Help: very slow software RAID 5.
Dean S. Messing
deanm at sharplabs.com
Tue Sep 18 23:04:59 UTC 2007
Alan M. Evans writes:
: On Tue, 2007-09-18 at 18:26 +0200, Test wrote:
: > This off course is very logical...
: >
: > Raid5 writes to all 3 disks at about the same time plus it has to write
: > the crc/verification data which also causes some overhead.
: >
: > so the average speed = 55+71+75 / 3 = 67...
: >
: > So your speed measurement is correct...
:
: I wouldn't have expected such poor performance for *reads*, which is
: what the OP complained about specifically. Even the web link you
: provided below states:
:
: "The read performance of RAID 5 is almost as good as RAID 0 for
: the same number of disks. Except for the parity blocks, the
: distribution of data over the drives follows the same pattern as
: RAID 0."
:
: So RAID5 should, presumably, be able to split the reads over multiple
: disks and achieve much better than disk-average performance when
: reading.
That was roughly my thinking (being a RAID N00BIE)
But even for writes, my thinking is (was?) as follows:
If I write 100 MB of data to the RAID 5, then the 100 MB gets split
(roughly) into
a 50 MB piece for the 55MB/s disk,
a 50 MB piece for the 71MB/s disk, and
a 50 MB piece for the 75MB/s disk.
Two of these pieces are (striped) data, one is parity.
The slowest drive determines the time it takes to complete the write:
T = max( 50MB/(55MB/s) , 50MB/(71MB/s) , 50MB/(75MB/s) ).
So the entire 100 MB of data is written (neglecging parity calcs) in
T = 0.909 seconds
and the avg. data rate is 100MB/.909s = 110 MB/s.
Where's my mistake?
What are others who run software RAID 5 seeing compared to the
individual partition speeds?
Dean
More information about the users
mailing list