I'd caution everyone about using Joyent. I company I work with used them for hosting (spent thousands on Accelerators) and had horrible IO. Clearly some other client was using all the disk IO related performance issues, yet their tech support claimed it was working properly. Note, this was all on Joyent's "new" architecture which was supposed to prevent these problems.
Joyent's tech support was arrogant, annoyed, and unhelpful. Caveat Emptor.
Bottom line was that Joyent thought that because it was using Solaris it could massively overload the number of zones per box. This was generally OK for CPU and memory sharing, but for disk IO it simply didn't scale well enough, yet Joyent's entire business model (and pricing) was based on the idea that you could squeeze a lot of zones onto one large box.
Example: Typing ls at the console would lag horribly, sometimes taking 3-4 seconds to return. Other times it would be lightning fast, and was thus declared a non-issue.
Heroku, on the other hand, has been fantastic. They even have responsive, helpful people on IRC most of the time.
There are surely a lot of very smart people there, but the ones I worked with had all drunk the kool-ade that Raid 10 on 15K spindles with ZFS was so awesome that there couldn't be IO problems.
When you have a company that is using technology that is essentially beta (like ZFS) and you have to pay large amounts of money to Sun every time you need support, that creates an incentive to pretend the problem doesn't exist. Joyent also showed some signs of serious cash flow issues (half off if you sign up for a year) which may explain why they couldn't pay Sun enough money to identify and/or address the issues.
Note: Switching the app that I mentioned to a $100/month slice (off of a $400/month zone) resulted in a massive speedup and noticeably faster disk IO, db performance, etc.
Joyent's tech support was arrogant, annoyed, and unhelpful. Caveat Emptor.
Bottom line was that Joyent thought that because it was using Solaris it could massively overload the number of zones per box. This was generally OK for CPU and memory sharing, but for disk IO it simply didn't scale well enough, yet Joyent's entire business model (and pricing) was based on the idea that you could squeeze a lot of zones onto one large box.
Example: Typing ls at the console would lag horribly, sometimes taking 3-4 seconds to return. Other times it would be lightning fast, and was thus declared a non-issue.
Heroku, on the other hand, has been fantastic. They even have responsive, helpful people on IRC most of the time.