I don’t ascribe to the world view that “all things will bit-flip to software loosely coupled to commodity industry standard packaging”. People who speak in absolutes, people who are polemic – they get a voice and an audience, because some people’s mind can’t handle nuance (look at the political process in the US :-)
Here’s how I see it:
- There are absolutely use cases where storage appliances (tightly coupled software/hardware combos) continue to fly (see the AFA battleground).
- There is no question that the networking domain is being impacted by merchant silicon, but if you saw the results of our Nexus 3000 and Nexus 9200 testing (used in VxRack), you would see that custom silicon has a long road ahead – even at the same time that merchant silicon becomes an increasingly important part of the networking ecosystem. More on this another day.
- There is no question that there is room for all sorts of new very custom hardware in the NAND and NGNVM (next generation non-volatile memory) ecosystem (think of things like DSSD). This is an area which is early in the hardware innovation cycle (see the SIN(X), COS(X) analogy here).
…BUT – I want to be clear – anyone who says “software defined + industry standard hardware” is a bad idea, or not economical at a broad set of use cases and scales…. Well, to me they sound like doth protest too much.
I heard that position recently on a podcast from someone…. but no surprise, that someone is someone who only plays in the traditional external storage market, surprise surprise! :-)
There is no question that SDS and SDN are having a big impact on the infrastructure domain – and while it is still early days – their impact is accelerating, not the other way around.
And I want to be clear – this is true in all forms of “packaging”: 1) software only (most flexible); 2) software + validated list of hardware (still pretty open); 3) firmly packaged with commodity hardware (most supportable for most customers). While most customers indicate they want the first, they tend to end up wanting to buy the latter. More on these different forms of packaging later, though I’ve discussed my observations from a TON of customers here: Is the dress white and gold – or blue and black? SDS + Server, or Appliance?
EMC and VMware are both leaning into the transition, and leading this Software Defined charge – and today’s big ScaleIO 2.0 release is only the most recent (and not the last) milestone.
- VMware NSX 6.2 – a huge release, and bringing micro-segmentation and overlay networks to thousands of customers, using vSphere, but also a broad range of non-vSphere use cases.
- VMware VSAN 6.2 – a huge release, and bringing the most vSphere integrated transactional SDS to thousands.
- EMC Isilon SD Edge – a huge release, bringing the industries best and most widely deployed scale-out NAS to thousands via an SDS option… and simple easy free and frictionless download here.
- EMC Elastic Cloud Storage 2.2 – a huge release, improving the industries most advanced web scale, geo-dispersed S3-compliant object storage, with almost 2 Exabytes deployed, and growing fast. Need a scaleable, simple blob store for the new world of container abstraction and cloud native applications – check, ECS snaps right into Cloud Foundry via a simple service broker here… and simple easy free and frictionless download here.
- … and now, EMC ScaleIO 2.0 – a huge release, expanding the reach and capability of the most scalable open transactional SDS on the market.
ScaleIO 2.0 is used by customers who want something wicked fast, wicked low-latency, and can scale like nothing else in that transactional domain.
ScaleIO is used by customers who want something that is open. ScaleIO integrates with OpenStack Cinder, with VMware, Hyper-V, you name it. ScaleIO has broad kernel-mode integration with the big Linux distros. It has Mirantis Fuel plugins (Ubuntu JUJU charms coming very soon), and rich and mature vCenter plugins. Need something transactional for the new world of container abstraction and cloud native applications to go with your ECS blob store? Check, ScaleIO is the answer. ScaleIO already has integration with Docker volume management, with Mesosphere.
ScaleIO 2.0 is available immediately via free ($0) and frictionless (not a single barrier for you to use) at this link: http://www.emc.com/getscaleio. This has no timeout, no feature limit, no capacity limit. You don’t get support. If you want to use it in a production environment when you need support – buy it.
So – what’s new in the release?
Well – there are 3 “anchors” – Enriched Securty, Enhanced Resilience, Extended Platform support, and as we grow the ScaleIO installed base – they increasingly are guiding us based on what they want. There’s a lot of their requests in here :-)
Enriched Security
- Support expanded network protocols (IPv6) – supported pervasively (inside ScaleIO clusters and for clients).
