Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

Version 1 Next »

TERRACOTTA
* Hibernate
** cache raw compenent types, no objects
** 1st level cache (tied to db connection, goes away when db connection
goes away)
** 2nd level cache (avoid going back to db, pluggable, ehcache most
common, terracotta another option)
*** needs to be across cluster for writing data
* ehcache
** option for Hibernate 2nd cache
** uPortal boxes all have to talk to eachother, get 5-10 becomes
problem, exponentially chatty
* Terracotta (open source)
** option for clusterwide Hibernate 2nd cache
*** could use it directly as well
** advantages for clustering, even just 10% worth it
*** seperate service, not run on uPortal boxes
*** less "chatty", all uPortal boxes talk to same terracotta box, this
allows "sharding" (see below)
*** keeps data as Hibernate 2nd level objects (however those are done),
like shared memory for all uPortal boxes "network attached memory" (like
NAS for memory)
*** mark a particular part of application object as "distributed"
*** could be called object oriented db, persists all of its cache to disk
*** data persists until orphaned, then its garbage collection
** use to cluster Tomcat http sessions
*** does it better that Tomcat own session clustering for same
"chattiness" issue
*** still doesn't get around serializable requirement, but has config
option in case you can't serialize in Java code
** Big Savings: "sharding", only possible with less chatty architecture,
terracotta will only push one box's cache to other boxes when other
boxes look for it
? should people worry about clustering with uPortal?
** Cris doesn't see big value to clustering http options, e.g. Madison
only gets 1.7 requests per login, little chance for failover there
** important for Sakai, e.g. things go boom in the middle of an exam
** UofC will be in between
** uPortal, 2nd level Hibernate cache for layout
** trend in uPortal has been that session cache is becoming more
serializable
* Stuff going on
** Sakai
*** Cris used in Sakai instance for John Wiley&Son, will be part of
Sakai 2.7 (code freeze Nov. 20 to meet late Winter/early Spring release)
*** capabilities will be in Sakai, up to implementers to turn it on
*** with it on you should be able to fail over without having to log back in
*** for app specific things, e.g. an exam, you can terracotta-enable
tool by tool, some tools don't have much state and don't need it
** Hibernate 2nd level thing is huge
*** uPortal session is possibility, Eric has been thinking about it
*** Cris interested in finding uPortal instance where big gains can be
made with terracotta
** there is a way to plug terracotta behind an ehcache
** terracotta can be used without Hibernate, use configs to cache
without app even knowing (good and bad)
*** "super static" - local static class attr, cluster in terracotta
config, now static across cluster, not just local JVM
** Webinar on changes made to Sakai October 29th 10:30 am pacific
** alternative to db, does data really need to be in a db? e.g. Sakai
"presence" data
** questions? Talk to Cris, also the Terracotta guys are very accessible

  • No labels