On 09/04/13 20:07, Tyler Hicks wrote:
Make the switch from the blkcipher kernel crypto interface to the ablkcipher interface. encrypt_scatterlist() and decrypt_scatterlist() now use the ablkcipher interface but, from the eCryptfs standpoint, still treat the crypto operation as a synchronous operation. They submit the async request and then wait until the operation is finished before they return. Most of the changes are contained inside those two functions. Despite waiting for the completion of the crypto operation, the ablkcipher interface provides performance increases in most cases when used on AES-NI capable hardware. However, sequential I/O with one or two threads on slow storage media does exhibit a sizeable decrease in performance.
With the testing I've run with the ablkcipher patch I observe a noticeable performance win with fast media such as SSD or tmpfs and a slight improvement with fast 7200 rpm HDDs. With slower HDDs, I don't see any negative performance impact with ablkcipher. Exercised on a recent AES-NI capable Intel 4 core server with 7200 rpm HDD and Intel 330 SSD and also with a Intel i3-2350M Lenovo X220i laptop with a 5400 rpm HDD and Intel 330 SSD.
I have also given this patch considerable amount of soak testing overnight with continuous parallel kernel builds while also running the eCryptfs tests without any issues.
As it stands, I am happy with v3 of this patch. -- To unsubscribe from this list: send the line "unsubscribe ecryptfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
