Yes, as a high-performance in-memory data grid, ScaleOut StateServer runs best in a fast, responsive environment. The following guidelines will help to ensure the best performance:
Provision VMs without overloading the available CPU or network bandwidth on the physical server.
Ensure that multicast is enabled in the virtual environment if automatic discovery is configured (the "Use Multicast" checkbox in the SOSS Console's Host Configuration tab)—some virtual environments block multicast, which SOSS uses for dynamic host discovery.
While SOSS can run on multiple VMs on a single physical machine, this configuration is not highly available, because a single hardware failure could cause the loss of multiple SOSS servers, thus potentially causing a loss of data. When high availability is a concern, we recommend running only one SOSS server per physical machine.
Avoid dynamically moving VMs (using VMware’s vMotion feature, for example) that are running the SOSS service, as the pause in point-to-point network traffic may be interpreted by the SOSS service as a failure scenario.
Avoid taking and/or consolidating snapshots of VMs, as snapshot activity may quiesce the guest system for an extended period, which may be detected by other instances of the SOSS service as an outage. Note that VMware backup solutions (including third party products like Symantec NetBackup or Dell's vRanger) may perform snapshotting behind the scenes.
VM-level backups of SOSS hosts typically aren't needed and aren't valuable--SOSS doesn't persist anything to disk, and if the VM's memory is included in the snapshot then it will immediately become out-of-date. Ordinary guest OS-level backups are sufficient for backing up any configuration settings.
In general, ScaleOut servers should leave the distributed store before any planned maintenance which could lead to a pause or drop in network traffic or guest responsiveness.
If possible, use VMware’s latest virtual network adapter (VMXNET 3) for the best possible networking performance in your VMware guests. See VMware KB article 1001805 for details on choosing a network adapter for your virtual machines.
The net_perf_factor parameter allows servers to tune their sensitivity to network delays. For example, reducing net_perf_factor may allow VMotion to run without interfering with SOSS. If you decide to tune net_perf_factor, we would recommend starting at 75 or 85, and moving downward if there are still issues. Please do not tune net_perf_factor lower than 50, as this would increase recovery time by more than a factor of 2 and delay recovery from legitimate outages. net_perf_factor is described in more detail in the SOSS Help file’s “Configuration Parameters” topic.