On Wed, 16 Jun 2010 10:45:00 +1200, Robert Collins <[email protected]> wrote: > Well its written in an entertaining and a little condescending style. > The class of algorithmic analysis that is relevant is 'cache oblivious > algorithms' and is a hot topic at the moment. > > Well worth reading and thinking about. > > -Rob
Aye, with a fair bit of the FUD mixed in as well. The first adopters were facing Squid-2.5, which is at best 50% slower than the current Squid. There is one line he uses which is quite telling: "Once created, objects are often cached for weeks if not months, and therefore the binary heap may not be updated even once per minute; on some sites not even once per hour." This is not the behaviour we see in regular Squid installs. Where >60% of objects are created one second, expired the next. The first step towards making a cache as fast as that is to remove the Squid insistence that everything gets passed through Store. Outright blocking the non-cacheable objects from going near store would probably be best. If you recall this a 4-way b-heap is essentially what we were discussing for a chunked memory replacement algorithm before deciding to drop chunking and let the VM do the work with its probably better implementation instead. Amos
