On 19 October, Dell EMC made an announcement about the All-Flash Isilon nodes. This product line has been code-named “Project Nitro”. The announcement was timed to occur during the first-ever Dell EMC World event.
I’ll walk through what was announced, and then provide additional details.
What Dell EMC Announced
If you’d prefer to read the official Dell EMC press release, you should, but I’ll summarize it here.
Dell EMC announced an All-Flash version of their Isilon scale-out NAS platform. They said it changes the All-Flash game, not only by providing multi-protocol access, but also by bring All-Flash to unstructured data.
A single 4U Isilon All-Flash system provides:
- 4 nodes
- up to 924TB capacity
- up to 250K IOPS
- up to 15GB/s total aggregate bandwidth
The Isilon All-Flash can offer up to 80% storage utilization. Customers can get further storage efficiency, by using Isilon’s SmartDedupe and reducing the size of their data by as much as 30%.
One Isilon All-Flash cluster can support up t0 100 systems, providing:
- 400 nodes
- 92.4PB capacity
- 25M IOPS
- up to 1.5TB/s total aggregate bandwidth
The Isilon All-Flash will seamlessly integrate with existing Isilon clusters and allow for automatic tiering of data.
What We Actually Know
First and foremost, if you’re at all familiar with the Isilon platform, you’ll have immediately noticed that the Isilon All-Flash is built on completely different hardware than existing Isilon nodes. The four nodes in a 4U form factor is a departure from providing a single node in either 2 or 4U, depending on how many drives it has.
SSDs can take up less physical space than HDDs, which is a large part of how Dell EMC is able to achieve this higher node density. Each Isilon All-Flash node is half-width and 2U in height, allowing for four nodes in 4U, in two rows of two nodes each. Each node contains 15 SSDs.
With this new hardware comes some new requirements. In the past, the minimum size of an Isilon disk pool was three nodes, and a pool could be expanded one node at a time. The Isilon All-Flash will require a minimum of four nodes and can only be expanded in pairs of nodes. (More on this later)
It’s definitely worth noting that the capacity numbers that Dell EMC mentions in their announcement are all RAW capacity numbers. The SSD options available are 1.6, 3.2, or 15.4TB drives, meaning that customers will be able to get 96TB, 192TB, or 924TB of raw capacity in a single 4-node, 4U chassis. Once the usual Isilon data protection scheme is applied, that 924TB of raw capacity will likely be closer to 600TB of usable space. Isilon will allow customers to achieve the 80% storage efficiency quoted, but that requires more than four nodes.
While I’m on the topic, drive sizes cannot be mixed within a node. I’ve asked multiple times whether different drive sizes can be used within the same disk pool (a grouping of like nodes). The repeated lack of an answer to the question leads me to believe that the answer is “no”. This means that if you want to use different sizes of drives, you’d need a minimum of four nodes per each drive size you wanted to use, and would also need a SmartPools license. (In addition to providing policy-based automatic movement of data between tiers (disk pools), SmartPools is what allows different node types to be used in a single cluster with a single filesystem.)
With the new hardware, Dell EMC will offer customers a choice of which of two different back-end networks they’d like to use: InfiniBand (IB) or 40Gb/s Ethernet (40GbE). (An Isilon cluster uses a separate “back-end” network for all communication between nodes.) The IB connection will be the standard QDR option used in existing Isilon nodes today. For the 40GbE option, the choice of switch vendor is “still TBD”.
This back-end option choice matters. If a customer plans to add Isilon All-Flash nodes to an existing cluster, they’ll have to choose an IB back-end, as that’s the only back-end supported for existing nodes. Stated differently, if a customer chooses the 40GbE back-end option, they will not be able to add their new Isilon All-Flash nodes to an existing Isilon cluster.
Another major “under the hood” change to the new Isilon nodes is in the implementation of the filesystem journal. Unlike other Isilon nodes, the Isilon All-Flash will not use an NVRAM journal, but will instead take advantage of the SSDs for the journal. In order to provide the same level of data protection that the NVRAM journal does, Isilon All-Flash nodes work in pairs to mirror the journal. This new journal handling explains why new nodes need to be added in pairs.
