if keys are uri which could go upto a few hundred bytes then md5 would make sense? It all boils down the number of entries per bucket. I could hash on only 8 bytes and use the other 8 bytes for actual comparison.
On 9/7/07, Dustin Sallings <[EMAIL PROTECTED]> wrote: > > > On Sep 7, 2007, at 0:34, Venkatesh KS wrote: > > memcached supports variable length keys. But I am just curious as to why > not use md5 instead of lengthy keys. The keylen requirements for my cache > (which is very similar to squid proxy) is very high and I am planning on > using md5. I will read up to find out about the false positive probability. > But going md5 way will certainly cap the keylen to 16bytes. > > > You certainly can, but md5(``somestring'') is a lot more costly to compute > than ``somestring'' > > -- > Dustin Sallings > > >
