Do you have an Xbox – 360 or XboxOne? Do you use XboxLive? I do – and as an aside, I’m looking forward to a little Rise of Tomb Raider this weekend, along with Fallout 4 until my eyes bleed :-)
If you use Xbox Live, you are an EMC Isilon customer by proxy. In fact, if you use all sort of online streaming content sources, whether it’s audio, video, and other forms – odds are pretty good you’re using EMC Isilon.
This season, when the Blue Jays were making their push to make it to the World Series (congrats KC), I was all over the world and depended on MLB.TV for all my baseball goodness, from Hong Kong to Singapore to Barcelona…
… Likewise, I was an EMC Isilon customer by proxy! … And, thanks MLB.TV, and there’s always next year Blue Jays!
EMC Isilon is an amazing technology, and amazing engineering team that keeps on giving. The core innovation of the simplest (and IMO best – though I’m biased) scale-out unstructured data platform. It’s a Type 3 architecture using the Storage Architecture phylum here – and man, does it scale.
It’s funny that the EMC brand is still perceived by most as the premier “transactional storage, ‘Tier 1’” brand. Yup – we are that. But we’re so much more :-) While classic “Tier 1” block storage as a category is in slow decline by % of data growth – it remains the foundation on which so many of the services we depend on every day – think of using a bank, or making a phone call, or ordering a flight. In fact, the whole category of “transactional systems” is pretty flat overall, with huge battles of workload substitution (AFAs, SDS, Hyper-Converged systems) being the dominant market force (and refreshes and market share battles for the workloads that keep off the zombie apocalypse, where the VMAX 3 is gaining huge momentum) - but since “transactional storage” is a very important, very visible economic battlefield (“the SAN market”), it has lots of eyeballs.
So - where is the growth of information and net new workloads occurring that used to be centered around databases and VMs? Unstructured. NAS. Even more true of Object, HDFS, and similar things – all whose core characteristic is this – the information not accessed by a SCSI command :-)
This information is around us all the time, but we don’t think of it every day. It is growing in leaps and bounds.
While Isilon is known primarily for NAS (think “home directories”), the biggest single growth areas are as the underpinning of Enterprise Data lakes (simultaneous access via NFS and HDFS) – deeply integrated with Cloudera, Hortonworks and Pivotal. Along with HDFS – huge demand as the could bucket for Splunk, often teamed with XtremIO for the hot bucket data. These use cases are climbing like crazy.
While people think first that deploying non-shared storage will be cheaper (and I’ve learnt people need to learn otherwise through experience) – quickly they realize that there is great cost efficiency and leverage (particularly where they can scale compute/storage independently – which also is material in some licensing models) in being able to have a kick-ass pool of HDFS:
… And we integrate (and even bundle!) Cloudera, Hortonworks, and Pivotal Big Data Suite (including integrating in most of their management tools). More to do – but we will keep going down this path.
EMC Isilon has fared well in this macro trend towards growth in unstructured data in all it’s forms.
The largest single customer ordered 100PB at ONCE in Q1 of 2015. Amazingly, this is dwarfed by EMC’s largest single ccustomer, who has more than 1.5EB of our Object storage deployed. I would wager that most Virtual Geek readers touch these two persistence pools regularly.
So – if it’s so great, how do you make it even better? You make it simultaneously BIGGER, and also SMALLER. You make it be able to STRETCH FURTHER and make something simple even SIMPLER.
Today we announced several massive innovations in Isilon-land.
- The next generation of OneFS – a huge leap forward in the CORE. Massive overhaul to the core codebase, new kernel, bigger scale, massive lift in all system maximums. Perhaps most of all – Isilon upgrades have always been easy with rolling updates– but with complete support for non-disruptive NAS protocols (NFS 4, SMB 3.0), the upgrade to this next generation will be the last upgrade that will have any disruption for Isilon customers.
- Isilon Edge SD – our first (but not last!) 100% software-defined Isilon release
- Cloud Pools – connecting Isilon with Public/Private object stores – easily and transparently.
My favorite part? Like all our software products – it’s available easily, openly. The “Free and Frictionless” march across the EMC software portfolio continues. This means that you don’t have to listen to anyone. Not me, not to the EMC or EMC partner sales rep. Download – and try for yourself. It’s putting our code where our mouth is :-)
To understand more about each of these, read on!
First – the core next generation OneFS update.
- New kernel. Better performance, including the background function.
