Using RAM Disk for persitence or replication storage
We are evaluating Couchbase as a caching solution that can handle over 10,000 ops/sec (~roughly 6000 gets/4000 puts with 8 mins TTL for 8KB message). We used out-of-the-box configuration using a single couchbase (default) bucket with 1 replica using 4 redhat nodes that provided 32GB each of memory with local disk for persistence. On steady state, the disk I/O was 50% busy but when I removed a node from the console and started re-balance, the disk I/O went to 100% busy. The re-balance took a long time (over 40 mins) and eventually failed. Are there pointers for app/system configuration for achieving better disk I/O rates with couchbase ?
We then used a RAM disk and mounted it for couchbase to use as (non)persistence to get over the disk I/O bottleneck. We really wanted replication more than persistence. This time around, the re-balance took less than a minute and "get" errors were really low.
I was wondering if there are potential pitfalls by using RAM disk. Has anyone used this configuration in a production environment in this manner ? Will it be supported by couchbase ?