Isilon Infiniband: Go Big Storage Or Go Home

By | Backup, Isilon, Replication, Storage | No Comments

It wasn’t so long ago that a 1G disk drive made your computer the envy of every other nerd on the block. Now the tech enthusiast sees green when you have 4 to 10 TB. This is just at home!

Currently in the enterprise space we are seeing regular caches of data well in excess of 100TB. Oddly, the file systems holding this kind of data struggle when the files get much above 16TB in size. So we start carving our repositories into all sorts of folders. Sometimes this makes sense, but more often than not it is to get around the technology limitations of the storage subsystem being used.

The ideal storage system would be able to scale in terms of performance and capacity. This solution could mix and match SAS, SSD and near line SAS technologies so the right data is on the correct tier. Snapshots and replication, native, of course. The solution would allow us to create folders – but not be bound to them. We would be able to slide data from one folder or file system or from one protocol to another to another without having to reorganize the storage. There would be no taking losses due to repartitioning or any loss of performance due to hot spots. This storage would have internal data protection which doesn’t require RAID sets or aggregates that require bunches of parity drives or hot spares. More drives would just mean more storage efficiency. How awesome would it be to literally have 1000’s of drives working together for a single workload with petabytes of continuous space under your home drive or operations folder or development folder?

No crazy ideas here.
Not an odd desire.
Not a unique problem.

Yes, there are ways to build this “ideal” storage system described above yourself – if you have time for a serious science project.

But for the professional IT shop, who are looking for something that is fully supported, shrink wrapped, and ready to run – ENTER ISILON – stacks of nodes with lots of drives and a really smart OS, purpose built for serving (a lot of) files.

I have always seen the benefits of node based architectures, but also the downfall of the “backplane on a LAN” approach. Any kind of storage balancing or motion around the array could crush the network and then starve the end users coming over the same network, or in some cases; the same ports. Isilon sidesteps this common issue and threw an Infiniband network behind its nodes. Now we are talking about a backplane. The best part is that it is expandable as the number of ports on the switch. As the switch technology develops, so does the size and performance of the entire array.

OK, so we all don’t need to go this big, but what a great way to get there if you did. Crazy scalability, crazy performance, all shrink wrapped with support. I just have to figure out how to get one for my home theater…

Photo Credit: ChrisDag

Isilon Storage (Still) Supports “Block Headed” IT

By | EMC, Isilon, Storage | No Comments

Call me crazy, but imagine this: a high performance, highly available storage system that uses off-the-shelf components and no RAID. And add to that ease of use and the ability to scale to petabytes in minutes. On top of that, throw in some features like snapshots, site-to-site replication, and intelligent auto-tiering. Seeing it yet?

The picture I’m painting is a system that has NetApp throwing stones and EMC wondering what’s next for Celerra, now that it has spent $2.25 billion on keeping it out of NetApp’s hands. If you haven’t figured it out yet, I am thoroughly enamored with our friends at Isilon:)

Okay, so I asked you to imagine a highly available system without RAID—yet, not without data protection. Isilon’s scale-out architecture distributes incoming writes across nodes mirroring what they deem small files (less than 128KB) and striping files 128KB and larger in 16KB chunks across each node in the grid with parity.

Here is where it gets interesting, though: Protection can be set on a block and node protection basis, so you can define how many drives or nodes you want to be able to survive losing. When you lose a drive or a node, the grid rebuilds from parity across free space on the remaining nodes.

But that’s only the first reason for my enamoredness (yep, it’s a word) …

I’m a dyed-in-the wool “Block Head,” and I know Isilon is a NAS platform, but they really got my attention with the OneFS OS. They handle all the major protocols: NFSv2, v3, and v4, pnfs, FTP, HTTP, and CIFS/SMB. Authentication support for LDAP, NIS, Active Directory and local users and groups are all supported. For backup: NDMP is supported, directly to tape through a backup accelerator node or across the front-end Ethernet via 3-Way NDMP.

For us “Block Heads,” Isilon supports iSCSI—after all, a LUN is just a file, right?


144 Node Isilon Cluster

Photo Credit: Paul Stevenson