I have 8 Western Digital WD15EADS 1.5T green drives in a software
raid6 6+2 config using 1024K chunk size.
They are connected to a SuperMicro SASLP-MV8 card sitting in the 16x
PCI-E slot of an old but stable Asus K8N4E-Deluxe board.
Every single drive appears to read at >90Mbyte/sec:
hdparm -t /dev/sdb
Timing buffered disk reads: 282 MB in 3.02 seconds = 93.49 MB/sec
But when I try to hdparm -t /dev/md2, I max out at around 180Mbyte/sec
and the read seems to be nicely distributed across all drives
according to iostat -k :
avg-cpu: %user %nice %system %iowait %steal %idle
5.00 0.00 95.00 0.00 0.00 0.00
Device: tps kB_read/s kB_wrtn/s kB_read kB_wrtn
sda 0.00 0.00 0.00 0 0
sdb 115.00 23552.00 0.00 23552 0
sdc 110.00 22528.00 0.00 22528 0
sdd 115.00 23552.00 0.00 23552 0
sde 115.00 23552.00 0.00 23552 0
sdf 113.00 23040.00 0.00 23040 0
sdg 112.00 23040.00 0.00 23040 0
sdh 119.00 24324.00 0.00 24324 0
sdi 121.00 24576.00 0.00 24576 0
md2 47783.00 191132.00 0.00 191132 0
sda is my boot drive connected to the onboard nforce, not part of the raidset
When loading the mvsas module, it seems the bandwith is limited:
mvsas 0000:01:00.0: mvsas: PCI-E x4, Bandwidth Usage: 2.5 Gbps
which seems strange as it should be able to do 250Mbyte/sec per lane,
so 4 lanes should amount to 1 Gbyte / sec even for the oldest PCI-E
v1.x standard.
My system is Fedora 12 x86_64 using a custom 2.6.32.3 kernel with the
following patches from Andy Yan:
[PATCH 6/7]MVSAS: Enhanced hot plug handling
[PATCH 5/7]MVSAS:Optimization for DMA buffer
[PATCH 4/7]MVSAS:Make code more flexibe for different chip model.
[PATCH 3/7]MVSAS: bug fix with big endian
[PATCH 2/7]MVSAS:add supporting MSI feature
[PATCH 1/7]MVSAS: Update chip initialization
I did not use [PATCH 7/7]
I can report these patches solved all of my raid6 stability issues
(drives kicking out of the raid, /proc/mdstat not reporting faults,
xfs corruption) and have been running stable for the past 3 weeks with
all sorts of stress testing.
Am I missing something with regards to tuning to get the full
bandwidth out of this marvell controller ?
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
