On Fri, Oct 31, 2008 at 6:16 PM, MP <[EMAIL PROTECTED]> wrote: > If I were to use memcached (where memcache servers are different from > the app servers), then on every request I'll have make multiple > network calls (from app server to memcache servers) to fetch the > cached data. >
There's no problem whatsoever with that approach, on our web application, http://nonoba.com, we do between 20-50 memcached requests per page, depending on the page, and depending on you being logged in or not, and some of those requests are ~50-key multi-gets. What we see is that a single database request on a page is much slower than all those memcached requests together. We try to minimize the size of the data we put in memcached though. I don't know how large your page fragments are, but something you could experiment with is to cache all fragments on each local webserver, and only cache a cache-key in memcached, i.e. a two-layered cache where the items are held in local memory, and memcache is used to synchronize these. /Henrik
