Inline. J-D
2011/5/31 Gaojinchao <[email protected]>: > Per I know: > 1.zookeeper is sensitive to resources(Memory, Disk, CPU, NetWork). Mostly just disk. > If there is some underprovisioning on server, then > a) Server may not respond to client requests in time. > b) Client assumes server is down, closes the socket and it connects to other > server. > > 2. Hbase is sensitive to RAM. But CPU use less. Unless you swap, in which case your performance is bye bye. > > 3. M/R(it againsts HDFS ) is sensitive to CPU and Memory. Usually MR is IO-bound. > > My question is that: > 1. Does it have a race resource between HDFS, ZK, M/R and Hbase? MR can steal IO from HBase and ZK, in the latter case it's really bad. > > 2. Suppose that one machine fault, the cluster should be safe. > What do I need consider(eg: cpu, network use rate or configure)? More machines, I'd never run HBase in production on just three nodes. > > 3. Is there anyone who has used below configure for product? On such a small cluster there's no good configuration IMO, you'd need to get a few more machines then you would put ZK, Namenode, JobTracker and HMaster on one node and all the others would have DataNode, TaskTracker and HRegionServer. > > > > -----邮件原件----- > 发件人: [email protected] [mailto:[email protected]] 代表 Stack > 发送时间: 2011年6月1日 4:13 > 收件人: [email protected] > 主题: Re: about disposing Hbase process > > Sorry Gao, what is your question? > St.Ack > > 2011/5/31 Gaojinchao <[email protected]>: >> For one our application, There is 3 node. >> All process disposing and machine configure is as below. >> >> Who has experience about this? >> >> The use rate of cpu is about 70%~80%, Does it make HBase or zookeeper starve? >> >> >> >> Machine: >> >> cpu: 8 core 2.GHz >> >> memory: 48G >> >> Disk: 2T*8 = 16T >> >> Node1： >> DataNode >> HAJobTracker >> TaskTracker >> QuorumPeerMain >> HMaster >> HRegionServer >> >> Node 2: >> NameNode >> DataNode >> HAJobTracker >> TaskTracker >> HMaster >> HRegionServer >> QuorumPeerMain >> >> Node 3: >> QuorumPeerMain >> HRegionServer >> TaskTracker >> DataNode >> >
