We've been looking at running Membase behind a client-side moxi, and have noticed we see what appear to be performance issues tied to our moxi instances (running 1.6.5).
What we see: when running against a 4-node Membase cluster, with 1 instance of moxi running on a client (Linux), we see a batch load script run about 800 transactions/sec. We tried some tweaking of the various thread and concurrency settings, but didn't see any obvious improvement. The odd thing however is: we get noticeably improved performance (almost linear) by adding extra instances of moxi, and configuring our script to hit all of them.
Are there any known issues, or does this trigger any points in anyone's mind?