Comments (4)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 zhuangzy commented Permalink

only a few LUNs may hurt I/O performance, every hdisk has a queue. For production environment, at least 4-8 LUNs sounds good.

2 nagger commented Permalink

Hi zhuangzy. Absolutely but I would go a little higher like 10 to 16 as a minimum. Don't forget the LUNs should be spread across a number of disks at the backend and probably have a good size diks sub-system cache too. So 10 LUNs does not mean 10 disks at the back end but a large RAID5 stripe. As always, set the disk queue depth nice and high and we should spread the LUNs across lots of spindles too with as little overlap as possible. Of course, users with very large SSP pool size will be thinking in much larger numbers and proobable spreading the I/O across lots of back end disk sub-system devices.

3 jmueller commented Permalink

Nigel, <div>&nbsp;</div> Does it make sense to add one more(second) LUN? <br /> iostat -D hdisk2 1 | awk '/queue/{print; getline;print $0}' <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 3.6 0.0 33.7 16.0 22.0 1117.0 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 2.5 0.0 33.7 11.0 18.0 956.6 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 2.1 0.0 52.0 13.0 27.0 1500.9 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 2.4 0.0 52.0 11.0 24.0 992.4 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 2.4 0.0 52.0 10.0 18.0 795.3 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 1.2 0.0 52.0 6.0 14.0 793.3 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 1.5 0.0 52.0 9.0 18.0 1329.2 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 1.7 0.0 52.0 9.0 16.0 1127.3 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 1.8 0.0 52.0 8.0 17.0 1068.7 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 3.4 0.0 52.0 15.0 20.0 961.2 <br /> queue: avgtime mintime maxtime avgwqsz avgsqsz sqfull <br /> 3.0 0.0 52.0 14.0 21.0 1147.4 <div>&nbsp;</div> /dev/db2tkpdatalv 1440.00 165.21 89% 2914 1% /db2/TKP/tkpdata <br /> lsattr -El hdisk2 -a queue_depth <br /> queue_depth 40 Queue DEPTH True <div>&nbsp;</div> Best regards <br /> Joachim

4 nagger commented Permalink

Answer to jmueller. I STRONGLY recommend you are 9 more LUNs. If you have enough spindles at the disk sub-system back end also spread them across lots of spindles. Your one LUN, one test program and one SSP client VM is an extreme end special case.