- Extends next-gen NAS support to SMB 3 (as well as NFS 4.x which was already supported). Critical to supporting client re-connection that is non-disruptive
- Improved Insight IQ and general file analytics data
- Rolling upgrades become completely non-disruptive – not only based on client protocol support – but a great “rollback” feature – being able to revert easily to prior platform releases. Frankly – I’m playing with a Windows 10 Surface Book as well as a new El Captain Macbook – and desperately wish sometimes they had this capability :-)
Second – the first (but not last!) 100% software-defined Isilon release.
What is it? It is a software-only implementation of the next generation OneFS data path Remember, SDS has two ideas: 1) SDS Control Planes– abstraction of the control plane from the data plane, in a way that is open – ViPR Controller being an example along with it’s Open Source trunk, CoprHD; 2) SDS Data Planes which deliver the storage persistence function and data services, in software-only form. It’s designed to be deployed on vSphere, and has great vCenter integration to make deployment and use a piece of cake.
While we’re still nailing down the deployment specs, we think the idea of linking it to the VSAN ready node HCL is likely the way to go – not only because they specify SSD configurations (needed for write destage with solid latency), but because they have been hardened and validated for material storage use (think controller specs).
Unlike the Isilon Simulator VM – which was single node, and had no production use, here you can have up to 6 nodes, 36TB of capacity – and all of the features/functions in the next generation of OneFS used in an Isilon node.
Why do I make the “not last” comment? There is interest – not from all customers, but some of the largest, to have a OneFS + bare metal SDS variant. Personally, I’ve made the comment publicly that everyone that **thinks** they are Facebook are not. I see the “cycle of SDS stupidity” (if you want to understand what I mean – read this) play out over and over again.
That said – “give the customer choice” is generally a good call – and the best way to help customers decide if they really want appliances, or whether they want SDS + DIY hardware is to give them the choice.
Third – perhaps most importantly for the longer term – Cloud Pools will be GA.
The ability to have a filesystem span into an external geo-dispersed object store is NOT NEW.
What is new is that in this new release, it’s SIMPLE.
The process to configure a Cloud Pool policy to tier files based on a simple, flexible policy to an external object store (AWS S3, Azure Blobs, and EMC ECS, including ultimately the Virtustream/vCloud Air ECS public cloud service) is revolutionary for Isilon customers.
The limit of the system capacity of Isilon clusters is 50PB in this next generation OneFS release when looking at just a cluster – but this is NOT a filesystem/inode limit, rather a function of the cluster itself.
With Cloud Pools – the maximum filesystem size is truly enormous – well, well north of 8 Trillion files.
Furthermore, the magic is that the fact that some are located in a cloud object store is completely transparent. There’s some great IP in there about handling a longer latency to the object, and making it seem like one simple massive filesystem.
This means it’s a basic economic question: what is the $/GB for a given bit of content – and how does the economic of public large scale cloud object stores change over time… but you can now use them transparently for the set of use cases that you would tackle with Isilon.
Now… I’m VERY curious to see how this works in the market. At list prices, S3 is listed at about $0.03/GB/month and S3 Infrequent Access is priced at about $0.0125/GB/month (and I’m leaving out PUT/GET and transfer prices).
One thing that is great about public cloud utility models is price transparency – the pricing is on the AWS calculator here.
What do I mean?
Well… VNX is regularly in the $1/GB price band (assuming a 3 year life, that’s $0.027/GB/month). Isilon is regularly in the $0.50/GB price band at scale (assuming a 3 year life, that’s $0.013/GB/month). Elastic Cloud Storage is regularly in the $0.20/GB price band at scale (assuming a 3 year life, that’s 0.0056/GB/month).
These are all list. I’m sure AWS discounts with volume. We do also.
Also – none of these include the costs to operate. They also assume a constant steady state use. The economics of public cloud models are interesting – people struggle to compare – because they are used to capital vs. opex oriented models – and the unit of measure is different.
What’s obvious (and this is true for people with experience operating object stores at scales in public clouds) is there is a cost of elasticity built into the models.
I LOVE the fact that customers can choose to target AWS S3, or Azure, or the public ECS instances – but I suspect there will be a ton of demand for private ECS target use cases as well!
The next generation OneFS code and the Isilon Edge SD GA code will drop early in 2016 – but would love to hear what you have to say! Are you an EMC Isilon customer? How’s it going?
Comments