Lastly, the Isilon All-Flash nodes will require an as-yet unreleased, still under development version of OneFS. This new OneFS version is code-named “Freight Trains”. This means that customers who want to add Isilon All-Flash nodes to an existing cluster will first need to upgrade that cluster to that new version of OneFS.
Dell EMC is taking pre-orders of the Isilon All-Flash now.
GA for the Isilon All-Flash nodes is scheduled for the first half of 2017.
The folks I’ve spoken to believe that the absolute earliest we’ll see GA of the Isilon All-Flash would be mid-May of 2017.
My thoughts on the Isilon All-Flash are below, listed in no particular order.
- When I first heard of Dell EMC’s plans to build an All-Flash Isilon node, my initial reaction was “Why?” I couldn’t think why anyone would need that extreme performance boost for file data. Then I realized that I was thinking of home directory or long-term file storage uses cases, which don’t need All-Flash performance. There are, however, other use cases that would benefit. More organizations are using Hadoop as their analytics engine and Isilon is already a great fit for Hadoop. Several of my customers are doing gene sequencing, and that uses large files.
- There’s a lot of customer interest in All-Flash for Isilon. Of course, now that they know about it, no one really wants to wait six months for it to be available.
- Given the above point, I’m really not sure if Dell EMC made a mistake announcing Isilon All-Flash so far ahead of its availability, or if it was a brilliant move. On the minus side, some folks might get tired of waiting and look for alternatives. On the plus side, some folks might decide that this is a good reason to stick with Isilon. Only time will tell which way this one goes.
- I think the Dell EMC Marketing folks had way too much of a say, and the Engineers far too little say in what was actually announced. I get that the “nearly 1PB of capacity in 4U” is a much more exciting message than “after drives are formatted, the filesystem overlaid, and data protection is applied you’ll get 600TB in 4U”, but the second message is actually true… When I was at EMC, we prided ourselves for selling based on usable capacity while competitors would only talk about raw capacity.
- Given the number of issues customers have had with OneFS upgrades recently, I don’t believe that the “seamless integration with existing Isilon clusters” claim will prove to be true. Don’t get me wrong — both SmartPools and CloudPools are great, and do allow for the seamless addition of a new tier of storage to an Isilon cluster — provided that you don’t need to change the software running on the existing cluster. With the requirement to upgrade to what will essentially be a “dot zero” release that has had minimal field-testing, I’m not sure how seamless things will actually be. (I’d love to be wrong about this point.)
- I really want to get my hands on this hardware and see what it can do. I feel like it’s going to work well (and amazingly fast) in a standalone cluster.
(Hey, Dell EMC, any chance you’d like to help sponsor a storage/virtualization blogger’s home lab by making a small donation of a single 4U appliance…?)
- While I keep hearing that the Isilon All-Flash can be pre-ordered, I’m not sure whether they’ve actually taken any pre-orders yet. I base that solely on the difficulty I’ve had pinning anyone down on even a price range for the new nodes… (And by “difficulty” in this case, I mean, “no one I’ve talked to at Dell EMC seems to have any idea what this will be priced at yet”.)
- Despite any misgivings I might have about how smoothly this will actually roll out, I do believe this will be a great addition to the Dell EMC storage line. The more time I spend thinking about it, the more use cases I can think of for it.
What are your thoughts on the Isilon announcement? What use cases do you see for it? Share your thoughts in the comments.
Pingback: A Closer Look at the All-Flash Isilon, aka Nitro - Gestalt IT
For 4 nodes storage efficiency would be 6+2 = 75%. It should climb to 16+2= 89% as the cluster grows. So that 4 nodes is more like 700-800TB depending on the size of the pool.
Pingback: Summary of Monday’s Announcements at Dell EMC World 2017 | GeekFluent