-
1. Re: Evictions with low number of entries
vblagojevic Sep 10, 2013 10:41 AM (in response to smalone)Sean, yeah this is possible in LIRS if I recall correctly. I would have to review algorithm in details to confirm but in the meantime here is one question for you: why even turn on eviction if you only have 85 possible entries?
Regards,
Vladimir
-
2. Re: Evictions with low number of entries
smalone Sep 13, 2013 12:44 PM (in response to vblagojevic)Hey Vladimir,
Thanks for the response. We have a cache on this table because of the number of queries being ran on that key. It is about 60 million queries a day. Removing the cache would be quite detrimental to our system. So, if you're saying that the LIRS algorithm can cause the number of actual entries in the cache to be substantially lower than the max entries size, what is a good heuristic for setting our max size?
-
3. Re: Evictions with low number of entries
vblagojevic Sep 13, 2013 1:07 PM (in response to smalone)Sean,
I am not saying to remove cache just turn eviction off for your cache. Configure your cache without eviction [1]. Does this make sense or are we not understanding each other? :-(
Regards,
Vladimir
[1] Eviction - Infinispan 6.0 - Project Documentation Editor
-
4. Re: Evictions with low number of entries
smalone Sep 13, 2013 1:22 PM (in response to vblagojevic)Ah! I'm sorry. I misread your comment. We cannot turn eviction off for all of our caches. We have several hundred caches. I was only using that one as an example. It's happening to all of them. We have memory constraints that require us to use evictions. I was just wondering why this case might be happening.
-
5. Re: Evictions with low number of entries
vblagojevic Sep 17, 2013 11:02 AM (in response to smalone)Yes LIRS will evict on segments and I think because your max entries is 1000 many more segments are created than actually needed. Try lowering max entries and/or try some other eviction algorithm.
Regards,
Vladimir
-
6. Re: Evictions with low number of entries
smalone Sep 17, 2013 11:17 AM (in response to vblagojevic)Hey Vladimir,
Thanks for responding again. So maybe I'm misunderstanding what the documentation says. It seemed like it was saying that segments are on a higher level than the max entries. So there could potentially be multiple cache maps in a single segment. Are you saying that this is incorrect and that in fact there are multiple segments in a single cache map? So this would mean that depending on the size of the cache could determine how many segments there are? If this is the case, what determines the size of a segment in order to determine how many reside in a cache map?
Thanks,
Sean