Messages in this thread | | | Subject | Re: Update to the /dev/random driver | Date | Thu, 30 Apr 1998 21:18:04 -0500 | From | Billy Harvey <> |
| |
... > "Cacheing" some random bytes will help, as we don't do this internally > in the random driver today. ...
This seems like a good place to bring up an idea I have intermittently considered. I have wondered why we don't put the idle task to work a little better. If processes could register themselves to be called only when the system would otherwise do useful work, (or when specifically called), then they could do probable look-ahead work.
An example can be shown with random_bytes. Normally we don't cache any, but why not? As Ted stated, the crypto function dominates the time, so generating two random bytes when only one was called for will not give us much benefit except for the calling overhead. However, having a few bytes around for the next call in advance would make the calling overhead the limiting factor. The time to process such compute-intensive processes is when we have nothing better to do. Then we squirrel them away for later.
There are probably other processes that could benefit from otherwise idle time.
I make this distinction versus the concept of nice-ing a process. For example, the RC5 process on my system still takes up too much of my useful CPU time. I only want some things working when I have *nothing* else to do.
Any comments?
Regards, Billy Harvey
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |