Hey all, Jonathan, I think this is all good stuff
On my end, I'd like to give a presentation on the integration of Zookeeper in HBase that Nitay Joffe and me have been working on. Main subjects would be: - The main features that are implemented for 0.20 (.ROOT. location stored in ZK, HMaster failover, Region Servers leases replaced with ephemeral nodes, etc) - How HBase manages ZK instances. - The issues we encountered during the integration and how we solved them (mostly stuff about configuration) - How using ZK improved HBase. One example is that by removing the RS leases, a dead node is detected in the following 2-3 seconds instead of, in the worst case, the full length of a lease. This way, we can reassign the regions faster. - The main features that are implemented for 0.21 like using ZK to distribute the HMaster role to Region Servers. - If in June HBase 0.21 is released or soon to be, I would cover the new issues/improvements. J-D On Thu, Feb 5, 2009 at 3:59 PM, Jonathan Gray <[email protected]> wrote: > Following some discussion on IRC, I'd like to propose giving a presentation > for the Hadoop Summit. JD is also interested. > > My current idea would be to title it something like "HBase Goes Realtime" > or > something about HBase being Web-ready. > > Would include discussion of: > > - Improvements in the (released at this point) 0.20 and potentially > 0.21/0.22 releases by June. > - Specifically those related to performance and the use of HBase for > realtime serving of a web site. (New file format, cell caching, block > caching, indexes, rpc, whatever else we tackle by the summit). > - Use of HBase in production at Streamy. We're using it in a number of > different ways and should be able to cover most of the issues related to > using HBase for the web. Specifically, we'd touch on the migration from > relational databases, decisions we made along the way, trade-offs in > architecture and schema design, etc. but all in all a "success story" on > using HBase to serve a realtime web site. > - Touch on improved scalability. > > I will let JD speak for himself, but I believe he was thinking of talking > about Zookeeper and HBase. We'd then be able to cover the stability and > fault-tolerance improvements made in HBase in this talk. > > I think that would cover our bases. > > Let me know what you all think. > > JG > > > -----Original Message----- > > From: Ajay Anand [mailto:[email protected]] > > Sent: Thursday, February 05, 2009 11:25 AM > > To: [email protected]; [email protected]; zookeeper- > > [email protected]; [email protected]; pig- > > [email protected] > > Subject: Hadoop Summit 2009 - Call for submissions > > > > We are planning the 2009 Hadoop Summit, to be held the second week of > > June in Santa Clara, CA. > > > > > > > > Please send me ([email protected]) your presentation proposals and > > suggested topics. > > > > > > > > Areas we plan to cover include: > > > > - Core Hadoop - areas of development and key contributions > > > > - Subprojects and related projects: including Pig, Hive, > > Hbase, > > Mahout, Zookeeper and others > > > > - Administration: Monitoring tools and framework, services, > > experiences with administering Hadoop clusters > > > > - Adoption: Deployment examples, hosted services > > > > - Applications on Hadoop: examples of innovative applications > > being developed and deployed on Hadoop > > > > - Research: Research topics being explored, university > > research > > examples > > > > > > > > Thanks, and looking forward to a strong event, > > > > Ajay > > >
