1847196
Members
4629
Online
110263
Solutions
Don't worry about 100% utilisation on a SAN disk - it's pretty meaningless - all it tells you is that during the interval, that disk was doing IO for 100% of the time (not necessarily a problem for an EVA LUN which could be backed by between 8 and over 200 physical spindles...)
I'd look first at what you see on disk2/3 - what do you have on those disks?
HTH
Duncan
I am an HPE Employee
I'd look first at what you see on disk2/3 - what do you have on those disks?
HTH
Duncan
I am an HPE Employee
Ron,
>> My question is what is the magic number and how far can you go?
Well most EVA ports have a queue depth of 1536 (what model do you have?)
But that doesn't mean you can set the qdepth to 1536! You have to take into account _all_ the LUNs and _all_ the hosts that access the EVA, and make sure there's no way you could possibly have more than 1536 outstanding IOs on all LUNs to all hosts on a given port.
So if this host sees 4 different LUNs through one port on the EVA, and no other systems use the EVA, then you would be able to set the queue depth for that LUN on that port to 1536 / 4. Of course if you had a Windows system accessing 2 LUNs on that port as well you'd need to take that into account as well - I seem to recall that Windows default queue depth is 32, so in this case your queue depths would be (1536 - (2 * 32)) / 4. Hopefully that makes sense.
So the first thing to ask is "is the EVA accessed by other systems as well as this one?" If the answer is _no_ then you could probably setup Target port based queue depth management as described in this whitepaper (which you could do worse than read anyway):
http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c02054623/c02054623.pdf
HTH
Duncan
I am an HPE Employee
>> My question is what is the magic number and how far can you go?
Well most EVA ports have a queue depth of 1536 (what model do you have?)
But that doesn't mean you can set the qdepth to 1536! You have to take into account _all_ the LUNs and _all_ the hosts that access the EVA, and make sure there's no way you could possibly have more than 1536 outstanding IOs on all LUNs to all hosts on a given port.
So if this host sees 4 different LUNs through one port on the EVA, and no other systems use the EVA, then you would be able to set the queue depth for that LUN on that port to 1536 / 4. Of course if you had a Windows system accessing 2 LUNs on that port as well you'd need to take that into account as well - I seem to recall that Windows default queue depth is 32, so in this case your queue depths would be (1536 - (2 * 32)) / 4. Hopefully that makes sense.
So the first thing to ask is "is the EVA accessed by other systems as well as this one?" If the answer is _no_ then you could probably setup Target port based queue depth management as described in this whitepaper (which you could do worse than read anyway):
http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c02054623/c02054623.pdf
HTH
Duncan
I am an HPE Employee