Gluster storage platform

[Handruin]

Has anyone played around with the Gluster Storage Platform? I just started reading about and playing with this tonight and I think this would make for a fantastic way of having a distributed NAS environment that could grow and mirror for as many nodes as you decide to build on it.

I just built an example setup using VMs with 4 nodes and it was so easy to add nodes using a pool-based IP system. You can mirror or span a virtual file system among 2 or more nodes. Adding nodes is brain dead simple and you can grow almost infinitely. I created 4 virtual machines each with 10GB of space and then created a stripe across them.

With this kind of setup, each time you build a NAS and fill it up, you just add another one to your pool of systems and keep going. No need to manage a complex system of storing files. For important stuff, mirror it among two or more nodes for increased protection. I tried a test with the mirror and was able to power down one of the two nodes but the data was still accessible from the other mirrored node without problem. The stripe that I had built with 4 nodes was not accessible with one of the four nodes powered down as expected.

Just thought I'd share if anyone else has an interest in this kind of scalable and elastic platform.

 

[BingBangBop]

Never even heard of it before. However at $1500 per storage node per year it isn't exactly cheap.

 

[Pradeep]

Never even heard of it before. However at $1500 per storage node per year it isn't exactly cheap.

That's for the support contract. The product itself is open source and free to use. For deployment in an enterprise I would need 4 hour response and that's $8500 per node, however it looks like they have an "elite" akak site license for bigger deployments.

Just a single "optional" feature on a lower end NetApp box can easily run you into the five figures, and that's usually just a licensing thing to enable what you already have in the filer.

Interesting how you can tune it for your needs. Wonder how the "cloud" option works. Increase response time to maximize lag, discard random chucks of data to never be seen again?

 

[Handruin]

The idea sounded neat to me for smaller scale home NAS units more than enterprise. I never looked into their licensing to consider it for work or enterprise needs, but I can see those as viable concerns for you guys.

After experimenting with it some more I did notice some problems. I performed some node failure testing and it didn't seem to recover as gracefully as I expected (or perhaps hoped). They claim the software is self-healing, etc. I tested the failure of the management node and after recovery it no longer found my other 3 nodes that it was configured with. In a separate test, I failed one of the nodes that was part of a mirrored set while copying data to the other half of the mirror. I did not expect my copy to completely halt and time out in this situation. I expected this could tolerate a failed mirror and continue on the source, but perhaps for safety of data, they don't allow it to continue during a write operation.