[dm-crypt] Re : Re : Poor performances with nfs and Kernel 3.x

Matthias Schniedermeyer ms at citd.de
Thu Feb 9 10:34:23 CET 2012


On 09.02.2012 08:37, Arno Wagner wrote:
> Possible fixes:
> 4) get a faster CPU, or one with at least 2 cores if this is 
>    a single core.

An AMD Bulldozer(AFAIR) or Intel Core i5 (of Westmere or newer 
generation, so Core i5 5xx, or Core i5/7 2xxx) or better would have 
AES-NI. Hardware Acceleration helps tremendously.

I have tested copying between a Core i7 2600 and a Core i5 560, i get 
100MB/s either with NFS(v3) or via scp (with the 
intel-acceleration-engine for openssl, so openssl can use AES-NI which 
it doesn't "as is", altough i also get the necessary speed if i use 
"arcfour").

The Kernel i tested with was 3.2 on both sides.
I ran the test from HDD over Gigabit to tmpfs.

To be precise i have to say that i ran the test with loop-aes, but i 
expect i will get comparable speed after i switch to dm-crypt.

When i compared loop-aes to dm-crypt for raw-speed on a tmpfs i got more 
speed out of dm-crypt(cipher=aes:64-cbc-lmk)).
I got 360MB/s for loop-aes vs. 540 MB/s for dm-crypt. But loop-aes 
needed less cpu. 1 core with 100% for loop-aes vs. 1 core with 100% and 
3 cores with 66% from dm-crypt.
As only about 20% of max-speed are needed for 100MB/s i think i would 
get about the same speed if i had used dm-crypt.







Bis denn

-- 
Real Programmers consider "what you see is what you get" to be just as 
bad a concept in Text Editors as it is in women. No, the Real Programmer
wants a "you asked for it, you got it" text editor -- complicated, 
cryptic, powerful, unforgiving, dangerous.



More information about the dm-crypt mailing list