On Sun, Feb 09, 2003 at 11:13:53 +0100, David Olofson wrote: > On Sunday 09 February 2003 01.18, Steve Harris wrote: > [...silence and RT...] > > I suspect (but have no intention of finding out) that suddenly > > waking up a large subnet will have large chace effects too, so it > > wont be a case of just the CPU load varying. > > I suspect that unless most of the memory used by each "mode" of the > net fits in the cache, the difference won't be all that big, since > every block cycle causes massive amounts of cache misses anyway. > Also, as soon as you have hand disk recording going on, you'll have a > butler thread throwing RT data out of the cache all the time. (Same > effect that has surprized some RTL and/or RTAI users by causing > *higher* latencies for the RT threads on faster CPUs.)
There is also the instruction cache to think of. Practical experience suggests that suddenly waking up a chain of plugins causes a spike of CPU load which settles down. It may be possible to moderate this with carefull coding however. - Steve