UNSOLVED
D
De1uK5H8uw12444
3 Posts
0
3589
March 29th, 2012 02:00
Write Performance Raid 5
i have read Performance calculation for Clarrion Flare 30:
id like to make me an idea how to explain the Write Performance in a Clariion Storage:
if we have a Raid 5 Raid Group: (4+1 Disks)
a workload of, for example, 200 IOPS generates on the disks a load of 200*4= 800 IOPS:
i thought about that, my idea is as follows:
every write process writes 64k to one of the 5 Disks, another 3 Disks have to read there data for parity calculation which is written to the last disk: so for the write process of 64k all 5 disks have do move there header, reading and writing.
If that is the case: i d like to know: why is the write cache not collecting 4*64k = 256k and writes it simultaneously to 4 disks and the calculated parity to the last disk: so the workload 200 IOPS could theoretically be: (200*5)/4= 25O IOPS on the disks, not 800 IOPS.
So can you please answere me if my theory is right, and why the write cache can not improve the performance?
How does this scale if we use spanned metalun 20 Disks: 4* (4+1 Raid Groups)
our performance tests confirms that sequential writing 64k is as fast as writing to 5 Disks (4+1),
the 20 Disks only had a better performance on random i/o: how could that be explained? Is the write cache only working on random i/o?
