Subject: Re: Throttling IO Requests in NetBSD via Congestion Control
To: Steven M. Bellovin <firstname.lastname@example.org>
From: Sumantra Kundu <email@example.com>
Date: 08/21/2006 14:40:38
> In the limiting case, that sets
> an upper bound on output rate, regardless of other demand. That would allow other
> processes (which would generate read requests) to make more requeasts.
+ Since we do not correlating the rate of read requests, putting
processes to sleep, independently, indeed places an upper bound on the
write throughput. But, this wil not be the case, if we also monitor
the number of pending read requests (to be implemented in the next
part).. In that case, writes can happily continue if there are not
"enough" reads or slow down otherwise..
> Have you
> explored lowering scheduling priority instead?
No.. but with one reader and one writer in the system, will that
create substantial performance difference? Moreover the reader is
supposed to be running when the writer is sleeping !!!