Comments (11)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 CarlNZ commented Permalink

Hi Anthony - is there a way to change the timeline within performance monitoring from the default 5 minute period ? <div>&nbsp;</div> cheers - Carl

2 anthonyv commented Permalink

Hi Carl. <div>&nbsp;</div> Great question. <br /> Right now we cannot change this. <br /> There are plans to dramatically widen what this panel can do in future releases, so watch this space.

3 ejc commented Permalink

Hi Anthony, <div>&nbsp;</div> In the "Interfaces" panel, we observed a peak in the FC IOPS of over 57K while the IOPS in the "Volumes" panel was around 10K. This is a 4-node V7000 cluster (two control enclosures)... so I suspect that the increased FC IOPS were a result of intra-cluster traffic over the FC paths. Appreciate your comments. Thanks!

4 anthonyv commented Permalink

Its a great question and quite a few things go into the answer. <br /> If a host writes to node 1 then the following things happen: <div>&nbsp;</div> 1) Node 1 gets incoming IO from host <br /> 2) Node 1 mirrors the write to node 2 <br /> 3) Later Node 1 writes the data down to the disk subsystem <div>&nbsp;</div> So you may well see peaks on the FC interfaces that are higher than the peaks you see on the volume stats.

5 ejc commented Permalink

Thanks, Anthony. To clarify, there is currently no externally virtualized disk. (There will be soon). <br /> We did create storage pools from disks attached to each controller. Might that account for the additional intra-cluster FC traffic? <div>&nbsp;</div> If so, do we need to be concerned with saturating the FC ports and at what level of IOPS?

6 anthonyv commented Permalink

Mmm... is there a pool spread across mdisks (arrays) on each of the two controllers? <br /> If not then I am confused. <div>&nbsp;</div> The scenario I drew was really for an SVC. <br /> Normally for a V7000, the internode traffic does not go via FC (it goes via PCIe). <br /> The SAS disk traffic clearly also does not go via FC. <div>&nbsp;</div> The two controllers in a clustered pair do not sync cache and the handshake traffic is tiny. <br /> The only way you can have this is with a volume spread across two controllers... <br />

7 ejc commented Permalink

Correct. We have pools containing Mdisks built from arrays on each control enclosure. Was wondering if/when we have to be concerned about the amount of FC IOPS for this "Non-Host" IO? Thanks... <div>&nbsp;</div>

8 anthonyv commented Permalink

Its not a concern. <br /> As long as port utilization does not get to saturation you are ok. <br /> For the vast bulk of clients they never get near it.

9 Suman Debnath commented Permalink

The IOPS that we see on the VOlume in the SVC or V7K this the IOPS the host generates ? <br /> Lets say, if I have a windows host.....and running IOMeter....lets say at any point of time IOMeter is showing 1000 IOPS, should we expect 1000 IOPS in the SVC/V7K GUI Volume performance stats?? <div>&nbsp;</div> Thanks <div>&nbsp;</div> Regards, <br /> Suman

10 anthonyv commented Permalink

Volume IOPS are indeed all IOPS generated by all the hosts attached to the box. <br /> Remember with IOmeter that it has a refresh rate (as does the V7000 GUI). <br /> They might be slightly out of sync.

11 Suman Debnath commented Permalink

Thanks Anthony....