One initial benchmark I ran was a slow query. It took 8 seconds on my laptop and a whopping 6+ minutes on the Joyent Accelerator.

So I contacted Joyent support.

Turns out Joyent has two different hardware configurations, one that is “old” and one that is “new”. The “old” ones have dirt slow IO.

After Joyent switched me to a 1GB accelerator on the “new” hardware, the same query now executes in between .8 seconds and 1.2 seconds (probably depending on the overall load on the box).

So I suggest you contact support and verify that you’re on the “new” hardware configuration at Joyent. It took me a while to get this to happen… Joyent tried to blame the query, mysql, my app code, my my.cnf settings, etc.

Thanks to this benchmark I’ll probably give the query at try on a Large EC2 instance just to compare…

Note: I think Joyent should be proactively getting people off of the atrocious “old” hardware config, but that does not seem to be the official policy. a senior tech at Joyent told me that doing so is a “least desirable option” from his perspective. What surprised me most about this was that Joyent was not proactive in moving people off of the old hardware. Once the decision to move was made, everything went smoothly, but it was not fun trying to get anyone there to acknowledge that there are IO issues. The first line of response is “it must be your query, your my.cnf, or your app code”.

That said, I suspect that the real issue is that 2.5″ laptop hard disks are quite slow (5400 rpm, usually) compared to desktop or server class hard disks.

Thanks very much for the generous offer! I’ll definitely ping you as my peanut (what some would call a brain) overheats :)

I don’t classify as a guru I think, but I’m doing application-to-application protocols over XMPP for 2 or 3 years now, so ping me (jabber id is equal to my email address) if you need anything.

Always ready to help out XMPP projects :)

Best regards,