nvidia onboard RAID0+1: 3x250/1x500	97	142	160	151	144	* 7s delta on 2-process copy... 66 second delta on 5-process copy
   Graid3: 3x250/2x500			213	137	106	105	105
   Graid3: 3x250/2x500 -r			187	144	103	105	107
   Graid3: 3x250				115	98	77	76	75
   Graid3: 3x250 -r			116	88	79	76	75

@@ Line 2: / Line 2: @@
   '''System: '''
-     amd64 6.2-RELEASE, Athlon X2 5000+, 2GB RAM
+     amd64 FreeBSD 6.2-RELEASE
 x Western Digital 250GB drives (WDC WD2500JS-22NCB1 10.02E02 SATA-300)
 x Western Digital 500GB drives (WDC WD5000AAKS-00YGA0 12.01C02)
@@ Line 10: / Line 13: @@
 (using dd if=/dev/random bs=16m count=200).
-For these tests, read-ahead cache value was GREATLY increased from 6.2-RELEASE's default of 8: '''sysctl -w vfs.read_max=128'''.
+For these tests, read-ahead cache value was GREATLY increased from 6.2-RELEASE's default of 8: '''sysctl -w vfs.read_max=128'''.  Linux's read-ahead cache value was similarly greatly increased from 256 to 4192 using '''hdparm /dev/md0 -a4192'''.
-The "split" balance algorithm for gmirror was also tested but resulted in universally horrible performance, so was not included on the graph.
+The "-r" option (use parity drive on reads) for graid3 was also tested but typically resulted in slight to severe performance decrease, so was not included on the graph, although its raw data is available below.
 The "-r" option (use parity drive on reads) for graid3 was also tested but typically resulted in slight to severe performance decrease, so was not included on the graph.
 Raw data follows:
   Graid3: 3x250/2x500			213	137	106	105	105
   Graid3: 3x250/2x500 -r			187	144	103	105	107
   Nvidia onboard RAID1: 2x250		60	108	90	96	100	* 15 secs delta on 2-process copy / 38 seconds on 5-process!
   Baseline: 1x500			76	70	72	72	72
   Gmirror: 3x250/2x500 round-robin	63	73	69	71	72

@@ Line 1: / Line 1: @@
   Graid3: 3x250/2x500			213	137	106	105	105
@@ Line 14: / Line 29: @@
   Gmirror: 2x250 split 128k		55	47	47	47	48
   Gmirror: 3x250/2x500 split 128k	54	50	47	47	47

← Older revision	Revision as of 02:04, 27 December 2007
Line 1:	Line 1:

	+	Graid3: 3x250/2x500 213 137 106 105 105
	+	Graid3: 3x250/2x500 -r 187 144 103 105 107
	+	Nvidia onboard RAID1: 2x250 60 108 90 96 100 * 15 secs delta on 2-process copy / 38 seconds on 5-process!
	+	Graid3: 3x250 -r 116 88 79 76 75
	+	Graid3: 3x250 115 98 77 76 75
	+	Baseline: 1x500 76 70 72 72 72
	+	Gmirror: 3x250/2x500 round-robin 63 73 69 71 72
	+	Promise TX-2300 RAID1: 2x250 61 52 54 55 57 * 15 sec delta on 5-process copy
	+	Gmirror: 3x250 round-robin 59 53 53 56 55
	+	Gmirror: 2x250 round-robin 58 49 51 51 50
	+	Baseline: 1x250 56 56 55 55 56
	+	Gmirror: 3x250 split 128k 55 46 49 55 54
	+	Gmirror: 2x250 split 128k 55 47 47 47 48
	+	Gmirror: 3x250/2x500 split 128k 54 50 47 47 47