Overview:
Load simulator virtual appliance (Linux) that will integrate with
vCenter to rapidly deploy Linux VMs that generate CPU, memory, network,
storage, and Fault Tolerance workloads.
Problem:
When deploying vSphere on new hardware (server, network, storage)
there is currently no simple way to stress test it before certifying it
for production use. Want to answer questions like:
• Is virtual and physical network configured correctly and not
causing perf issues? Are packets being dropped? Is the 1GB network
connection allowing up to 1GB of bandwidth?
• Is the server's memory reliable?
• How will FT affect my network design?
• What happens to my hardware when fully maxed out?
• How does storage handle IO workloads being generated from different hosts?
Solution:
Single simple tool to stress test new vSphere hardware / vSphere
installations. Could also be used to test vSphere share mechanisms.
Not intended to simulate actual workloads.
Detail:
Load Simulator virtual appliance with a web base UI connects to
vCenter and allows admins to rapidly deploy stress test VMs with various
CPU, memory, network, storage, and FT sizes. The stress test VMs
would generate CPU, memory, network, and storage based upon choices in
the UI. For example:
- CPU - Choose # of vCPUs and how much CPU that should be generated
on the VM. For example, 8 vCPUs running at 50% utilization.
- Memory - Choose how much memory for VM and how much of that
memory should be consumed by the VM. The app would then consume the
memory in the VM
- Network - Choose Portgroup for VMs to use and then choose how
much network the VM should consume. The VM's network traffic generator
would be smart enough to automatically pair with other deployed stress
test VM as a destination for the network traffic. A checkbox should be
available in the UI to choose if network traffic should traverse hosts
to another host where a partner stress test VM is running (vs intra-host
network traffic)
- Storage - Choose which VMFS volume the stress test VM should be
deployed to and how much IO the VM should generate. Could have
dropdown with simple workload type to generate.
- SMP FT - UI would allow choose VMs to have Fault Tolerance
enabled. This will be important as the next version of vSphere has been
stated to include SMP FT and customers will want to understand
implications on network HW.
- Cleanup button - A cleanup button would be available to remove the stress test VMs
Load Simulator UI would gather all vSphere Events in consolidated
view. Furthermore, it would gather key perf metrics from the vSPhere
Perf UI and roll them up into a perf chart. The Events and Perf view
would make it easy to identify issues with HW and perf during the test.
Example Use:
Building out new blade based host running latest version of
vSphere. Once environment is configured as per design spec, navigate
to the Load Simulator UI and configure the test to deploy 50 stress test
VMs in Small, Medium, and Large configurations. Monitor vSphere
events and performance using Load Simulator UI to see if any
bottlenecks are occurring. If issues are not occurring use the Load
Simulator UI to add 10 more VMs and then recheck perf. Let stress test
run for an hour and recheck perf. If all looks well, click on the
cleanup button and the stress test VMs are automatically deleted.
If you like this idea vote here:
https://flingcontest.vmware.com/ideas/load-simulator-virtual-appliance-linux-that-will-integrate-with-vcenter-to-rapidly-deploy-linux-vms-that-generate-cpu-memory-network-storage-and-fault-tolerance-workloads
Labels: contest, fling, load simulator, memmtest, share, stress, Vmware, vSphere