Zbr's days.
July
Sun Mon Tue Wed Thu Fri Sat
   
7
   
2008
Months
Jul
Nov Dec

About :: TODO :: Blog :: RSS :: Old blog :: Projects :: GIT :: Gallery :: Notes

Mon, 07 Jul 2008

POHMELFS crypto processing performance.

If you expected a miracle, it did not happen, so I just present a picture, where I compared plain async in-kernel NFS server (no encryption, no checksumming) versus POHMELFS, which performed SHA1 hashing and AES-128-CBC encryption of the whole data channel.
Block size used in iozone test is 8KB, filesize - 8GB, 1GB of RAM.

Encrypted + hashed POHMELFS vs plain NFS

/devel/fs :: Link / Comments (4)

Adam Langley wrote at 2008-07-07 18:34:

You can probably regain some of the perform hit by using faster crypto primitives. Try Salsa20 for the cipher (already in the stock kernels) and, maybe Poly1305 with AES or Salsa20 as the MAC. Poly1305 might be an issue due to the use of the FP registers in the reference implementation, in which you could case investigate UHASH8.

Piotr K. K. Michaloski wrote at 2008-07-07 18:49:

POHMELFS is the most promising network file system out there. When it will be ready for users? Do you want to push it into mainline kernel?

Zbr wrote at 2008-07-07 22:38:

>Piotr I do not know, maybe its time to push it upstream, but I do not want to bother with linux kernel politics. We will see soon.

>Adam I used likely the most common setup which can be installed. Sure one can use different and less CPU-hungry algorithms. I also do not like Dr. Bernstein programming and/or his communication mode, although he is great mathematician for sure.

Piotr K. K. Michaloski wrote at 2008-07-07 23:34:

"but I do not want to bother with linux kernel politics"

I understand your reluctance for pushing your code upstream - how many of your projects burned down in lkml flames? (kevent? distributed storage?)

Please solve this captcha to be allowed to post (need to reload in a minute): 17 * 77

Comments are closed for this story.