Hi, you mention that your 4-tile run is failing QoS. Did you get a passing (compliant) VMmark run with 1, 2 or 3 tiles?
The failure is due to resource constraints. The most common bottlenecks I’ve seen are storage and sometimes it could be the client systems’ CPU.
Make sure your Weathervane Auction VMs, especially AuctionDB and AuctionNoSQL, are placed on the fastest storage, and they are not sharing a datastore with other workloads.
Check the CPU utilization of your client hosts and system under test hosts to make sure they are not maxed out.
In the VMmark user’s guide, check the VMmark Hardware Requirements and make sure your environment meets all the requirements.
i have compliant run with 3 tiles on the same SUT & same settings.
but it seemed that i can't or extremely difficult to pass 4 tiles run(and more) due to weathervane auction qos failed.
in the client vm system,
i placed one client vm per server node;
each node has two intel 2680 v3 CPU installed, 192GB RAM and a SSD to store the client vm.
basically no other vm or services exist in my client servers and the CPU utilization of each nodes never max out.
in my SUT cluster(which is a 4 node server):
a hybrid vSAN storage(a disk group) is utilized as main datastore to place nearly all the workloads including the weathervane auction VMs.
(ssd is one intel s4600 960GB per node and with 5 SAS disks)
two intel cpu gold 5120 are installed on each node; the RAM and disk space are enough to execute many tiles.
the CPU utilization of each nodes reached 50% or less when i did the 4 tiles test.
i placed VM network on a vmnic adapter of intel 82599 and configured other traffics(vSAN, vMotion...etc) on another vmnic adapter of this 82599 NIC.
i hope the bottleneck of weathervane auction QoS is not on network?
A. is it compliant to place weathervane Auction VMs on other shared storage out of SUT to test if needed?
(because it would be hard for me to create another shared datastore under my SUT)
B. will the weathervane auction QoS problem be solved simply by increasing client VMs' vCPU or change to better processors?
thank you so much!
Thank you for such a complete answer. I’m glad to hear that you were able to get 3 tiles.
Each tile is a unit of load. The simple answer is that your system under test doesn’t support 4 tiles of load at a passing quality of service. From what you’ve described, you have more than enough CPU and processing power to go around. Most likely, you have a storage bottleneck from the VSAN HDDs. There are a number of published VMmark results that are VSAN. VMmark 3.x Results Look for Quanta Cloud, Sugon, and Supermicro. Check to see whether they are all-flash or hybrid.
I doubt network or client CPU are a problem for you.
A. Yes, it is compliant to place the VMmark VMs on any shared storage, as long as the VMs are still part of the VMmark cluster and the VMs can be storage vMotioned to the SvMotion target LUN and vMotioned to any host in the cluster.
B. I don’t think so. You could try a faster processor for the Client hosts possibly. You can still get a passing VMmark run if you increase the clients’ # of vCPU, but the result will not be compliant, in that it cannot be published on VMmark.com.