Write caching for asynchronous writes

Writing to the cluster asynchronously with write caching is the fastest method of writing data to your cluster.

Write caching for asynchronous writes requires fewer cluster resources than write caching for synchronous writes, and will improve overall cluster performance for most workflows. However, there is some risk of data loss with asynchronous writes.

The following table describes the risk of data loss for each protocol when write caching for asynchronous writes is enabled:

Protocol
Risk
NFS
If a node fails, no data will be lost except in the unlikely event that a client of that node also crashes before it can reconnect to the cluster. In that situation, asynchronous writes that have not been committed to disk will be lost.
SMB
If a node fails, asynchronous writes that have not been committed to disk will be lost.

We recommend that you do not disable write caching, regardless of the protocol that you are writing with. If you are writing to the cluster with asynchronous writes, and you decide that the risks of data loss are too great, we recommend that you configure your clients to use synchronous writes, rather than disable write caching.