[ http://jira.jboss.com/jira/browse/JBCACHE-971?page=all ] Manik Surtani reassigned JBCACHE-971: ------------------------------------- Assignee: Mircea Markus (was: Manik Surtani) -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JBCACHE-971?page=comments#action_12394445 ] Paul Cowan commented on JBCACHE-971: ------------------------------------ Hi all, I have produced an AdaptiveReplacementPolicy implementation as part of a process of determining which eviciton policy best suits our needs, based on a real-life access data set. The code is rather horrible and would need a lot of cleanup to be able to be made available to the world at large (I'm tossing up whether I should contribute it to jboss or whether it should be a separate standalone open-sourced algorithm implementation with an Adapter to implement it as a jboss algorithm) but it seems to work well. On our dataset, it results in an ~10% decrease in cache misses compared to LFU. However, there's a more fundamental issue than code quality which prevents me from even attaching the very rough first-draft patch here -- the ARC algorithm is patented by IBM and hence there's a whole legal can-of-worms involved with doing so (I assume). There are a few patents which cover this area; US patent #6,996,676 is probably the main one I can find. This is the reason Postgres switched from ARC to 2Q (which is, as I understand it, patent-free) as a caching algorithm in version 8.0.2. I am lodging a request with IBM to find out what's involved in obtaining a license to use this algorithm via their contact form at https://www.ibm.com/ibm/licensing/contact/patents.shtml -- I understand that they have a program to grant licenses to open source projects for certain patents -- and I will advise of their response. (The fact that this requires permission from IBM is one reason for possibly making this a non-jboss implementation, with a jboss 'bridge' -- then I might only need to obtain permission from IBM once and people who don't use jboss could still use that open-sourced implementation for other purposes. Fingers crossed they're interested) In the meantime I also intend to look at implementing 2Q for JBoss Cache. -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JBCACHE-971?page=comments#action_12394447 ] Manik Surtani commented on JBCACHE-971: --------------------------------------- Thanks for your efforts, Paul. I look forward to your contributions. -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JBCACHE-971?page=comments#action_12416943 ] Paul Cowan commented on JBCACHE-971: ------------------------------------ For the record (Manik's recent target version change reminded me to update this): I have now developed a cache library which includes implementations of the following algorithms: * ARC * FRC (non-self-adjusting version of ARC; not sure if this is covered by the same patents as ARC) * 2Q * MQ (Multi-Queue Replacement, by Zhou, Philbin and Li -- see http://www.usenix.org/event/usenix01/full_papers/zhou/zhou_html/index.html -- I am not sure if there are any patent encumbrances here, and have contacted 2 of the authors to try and find out) This library has been developed without being jboss-specific; I have produced an adapter for JBoss Cache 1.3 (which we're still stuck using) and then one for JBoss Cache 2.0, which I would be happy to contribute (my employer has given me permission to do so, and is in the process of formalising this internally). HOWEVER, I have not heard anything back from IBM in the last 5 months in relation to the ARC patent issues, which makes me loathe to release this (of course!). Is there anyone at JBoss who would be in a better position to shake the IBM tree and try to extract an answer? I would be very keen to release my library and think JBoss Cache would benefit from it. -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JBCACHE-971?page=comments#action_12417237 ] Manik Surtani commented on JBCACHE-971: --------------------------------------- In the meanwhile, could you contrib the patent-free implementations? I can't expect the legal permission route to be quick in any way. :-) -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira