HBase
  1. HBase
  2. HBASE-2125

tcpnodelay is false by default; it should be true?

    Details

    • Type: Bug Bug
    • Status: Resolved
    • Priority: Major Major
    • Resolution: Duplicate
    • Affects Version/s: None
    • Fix Version/s: None
    • Component/s: None
    • Labels:
      None

      Description

      Up on IRC, nagles algo came up. Thought was that it should be off by default in our RPC only we'd need to check that nodelay didn't do damage:

      22:16 < javarants1> appears to be enabled unless you set it in the config
      22:16 < dj_ryan> good good
      22:16 < javarants1> probably want to set hbase.ipc.client.tcpnodelay and ipc.server.tcpnodelay to true and test that
      ...
      22:27 < St^Ack> So, lads, tcpnodelay should be TRUE on client and server by default rather than as they currently are, false?  If so, I should get this change into 0.20.3 (I can't repro kannan's hang still)
      22:28 < Toad> nodelay true makes sense but you should make sure that you've got appropriate buffering then
      22:28 < Toad> otherwise you'll get a bajillion packets per second and you'll suck throughputwise
      22:29 < dj_ryan> we buffer entire rows and whatnot
      22:29 < dj_ryan> so shouldnt be a problem i think
      22:29 < Toad> yea, worth watching a wireshark though, make sure your packet trace looks the same
      22:29 < St^Ack> thanks Toad ... then that means punt till 0.21 (smile)
      22:29 < St^Ack> not in imminent 0.20.3
      22:31 < Toad> yea, I wouldn't switch unless you have time to look it over
      

        Issue Links

          Activity

          Hide
          Todd Lipcon added a comment -

          Apparently I'm forgetful in my old age

          Show
          Todd Lipcon added a comment - Apparently I'm forgetful in my old age
          Hide
          stack added a comment -

          HBASE-5943

          FYI, the Toad in the IRC conversation here has a new handle now. Its tlipcon now. The advice he gives here is the same as he gives a couple of years later in HBASE-5943.

          Show
          stack added a comment - HBASE-5943 FYI, the Toad in the IRC conversation here has a new handle now. Its tlipcon now. The advice he gives here is the same as he gives a couple of years later in HBASE-5943 .

            People

            • Assignee:
              Unassigned
              Reporter:
              stack
            • Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:

                Development