[dm-devel] [PATCH] PROBLEM: SSD access time with dm-crypt is way too high

[Michael Zugelder]

On Mon, 2011-01-10 at 12:19 -0500, Mikulas Patocka wrote: 
> Try this patch.

Hi,

just compiled it on top of 2.6.37 without the fb1e753 and ran seeker [1]
for 100 seconds:

raw disk:       0.2173ms
fb1e753 patch:  0.2643ms
your patch:	0.2676ms

Throughput is within 5% of max, even with the 10ms access time.

On Mon, 2011-01-10 at 18:50 +0100, Milan Broz wrote:
> We can now process ios in parallel on different CPUs, that code will
> rewrite the context struct on every IO and at least it will cause cache bouncing.

I should say that those numbers come from a CRYPTO_PCRYPT=n kernel.


Anyone got an idea for a good testcase? I tried replicating the issues
with with a loop'd file on a tmpfs and failed. Using a usb-attached HDD
also didn't work.


Michael

[1] http://www.linuxinsight.com/how_fast_is_your_disk.html