On Thu, 2005-12-01 at 12:37 -0500, James.Smart@xxxxxxxxxx wrote: > > sure.. but if the defaults are 0... why couldn't mode 1 by > > the default.. > > isn't mode 1 with coalesing value 0 the same as mode 0.. and > > if so.. why > > need a mode 0 in the first place ? > > There's no gain - and in fact there is now a loss. Queuecommand > now eats up more cycles attempting to service i/o completions on > every invokation. Nothing gets helped on the interrupt side. We > would still have interrupts firing, thus the interrupt handler > being invoked to clear the interrupt. > > We're going to consume more cycles for a) queuecommand calls that > don't have i/o completions available, and b) for handling spurious > interrupts. ok fair enough... what kind of granularity can this delay be set too? It might make sense to default to say 50 usec or 100usec always... Also, not sure how your hardware works but it would be really sweet if the kernel could set "this request should be coalesed" or "this request is latency sensitive, gimme immediate", so that we can set async write IO's for coalescing while not getting latency on reads... - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
