How do LEB providers build their SAN? Buy commercial products or build with free software?

chihcherng

SAN is needed for some virtualization applications like live migration. Because LEB providers obviously need to keep their cost down, I wondered how they build their SAN infrastructure. Buying commercial products like NetApp sounds expensive to me. Is it possible to build an SAN with Openfiler or Solaris alike? If so, how does it perform?

fan

@StylexNetworks is using SAN with Onapp I think.

rds100

For OpenVZ live migration SAN is not necessary.
I doubt many LEB providers use any sort of SAN. I think @prometeus has one, but he has it from before, didn't buy it to start offering LEBs out of it.

prometeus

Yes, last year I used the old fiber channel san with some nodes. Maintenance for that expired, all the spare disks were in use so I decided to get out of production it :-)

We are also using one of our Coraid AOE SRX for some vps. In both cases of course SANs were used because they were in place :P

concerto49

SAN isn't often a smart choice. It's often the single point of failure and due to the complexities the thing that fails the most. Amazon AWS more often than not fails at the SAN level each time it dies.

stylexnetworks

Care to back that up with some facts?

raindog308

@concerto49 said: SAN isn't often a smart choice. It's often the single point of failure and due to the complexities the thing that fails the most.

Completely disagree. A proper SAN should not have any single point of failure - frames, controllers, switches, disks, etc.. The good stuff - Hitachi, EMC, IBM, etc. - does not at any point. Of course, you won't be serving LEBs off it :-)

The only SPOF for a SAN is the administrator, which admittedly is a problem, though that's true of anything you let humans touch.

@concerto49 said: Amazon AWS more often than not fails at the SAN level each time it dies.

I don't think AWS uses SAN. I know Azure does not and I think their architecture is very similar.

concerto49

@raindog308 said: Completely disagree. A proper SAN should not have any single point of failure - frames, controllers, switches, disks, etc.. The good stuff - Hitachi, EMC, IBM, etc. - does not at any point. Of course, you won't be serving LEBs off it :-)

I'm not saying SANs have single point of failure. I am saying THE SAN IS the single point of failure. As in looking at the complete system, it is the SAN that fails out of anything.

@raindog308 said: I don't think AWS uses SAN.

They do. Read up on a few storms where they crashed and most of the time the issue lies in the SAN in any recovery or automated 'cloud' infrastructure they have.

This is why Cloud Sigma go with distributed file system instead of a SAN.

http://www.cloudsigma.com/en/platform-details/our-infrastructure

Distributed storage provides us with much higher levels of redundancy and resilience >than any SAN solution could. SANs represent large single points of failure and require >huge networking traffic. When used in the cloud this translates to major availability >losses if a SAN fails (and they do!) as well as reduced virtual server performance and >increased network lag.