[ https://issues.apache.org/jira/browse/HDFS-14403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16813640#comment-16813640 ]
Daryn Sharp commented on HDFS-14403: ------------------------------------ I've got a simpler and more robust patch I'll try to cleanup/put up today for consideration. It lacks specific tests but has more accurate and fine grained tracking of time w/o a lot of new data structures. The existing decay scheduler can be toggled to use a cost instead of call count. It tracks timings for: * reader enqueue * in queue * processing ** handler overhead ** invoke *** lock free *** lock wait *** lock shared *** lock exclusive ** response time Each timing can be individually weighted with a config. I was waiting for results from an initial deploy before posting publicly. I was going to file a jira and then I saw this patch... > Cost-Based RPC FairCallQueue > ---------------------------- > > Key: HDFS-14403 > URL: https://issues.apache.org/jira/browse/HDFS-14403 > Project: Hadoop HDFS > Issue Type: Improvement > Components: ipc, namenode > Reporter: Erik Krogen > Assignee: Christopher Gregorian > Priority: Major > Labels: qos, rpc > Attachments: CostBasedFairCallQueueDesign_v0.pdf, > HDFS-14403.001.patch, HDFS-14403.002.patch > > > HADOOP-15016 initially described extensions to the Hadoop FairCallQueue > encompassing both cost-based analysis of incoming RPCs, as well as support > for reservations of RPC capacity for system/platform users. This JIRA intends > to track the former, as HADOOP-15016 was repurposed to more specifically > focus on the reservation portion of the work. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org