Storage Field Day 7 – Exablox OneBlox: scale-out NAS for SME

Exablox logoExablox presented their OneBlox scale-out NAS system for small and mid-sized organizations at Storage Field Day 7. A typical customer might start with a moderate amount of data that doesn’t have too stringent performance requirements. In the next few years the customer performance requirements might double and the capacity requirements quadruple. Instead of having to buy a traditional array that would have to be sized for that 4x end-state performance and later adding extra disks for capacity, most customers would prefer to invest in small chunks and grow their system as and when it needs to. Lets see how Exablox’ OneBlox system will do this…

The architecture

During Storage Field Day 7 we discussed the current Exablox offering of the OneBlox 3308. This is a 2U appliance containing a Cavium processor, some DDR3, one internal 2,5” SSD which stores the object index, 4x 1GbE for connectivity and 8 empty 3,5” drive slots. In these 8 slots you can add your own SAS or NL-SAS drives ranging from 1-6TB. While the OneBlox support pretty much all available drives (including shingled drives), Exablox itself will not provide support for the drives you install, so make sure you have your own spares. Yes, you can add SSD drives to store the data, but the appliance is currently not optimized for SSD so you might be wasting money here.

R&D rack full of Exablox OneBlox

If you need more than 48TB of raw capacity you need to add additional OneBlox appliances. The additional appliance(s) automatically detect the first OneBlox appliance and will then form a ring architecture. There are no dedicated nodes for metadata; all appliances run all functions: storage, control and the GUI. This enables the smallest cluster size to be 1 single node. The current hardware can host 1 billion objects per appliance and scale up to 336TB of raw capacity per ring. This is the tested/supported limit with the current appliances; during the presentation it was (jokingly) said that there is no real known limit since Exablox’s R&D lab was too small to host all the appliances.

On top of the ring runs a POSIX-ish, single global filesystem. Currently you can only access the files on this filesystem using SMB shares; NFS support is still in the works and will be made available with an update of the OneBlox OS. Objects are replicated to different appliances in the ring; currently 3 replicas are made. This is simply a value in a configuration file, so if you ask Exablox nicely they might change it for you if you want additional redundancy. The shares can be snapshotted based on per-share policies; these snapshots can then be made available to end-users using a virtual namespace. By the way: if you move a drive from an OneBlox appliance to a different ring, this drive is detected and the objects can be ingested into the ring if desired.

Management of the OneBlox happens via OneSystem which runs in the Amazon cloud. There is absolute separation between user data and management data; your data remains on-site in the OneBlox appliances, the OneSystem connection is only there for management of the ring. Only some telemetry is sent off-site to OneSystem, for example SMART drive data, predictive analytics. Exablox uses this data to analyze your OneBlox ring for hotspots and to predict (drive) failures.

My two cents

I just love scale out NAS. It make life so much easier in the field: just add more nodes and you get both capacity, performance and resiliency against failure (up to a point of course). Due to the global file system there isn’t even the need to migrate or rebalance data manually; if you don’t have any quotas on a share they will automatically be extended.

You might have seen some posts on this blog about a particular scale out NAS from our big blue friends over in Hopkinton. While I will not go as far as put both products head to head, I think there are some nice similarities and differences. Both have a global file systems, both have seamless and fast cluster/ring expansion with additional nodes, both don’t use RAID to protect your data, both offer pretty much unlimited snapshot capabilities.

The Isilon system currently offers more protocols (HDFS, NFS), has much faster appliance hardware (which is customizable), has inter-node communication over 40Gbit Infiniband instead of 1GbE and allows you to configure the protection level on a file/folder basis. These are just a couple of points that come to mind; Isilon is definitely the more mature, fat-featured product. They however also have a minimum cluster size of 3 nodes per type of node and, since you get enterprise level support on every component in the system including the disks, you pay an awful amount of money for the smallest cluster. Don’t get me wrong, it’s a great system, but it’s a lot of money. What if you’re a customer that wants a long-term, NAS archive that’s currently <100TB, but you want to seamlessly scale in the future once that archive grows?

Exablox OneBlox opened up
Exablox OneBlox opened up; 8 (currently empty) drive slots near the top.

Exablox with their OneBlox offers a bring-your-own-disks approach where you buy the appliance and the support for the appliance. The disks you bring yourself and you support them yourself. Maybe it’s slower than Isilon in some areas and you can only access your files over SMB for the time being… but if that’s all you need, do you care if you can buy it at a fraction of the cost compared to competitors?

You can watch the SFD7 videos over here; especially the demo is nice to watch since it gives you a glance at OneSystem and how the ring expansion works. Dan also wrote a blog post on Exablox which is worth the read.

Exablox is definitely a company that I’m going to watch closely in the future. There’s some exciting news to be announced soon, so I suggest you do too!

Disclaimer: Even though GestaltIT paid for the flight, hotel and various other expenses, I was not compensated for my time and there is no requirement to blog or tweet about any of the presentations. Everything I post is of my own accord.