We need to make the disk behavior configurable, that is, whether or not you want items to go to disk only when the memory size is reached, or if you want all items to go to disk regardless of how much room is left in the memory auxiliary.
--- Timo Nentwig <[EMAIL PROTECTED]> wrote: > Aaron Smuts wrote: > >> I forgot to mention that we wrote our own disk > cache > >> which simply > >> stores every cache item in a single file on disk > >> (groups are directories). > > > > I thought of making an implementation like that, > but > > it can make too many files. . . . The indexed > disk > > Some ten thousands, yes. But since most of the cache > is held > in memory this is no problem. > > > cache is the fastest way to go. It's keeps a > handle > > on the data file and just skips into position to > read. > > We emphasize on persistence. We also patched JCS to > write cache items > immediately to disk and not wait until memory "flows > over". > > It's important for us that in case of server crashes > or whatever we > can restart the server and it gets up - which it > doesn't in high-traffic > situation without cache. > > > Yes, of course. You don't use your OLTP database > for > > the cache. You run additional MySQL instances for > > cached data . . . For JCS I run a remote server > > cluster. Each is backed by its own MySQL instance > > running on the same box as the JCS server. > > Okay, so I'm going to have a shot at the NFS idea > and if it > fails I'll try this. > > Thanks a lot! > Timo > > --------------------------------------------------------------------- > To unsubscribe, e-mail: > [EMAIL PROTECTED] > For additional commands, e-mail: > [EMAIL PROTECTED] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
