IO Performance Fine Tuning
Viewed 78 times
An administrator may encounter situations when a VPG, or set of VPGs, is frequently entering Bitmap Sync, or even undergoing continuous Bitmap Syncs. In these situations, either the IO throughput is repeatedly higher than what associated VRAs can transfer over the WAN connection, or the target datastore is being written at such a high rate that a delay exists for writing replicated data to disk. This article outlines steps that can be taken to fine tune performance, and minimize this eventuality.
In order to fine tune IO performance, to avoid repeated Bitmap Syncs, the following logical steps are recommended:
- Ensure that the WAN Compression option is enabled in the configuration of the affected VPG. With WAN Compression enabled, Zerto Virtual Replication will use less bandwidth over the WAN connection.
- Ensure that source site VRAs servicing the affected VPGs have 3GB of reserved RAM. To do so, see the KB titled "How to Increase the RAM Allocation for an Existing VRA". Doing so will allow the VRA's buffers to store a greater amount of IOs, prior to sending them over the WAN connection.
- Ensure that VMs configured within affected VPGs are spread across as many target hosts as possible. This will allow Zerto Virtual Replication to leverage the buffers on a greater number of VRAs, decreasing the likelihood of a Bitmap Sync due to higher IO throughput than a lesser number of VRAs could handle.
- Ensure that Zerto Virtual Replication has access to bandwidth sufficient to handle the IO load of the protected applications. To do so, it is recommended to follow the guidelines within the WAN Sizing document, available for download from the Zerto Downloads page in the Zerto Support Portal. In addition, in order to test a link for effective available bandwidth, it is recommended to use an application such as iPerf for diagnostics between sites. Instructions on using iPerf are available in the "Using iPerf Windows Command Line Tool to Measure Bandwidth between Sites" article. If the resulting values are less than expected, it may indicate a network-related performance problem.