Hi. I started a Navi Analyzer session an noticed my dedicated a LUN 3+1 R5 300 GB FC 15k LUN was avg over 1000 IOPS. This is on a CX3-80. How is this possible when the max disk IOPS is (150 x 4) = 600 disk IOPS Total?
I quickly looked over the other counters and they looked fine, esp Utilization which was under 40%. I would assume these are tied together.
The numbers your are talking about (150 IOPS) is not a maximum, it is a recommended sustained maximum. This means you should plan to never exceed this number for any extended period of time.
The drives themselves can burst to much higher rates, but it can have a serious impact on your performance if you rely on it.
I've seen 146GB 15k drive sustain 300+ IOPS for minutes at a time, but once we found it we redesigned the disk layout to prevent it happening again.
I'll have to defer to one of the performance experts from EMC on this, but my understanding is that utilization is one of those things that has a complex algorithm calculating it in the background and it cannot be relied on entirely. Usually it is a good indicator, but it isn't an absolute number.
You overloaded a LUN(3+1), it does not mean the SP is busy, just the backend of the CLARiiON Here are some parameters recommended by EMC CLARiiON Best Practices: FC disks (I/Os) Threshold value = 160 ATA disks (I/Os) Threshold value = 70 seeking FC disks Threshold value = 10% of diskcap. or > 30GB/s seeking ATA disks Threshold value = 10% of diskcap. or > 30GB/s FC disks response time Threshold value = 15ms (total iops > 20) ATA disks response time Threshold value = 15ms (total iops > 20)
Smaller I/O's can be services more per second than very large ones, so you might see 1000 iops with I/O's of 1kB and 30 iops with I/O of 1Mb. Bandwith and throughput are ... how do you you say this in english ? .... each other's opposites.... If one is high, the other is low.
Allen Ward
6 Operator
•
2124 Posts
1071
1
Posted August 19th, 2008 11:00
The drives themselves can burst to much higher rates, but it can have a serious impact on your performance if you rely on it.
I've seen 146GB 15k drive sustain 300+ IOPS for minutes at a time, but once we found it we redesigned the disk layout to prevent it happening again.