You can "warm" the cache if this is a concern, although 100K*100K item-item pairs won't fit in memory. I'd reserve this for a thousand common items or something like that.
Beyond this you'd probably need to do a little hacking to get exactly the behavior you want. On Tue, Jun 21, 2011 at 4:01 PM, Daniel Xiaodan Zhou <[email protected]>wrote: > Thanks Sean. Some of my clients' sites have 10K-100K items. Do you > know if performance is OK if computing item-item similarity online? > With caching, repeated access is fine. But the first user who triggers > the item-item similarity could experience some delay... > >
