[jboss-user] [JBossCache] - Re: Config Question
manik.surtani@jboss.com
do-not-reply at jboss.com
Tue Jan 22 16:03:27 EST 2008
"lovelyliatroim" wrote :
| I have a cache configured with a JDBCCacheLoader and in my cache i have a region where i dont want any nodes in the cache memory i.e they will be retrieved through the cacheloader. How do i configure the region to have no nodes in memory since if i set it to 0 then this means no limit?? Maybe -1??
|
Definitely a limitation. Could you please create a feature request in JIRA for this and vote for it?
"lovelyliatroim" wrote :
| Ok next question. Im running some tests where im writing 32kb image to the DB through JBossCache. This is the region where i would like it to have no nodes in memory but currently have it configured to 1. Now if i write say 5000 objects/blobs to the cache/db i get a good throughput figure around 95 per second, however if i run the same test for a sustained period of time this figure drops down to around 20 per sec. Thats a big drop. Now I have a feeling that 2 things might be causing or attributing to the drop. First being the asynch mode, havent looked to see how it works under the hood but i assume that the write requests are buffered somewhere in memory and the heavier the load the more write requests will be buffered, when i run 5,000 writes, its a short burst but when i run the test over a sustained period all these writes are adding on to the asynch buffer thus increasing the load in memory and slowing down throughput. 2nd thing that i think is causing a problem is the eviction of the 1 node i have in memory. When the test is running i get a flury of debug messages saying "node added" and then it pauses for 1-2 seconds and then i start seeing
| anonymous wrote :
| | Eviction of /image/3457 timed out, retrying later
| |
| And it generally comes in groups of 3ôs and then carries on writing, pauses again, outputs the above and carries on again. Now if i could configure my region to tell the cache to have no nodes in memory then this would probably remove this problem or is there something else i can do??
| Any thoughts on why the big drop over time??
|
What could be your problem is how often your eviction thread kicks in. If you have enough memory to deal with temporary spikes before the eviction thread clears down unused nodes, I'd set this not to run too frequently.
"lovelyliatroim" wrote :
| Another question i was thinking about, is there any way to give the asynch buffer more time i.e to catch up or how exactly does the asynch mode work under the hood.
|
I don't think it is replication, although you could tune it by using a ReplicationQueue (by default, even async replication happens immediately; the caller thread just doesn't wait for the response). See the user guide/sample configs for how to use the ReplicationQueue to defer this; it can be triggered either by time or size.
"lovelyliatroim" wrote :
| And my last question. ;) Its off topic a bit, do you know what JBoss AS 5.0 milestones are i.e when the final production release will be.(I see the road map but no dates http://jira.jboss.com/jira/browse/JBAS?report=com.atlassian.jira.plugin.system.project:roadmap-panel ) In a bit of dilema as to whether to use JBoss Cache 1.4 with JBoss AS 4.2.x or go with JBoss Cache 2.0 and AS 5.0 Beta3 at the moment!! My production release date is May....any thoughts on it, proôs or cons??
|
Sorry - that is a question for the JBoss AS guys. :-) JBoss AS will most likely ship with JBoss Cache 2.1.0, BTW.
Also, if you are using the cache directly (i.e., packaged in your webapp or ear), you could even use 2.0.0/2.1.0 with JBoss AS 4.x. See http://wiki.jboss.org/wiki/Wiki.jsp?page=JBossCacheAsCompatibility
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4122387#4122387
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4122387
More information about the jboss-user
mailing list