Below is an overview of a curent situation...
An ESX farm consisting of 10 physical servers, each containing many guests.
All Physical servers are connected to a DMX4 through FC switches (ED48000).
there are 20x 500Gb LUNs provisioned on the DMX allocated to all the physical servers.
Using Performance manager I cannot "see" the overall utilization of each LUN as a whole, only individual devices.
With multiple hosts and guests accessing the LUNs how can I "see" individual performance metrics?
for example comparing this to a Clariion where you can get statistics of each RAID Group/LUN within a RAID group to determine utilization/performance.
The customer has a VM that has issues, however I cannot "see" the specific device as a LUN on which this resides to determine utilization/performance, to eliminate any issues at the DMX.
What if any metrics can I view from the DMX that will give me that level of detail?
The customer does have an ECC server, but no agent on the ESX hosts or guests?
Would installing ECC Agent on the ESX servers give any further detail to help the situation?
Is there another tool set that could be used either from ESX virtual center or from an array perspective?
Any help appreciated.
There are three basic components that can make up a host LUN on Symm. First are the physical disks. You can carve out logical volumes from the physical disks in any RAID type you choose, and you can mix them all on the same drive, however this isn't the best practice.
Once you have created a logical volume on some number of drives, you can then present that volume to the host, or group several together into a meta volume.
You can get disk performance stats in ECC/STP, as well as individual volume stats, or you can use SPA.
I think you missed my point, I am not after "how to present a LUN to a Host" this is already done.
The devices are created from multiple disk groups and then meta'd together to form the 500Gb LUN.
If, for instance the meta consists of 10x 50Gb devices, and each of the Tdevs are from different disk groups of 8 physical spindles.
the resulting LUN is in effect spread over 80 disks.
I can "see" each of the physical spindles but that doesn't narrow down specifically to the one LUN I am interested in (only a minute part).
I can also "see" each of the Tdevs, but again that is only 1 tenth of the resulting LUN and with concatenation it may not represent what is happening to the LUN from a performance utilization perspective.
I want to view the LUN as a whole if possible - do I take the IO per device and average it? total it? or what to get what is happening on that LUN!
therefore if the LUN is 30% utilized I could say there is no issue with the LUN, however if it is ~=99% then there may be a problem.
What tool or series of metrics can I use to determine an individual LUN as a whole as presented to the Host?
What is SPA?
So you have VP meta volumes. You can get performance stats for the TDEV presented to the host, or for the TDAT pool as a whole, but it is not really possible to figure out what specific IOs on the TDAT are coming from which TDEV.
Since the TDEV is really only a cache device, it really doesn't have "utilization", but the FAs it is mapped to does, as well as the TDATs it is mapped to in the pool. So what you should be looking at a high level are the disks, and the front end CPUs for TDEV performance.
You can't have a TDEV meta spanning multiple pools, the TDEVs must not be bound in order to form the meta and once formed you bind the meta head to a pool.
Unless he's referring to a host based concatenation. Is there a terminology confusion happening here?
That is correct, a TDev cannot span multiple pools - not what I am asking!
I know how to bind and form meta's - again not what I am asking!
For a specific LUN (meta devices allocated to a host) how can I "see" what its utilization is?
I know I can get the utilization of each physical spindle - in my senario that would be 80 disks.
and I can get individual Device stats (Tdev), I am looking for a LUN as a whole type information.
In Clariion you run Analyser and then from the output file you can get stat's on each LUN per RAID Group, the RAID Group itself or each disk.
The various server Admin's want to get performance metrics on specific LUNs presented to a host
I am trying to assertain if there is any tool or cmd that will give me stats on a LUN,