[ https://issues.apache.org/jira/browse/HADOOP-5632?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12717086#action_12717086 ]
Khaled Elmeleegy commented on HADOOP-5632: ------------------------------------------ To test the scalability of the patch, i.e. whether the patched JT will be able to keep up with the load in a large cluster or not, I've used ~200 nodes cluster. Each node has two quad core CPUs, 16GB of memory, and 4 disks. On each node, I ran 10 TaskTrackers, simulating ~2000 node cluster. Each TT has 6 map slots and 2 reduce slots. I patched the hadoop trunk with the 5632 patch. I ran the sleep job from the examples jar with 500,000 maps and map runtime (sleep time) of 20 seconds. I measured the slot utilization and it was 99.8%. I followed the CPU utilization at the jobtracker and all the 8 cpus were like 20% busy on average. Jobtracker's CPU utilization varies along time, but the CPUs are no where near saturation. Also, I didn't observe lock contention, i.e. Load was, more or less, evenly balanced among all the CPUs. I cranked up the load to stress test the JT by reducing the map runtime (sleep time) to 1 second. Still, the JT was able to keep up with the load with no problem. > Jobtracker leaves tasktrackers underutilized > -------------------------------------------- > > Key: HADOOP-5632 > URL: https://issues.apache.org/jira/browse/HADOOP-5632 > Project: Hadoop Core > Issue Type: Improvement > Components: mapred > Affects Versions: 0.18.0, 0.18.1, 0.18.2, 0.18.3, 0.19.0, 0.19.1, 0.20.0 > Environment: 2x HT 2.8GHz Intel Xeon, 3GB RAM, 4x 250GB HD linux > boxes, 100 node cluster > Reporter: Khaled Elmeleegy > Attachments: hadoop-khaled-tasktracker.10s.uncompress.timeline.pdf, > hadoop-khaled-tasktracker.150ms.uncompress.timeline.pdf, jobtracker.patch, > jobtracker20.patch > > > For some workloads, the jobtracker doesn't keep all the slots utilized even > under heavy load. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.