Nuts & Bolts: New Datacenter(37signals.com)
37signals.com
Nuts & Bolts: New Datacenter
http://37signals.com/svn/posts/2471-nuts-bolts-new-datacenter
2 comments
For the record, we used to run MogileFS. The Isilon gear is a radical improvement and the price premium isn't anywhere near as bad as the numbers you presented. It was more expensive, but safety of our customer data is our most important job and we're more confident in the Isilon gear than we ever were with MogileFS.
If you had tens of thousands of customers, would you bet your company on a homebrew MogileFS setup? And are you sure you'd save enough money to justify a port?
Yes. I'm confident in my abilities to overcome absolutely any problem that might occur. For the same reasons I choose Linux and open source solutions over Windows or other proprietary solutions: there's always a way out. Isilon, Cisco, or Oracle only provide the illusion of certainty.
Your confidence in your abilities provide only an illusion of certainty. Maybe if you had enough time you could fix any problems that occur with open source software, but when you have a deadline it's nice to have experts to help you. You can find open source experts, but with a commercially supported product there is a much clearer chain of responsibility.
Experts? You mean the $10/hr guys 5k miles away that support Microsoft, Oracle, or NetApp products?
Responsibility is meaningless. Competence is what matters.
Responsibility is meaningless. Competence is what matters.
While I support your argument, you have to be careful it doesn't turn into arrogance.
I remember one time being promised by the tech lead on a project that an inexpensive networking solution would be just as good as the expensive kit we had used for development. Turns out he couldn't make it work - the cheap kit just wasn't as good as the expensive stuff we had developed against.
However, I was the one standing in the customer's office being shouted at by three of their unhappy managers as their time critical project was delayed until we got the proper kit ordered and the tech lead sulked back at our office.
Sometimes expensive kit is worth it.
I remember one time being promised by the tech lead on a project that an inexpensive networking solution would be just as good as the expensive kit we had used for development. Turns out he couldn't make it work - the cheap kit just wasn't as good as the expensive stuff we had developed against.
However, I was the one standing in the customer's office being shouted at by three of their unhappy managers as their time critical project was delayed until we got the proper kit ordered and the tech lead sulked back at our office.
Sometimes expensive kit is worth it.
They were using MogileFS, perhaps they outgrew it: http://groups.google.com/group/mogile/browse_thread/thread/a...
Do some research before you shit on some technical decision.
Do some research before you shit on some technical decision.
my experience (and my experience in this is a few years old, so take that for what it is) has been that commercial SAN/NAS products have a significant reliability or performance edge over their open-source competitors. Note, I said reliability /or/ performance; the key bit you get from netapp or emc that I can't do with a linux box is cache coherency across multiple heads.
If you fire the money cannon at NetApp, they'll give you two heads, each with write-back cache in it. when you make a write, it will write to each head before returning, so either head can crash or lose power at any time without interrupting anything, and you get the speed of write-back cache, which is pretty great.
Now, the problem is that NetApp and friends charge a /ridiculous/ premium for this. I'm not saying it's a good deal. I'm just saying that /right now/ as far as I can tell, there are things you get from paying netapp and the like that you can't get from freenas.
Personally, I'm cheap, so I use local storage, and end up paying a whole lot less for disk. But if I was successful at selling a product that was not tied to my costs? then maybe my decision would be different.
If you fire the money cannon at NetApp, they'll give you two heads, each with write-back cache in it. when you make a write, it will write to each head before returning, so either head can crash or lose power at any time without interrupting anything, and you get the speed of write-back cache, which is pretty great.
Now, the problem is that NetApp and friends charge a /ridiculous/ premium for this. I'm not saying it's a good deal. I'm just saying that /right now/ as far as I can tell, there are things you get from paying netapp and the like that you can't get from freenas.
Personally, I'm cheap, so I use local storage, and end up paying a whole lot less for disk. But if I was successful at selling a product that was not tied to my costs? then maybe my decision would be different.
tldr: they moved from rackspace to their own colocated servers at someone else's datacenter, they didn't build a new datacenter. not sure why anyone would really care about this...
It do interested me tought, especially how they designed their hardware architecture to match their need. Their apps is quite different from other web startup, I think each user got his/her own DB instance. We all heard stories how facebook or twitter which data is stored in one huge DB scale. I'd like to know how 37signals did it and looking forward to their next article.
Moving to their own servers makes sense though. No doubt that's going to save them over paying a huge premiuim for Rackspace. With three full-time people it's a good move.