JBUG.be Infinispan
•
0 likes•356 views
- Galder Zamarreño is a core developer for Infinispan and JBoss Cache and contributes to several other JBoss projects - Infinispan is a highly scalable open source data grid platform that is based on JBoss Cache but is more memory efficient and scalable through improvements like using a flat concurrent map instead of a tree structure - New features in Infinispan include support for distributed caching, asynchronous APIs, eager locking, a client/server module, and fine-grained persistence modeling
Recommended
More Related Content
What's hot
What's hot (19)
Viewers also liked
Similar to JBUG.be Infinispan
Similar to JBUG.be Infinispan (20)
Recently uploaded
Recently uploaded (20)
JBUG.be Infinispan
- 1. Galder Zamarreño Infinispan Core Developer JBoss by Red Hat JBUG Belgium, 9th July 2009
- 2. Who am I? Core developer for Infinispan and JBoss Cache Contributor and com itter on JBoss AS, m JGroups, Hibernate, JBoss Portal, etc.
- 3. Agenda W is Infinispan? hat Relationship w JBoss Cache ith Newfeatures Demo
- 4. What is Infinispan? Highly scalable data grid platform 100% open source licensed (LGPL) Based on som JBoss Cache code e JBoss Cache = Tree structured cache Replicated using JGroups Supports JTA, evictions, cache stores, etc. NewJSR-107(JCACHE) compatible API Cache extends Map Tree adapter API available for legacy apps
- 5. More scalable than JBoss Cache Internal structures m m ory efficient ore em Tree --> Flat concurrent map Eviction queue --> Ordered container Marshalling based on JBoss Marshalling Smaller payloads + Poolable streams Early benchmarks Significant performance improvements
- 6. Memory Consumption Comparison Test: Put 2 million serial objs into cache JBoss Cache 3.1: W 2gb, 1 m ith illion objs
- 7. Memory Consumption Comparison (2) Infinispan 4.0.0.Alpha4: W 700m 2 m ith b, illion objs
- 8. ...borrowing best bits from JBoss Cache Multiversion Concurrency Control (MVCC) Newlocking strategy in JBoss Cache 3.0 Readers never locked! Writers w on copy of cache entry ork Non-blocking state transfer Senders generate state without stopping Crucial when state is large
- 9. New features - distributed cache Consistent hash based distribution W allowus to scale to bigger clusters ill Lightweight, L1 cache for efficient reads On writes, L1 invalidated Dynam rebalancing ic Pluggable consistent hashing algorithms Already available in 4.0.0.Alpha5!
- 10. New features - asynchronous API putAsync(), putIfAbsentAsync() Do not block, return a j.u.c.Future Future.get() blocks till call completes Best of both sync and async worlds Future.get() provides sync guarantees Greater parallelism Already available in 4.0.0.Alpha5!
- 11. New features - Eager locking By default: locks acquired at com it tim m e Problematic if updating a shared counter New Acquiring locks eagerly in cluster : Explicit: via API cache.lock(k) // acquire cluster wide lock on k Implicit: via configuration Each m odification implicitly acquires cluster w lock if not ide already held. Already available in 4.0.0.Alpha5!
- 12. New features - client/server module Server module = cache wrapper over TCP Client module = cache proxy Highly pluggable! Transport: XNIO, Netty, etc. Protocols: m cached, custom etc. em , Failover and load balancing Usable w current m cached clients ith em Drop-in replacem m cached servers ent em
- 13. New features - fine-grained model Successor to POJO Cache JPA-like interface: persist, find, remove... W not rely on AOP, javassist...etc ill M robust and easier to use/debug ore
- 14. New features - Others Query module Execute Lucene queries against cache Based on JBoss Cache - Searchable Distributed executors Runnable/Callable executed on data set Moves code, not data, around cluster