On Thu, 2007-07-12 at 17:53 +0100, Sam Mason wrote: > > OTOH, disk I/O scheduling is considered a hard problem. > > Guaranteeing bandwidth/latency to specific processes seems hard to me.
Actually, guaranteeing bandwidth/latency isn't particularly hard. What is hard is preserving any sort of sensible behavior for the rest of the system at the same time. The problem with disks is not high latency per se -- we have very good models for seek latency -- as high *variance*. The dominant latency is actually rotational these days, and the problem is that the rotational delay is not terribly predictable. Problem is that this can raise or lower the time required for the operation by almost a factor of two, so you get forced into very conservative resource allocation here. -- Jonathan S. Shapiro, Ph.D. Managing Director The EROS Group, LLC _______________________________________________ L4-hurd mailing list [email protected] http://lists.gnu.org/mailman/listinfo/l4-hurd
