All I can tell you is the number of VM's PER LUN isn't a solution, it's a guess. It depends on the size of the VM, the size of the LUN.
If you have Fiber, it's a very robust system. You can have millions of files on a single LUN. ESX may manage the datastore on those LUNs, but even 20 or 30 VM's should be a significant factor. So these numbers are based upon a very conservative estimate. I myself have tested this with 40 VM's accessed by 8 ESX host, on a single LUN and experience no performance impact, vs only 5 VM's on a different LUN. I have tried many configurations, but it always comes down to your storage.
That is the single biggest, most important element. If you don't have good CENTRAL storage, nothing else matters. Use proper configurations with lots of spindles, and high IO disks, such as SAS. SATA will have a HUGE impact on performance, so don't think you can get the same performance from SATA disks that you can from SAS, because you cannot.
: this I number I get from my storage vendor?
SAN
You can use this as a starting point, but it's not a definitive guide. I can dispute many of these numbers, because we have been running more than this for over 3 years..with no issues. The issues come from ESX (and MOSTLY due to 32-bit vs 64-bit which has since been fixed). Early versions of ESX were a lot less resilient. ESX 4.0 has been much better, so take that into account as well.
this I get from client OS level metrics like perfmon?
Host. esxtop in ESX
3) Is it no longer recommended to alter queue depth settings? rparker states this here:
Yes, and I stand by that remark. You should NEVER alter a setting for the sake of altering. You may not even need it. So queue depth is ONLY IF you have issues, then you can adjust it to suit your needs. The new default is now 64, and it should be watched, and very few problems can be attributed to queue depth. That's PER HOST setting, so it still depends on your SAN. Depends on the other hosts attached, and increasing or decreasing this could have a negative impact on other systems, so that's why I say don't touch it.
Max VM's per LUN is a recommendation by VMware. They have to be conservative otherwise they can't support VM's. So these numbers must be considered as general guideline. If however you go over these numbers they recommend and you have problems, they may not be able to help.. but it still depends on MANY factors, one set of numbers cannot be applied to EVERY environment.
