Unsolved
This post is more than 5 years old
59 Posts
0
1338
Meta LUN shows high utilization
Hi,
we have a MetaLUN on a CX480 which consists of 24 members on RAID 5 (4+1) RGs. If I look at the MetaLUNs utilization there is an average of 85% with spikes up to 100% whereas the members show an average of 6 - 8% each.
How do I have to interpret this? Could this be a potential bottleneck?
btw, The related RGs show an utilization of 30% max.
Regards
Dieter
jps00
392 Posts
0
June 21st, 2011 06:00
Are there other Meta's involved in the RAID groups? Did you stagger the base components across the RAID groups?
Are any of the components 'stacked'? That is more than one component of your Meta shares the same RAID group.
Mendocino
59 Posts
0
June 21st, 2011 07:00
yes, all provisioned LUNs are Metas and build from all RAID groups.
The members of a meta are from different RGs.
The chart shows the utilization of the Meta head (upper line) and the utilization of the members.
Kumar_A
727 Posts
1
June 21st, 2011 13:00
In Analyzer, Utilization is calculated as (number of busy intervals) / (number of busy intervals + number of idle intervals). An object is counted as busy if there are 1 or more requests in queue or in progress for that object.
The metaLUN counts as busy if there is at least 1 active request, which will probably only keep 1 of the component LUNs busy at a time. So, it is not unusual to find high metaLUN utilization rates, but low utilization rates for the component LUNs.
100% utilization does not necessarily mean there is a bottleneck, just that there is a steady stream of work to the object. Low utilization implies the object is not working very hard, and there is lots of idle time. High utilization implies the resource is usually busy, doing its job and giving value. But high utilization is often a good indicator of the busiest objects, to help focus on potential problems, like:
1. If the object is being kept busy with a large queue all the time, it will result in high response time.
2. If the LUN is counted as usually busy, but is running only a few IOPS, it implies each IO is taking a long time for some reason.
kelleg
4.5K Posts
1
June 22nd, 2011 09:00
A better value to look at is Queue Length - those objects (LUNs or disks) that have high Queue Length are probably busier than other objects.
glen
Mendocino
59 Posts
0
June 27th, 2011 00:00
Thank you very much for your answers!
As the Queue Length doesn't show high numbers too (1.471 in average) I think the LUN hasn't a performance problem.
Dieter