Thread (15 messages) 15 messages, 7 authors, 2010-07-20

Re: messed up changing chunk size

From: Roman Mamedov <hidden>
Date: 2010-07-19 14:58:16

On Mon, 19 Jul 2010 14:24:12 +0100
Jools Wills [off-list ref] wrote:
quoted
quoted
I looked around and found that chunk size
of 512 should work better.
Not true, at least with RAID5/6 a chunk size of 64K performs faster, see
http://louwrentius.blogspot.com/2010/05/raid-level-and-chunk-size-benchmarks.html
http://alephnull.com/benchmarks/sata2009/chunksize.html
There is no mention in this benchmark tests if he modified the stripe 
cache. With an increased stripe size a larger stripe cache would be 
needed for good performance (at least from my experience)
That's true, but increasing the stripe cache helps across all chunk sizes, even
with relatively low 128K, see [1]. And I don't think that a large-chunk(512K)
configuration will significantly, if at all, outperform a small-chunks(64K)
one at any given stripe cache size (set to the same value in both cases).

There's also an opinion I heard from a couple of sources, that the stripe size
better be chosen so that either one stripe, or a whole stride fits inside the
CPU's L2 or L3 cache. Sounds logical, though I am not convinced that this is
what causes the performance difference between various stripe sizes.


[1]
http://peterkieser.com/2009/11/29/raid-mdraid-stripe_cache_size-vs-write-transfer/

-- 
With respect,
Roman

Attachments

Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help