On 1/10/06, Boyd Stephen Smith Jr. <[EMAIL PROTECTED]> wrote: > On Tuesday 10 January 2006 07:13, Cláudio Henrique > <[EMAIL PROTECTED]> wrote about 'Re: [gentoo-user] LUKS': > > What about the performance, is it too different from plain partition > > usage? > > I never noticed the difference when I was using aes-loop on a 2GHz laptop. > That said, it will depend on the algorithm you choose and the CPU you have > available. Also, I /think/ aes-loop was supposed to be faster than > dm-crypt, but I believe the kernel's implementation of aes (and maybe > other ciphers) has gotten faster since the last benchmarks I saw.
I tested this recently on my new AMD64 X2 system. The dm-crypt and loop-aes are very very close in performance. I can't really say which is faster, because for some configurations, dm-crypt was faster, while for others, loop-AES was faster. By configurations I mean using 2 disks, software raid, LVM, and dm-crypt/loop-aes, and playing with the order of the "layers" (do I make a raid of 2 encrypted disks, or encrypt a raid array of 2 disks, or ...), the block sizes, etc. And in some cases, loop-aes would be faster at writing, but dm-crypt would be faster at reading, or vice-versa. The one thing I think loop-aes does better is that it creates a separate thread for each encrypted device, so it can take advantage of SMP systems. Still, I ended up using dm-crypt+luks on that system. For performance, on the AMD64 box, the two disks could deliver a combined read throughput of around 130MB/sec. The highest throughput I got with dm-crypt or loop-aes was 115-118MB/sec read, 95MB/sec write. On my 2.13Ghz laptop, using loop-AES, the disk can only deliver a maximum of 50MB/sec, and loop-aes tops out at about 45MB/sec at 42% CPU utilization. The only time it becomes a real impact is when I am doing a backup, when I have decrypt the data from one disk, archive it, compress it, and then encrypt the archive when it is written to another disk. I do _not_ notice an impact when compiling, becase the amount of disk activity for a typical compile is insignificant compared to the CPU usage of the compiler. -Richard -- [email protected] mailing list
