Coming down to earth? Mixpanel's reasons for leaving the cloud

"Why use the cloud?" This is not a serious question: seriously, why worry about which computers are hosting you when you can be certain that some computers certainly are? And when the clouds are gathered by such reliable giants as Amazon, Microsoft, and Google, you can feel pretty confident in your servers.

Sure, you may love your sysadmin to high heaven, but nobody can do anything better than those guys.

And yet, during this sunny season of cloud computing, Mixpanel Engineering offers a dissenting opinion.

In this (highly-commented) article provocatively titled 'Why We Moved Off The Cloud', the admins at Mixpanel explain why they ended their relationship with Rackspace and moved to an old-fashioned, dedicated host.

Of course, Mixpanel's reasons are somewhat case-specific -- and they recognize this. For example, they observe, not everyone cares that the cloud offers

No access to bleeding edge hardware. At Mixpanel, some of our codebase is highly optimized low level C. We’ve profiled, tweaked, and made sure we’re not missing anything obvious. My point is the only way this code is going to run in less time is if we get faster hardware. Dedicated hosting providers usually stay on top of new hardware (specifically, the latest CPU’s and SSD’s). On the cloud, you’re usually stuck with whatever the provider got a volume discount on.

 
But their chief objection is simple, and, they say, 'intractable': you simply don't have control over the cloud's performance. It's not exactly that the cloud performs badly: this surely isn't true. Rather, the problem, they say, is that the cloud just doesn't always perform the same way. Like cable Internet, or your morning commute: other people determine your speed.

In their own, slightly stormier words:

The fundamental problem with cloud servers is that you’re at the mercy of your neighbors. If they decide to “dd if=/dev/zero of=/dev/sda”, there’s not a whole lot you can do about it other than migrating to a different physical server (and it’s really hard to decide whether to wait it out or migrate, especially because zero down time migrations are always little painful). 

The controversial post has already generated a lot of chatter (31 comments at the time of this post). The discussion is worth checking out -- most of the replies are strongly pro-cloud and anti-this-post -- and the tone ranges from 'that's not the fault of the cloud, that's the fault of bad cloud architecture' to 'well of course dedicated servers will have better performance, but what about downtime?' to 'what are you talking about, Amazon IS cloud computing'. Plus some useful details (e.g., BraveNewCurrency's) that simply deny Mixpanel's 'the neighbors steal your I/O' claim.

Read the full article with comments here.