- Integration with existing customer password authentication systems (AD/LDAP).
- Ongoing automated software component validation which allows a strong authentication of ScaleIO components and will prevent a rogue or unknown component from joining the network
- Increase intercommunication security between ScaleIO components (SSL) via a secure connection between external components (CLI, GUI, ScaleIO Gateway) and the MDM using Secure Socket Layer (SSL) encryption
- FYI, one common request is Data-at-Rest encryption. ScaleIO has always had a lightweight obfuscation technigue, but that ain’t D@RE. D@RE is coming soon to ScaleIO, leveraging our great CloudLink technology (which won’t create a SEDs dependency) is coming soon later in Q2.
Enhanced Resiliency
- Ensured data integrity across the entire data path – by calculating an in-flight checksum on the application data directly on the client and verifying it before it is written to ScaleIO storage media, and validating on read.
- Increased redundancy for ScaleIO system
- One of my favorites – ScaleIO 2.0 will maintain cluster performance when doing routine maintenance (via the introduction of a formal maintenance mode). This should be a familiar concept to many, but important. ScaleIO has crazy fast distributed IO and crazy fast distributed rebuilds – but sometimes you just don’t want to do that, because you’re just rebooting/updating a node. Maintenance mode prevents an automatic rebuild when a node is rebooted for maintenance purposes, and automatically tracks for any data changes which affect the rebooting node and syncs the delta when the node comes back online.
- There’s also the ability to proactively redistribute so that there is no period of increased exposure, regardless of how short that may be.
- While one reason that ScaleIO scales so well is because the metadata is not in the IO path, the meta data manager (MDM) is a critical ScaleIO cluster component. ScaleIO 2.0 supports a cluster of 5 MDM nodes (3 repositories + 2 tiebreakers), which enables continued operation after multiple MDM node failures.
- Address potential system problems faster (via ESRS). As EMC products mature, they become more and more integrated with the EMC support systems. ScaleIO 2.0 uses EMC Secure Remote Services (ESRS) system as the vehicle for both Call Home and Remote Support access.
- Reduce upgrade time between releases (NDU)
- This delivers on one of the most important promises of SDS stacks – elimination of the dreaded data migration during any hardware/software update. SAN migrations are not fun. The “external array” market (us included) are working hard on this – but its tricky. Some package “head upgrades” as non-disruptive – and they are… Up to a point. VNX and CLARiiON have done this for years (long periods of non-disruptive upgrades punctuated with upgrade events). Some of the new startups nicely market this as “no migrations ever”. I’m skeptical. IMO - they just haven’t been around for long enough, but time will tell. Eventually on-disk/flash/persistence format changes, or just too much hardware level change (think of the FC loop to SAS transition, now imagine the PCIe/NVMe transition) get you. At the high end of the external storage array market, VMAX and its competitors do neat things to make WWN portability disguise data movement – and perhaps that can be done well enough to make data migrations a thing of the past. Conversely, the distributed SDS models make migrations a thing of the past (one of the fundamental opex benefits of SDS and hyper-converged models). Like VSAN (including through on-disk format changes from 5.5 to 6.0), Scale IO delivers a non-disruptive upgrade process between 1.3x and 2.x versions of ScaleIO.
Extended Platform Support
- ScaleIO 2.0 has a huge leap forward in all the ways you interact with it. Yup, has a nice RESTful interface. Yup, clean APIs. Yup, Ansible integration. Yup, Go API bindings. Great vCenter plugin. But, the ScaleIO UI was never the best. ScaleIO 2.0 provide a simpler management experience (GUI), which enables the ability to perform volume and client management operations and view related performance statistics in that UI.
- Broadened container ecosystem support – including ScaleIO with Docker (CoreOS support). ScaleIO 2.0 supports of CoreOS – available via Request for Product Qualification (RPQ) only. CoreOS support enables very high density of compute in clouds using Docker technologies native to CoreOS. BTW – we’re furiously working on deep VMware Photon OS support, expect more soon!
- Expanded support for Linux. Responding to overwhelming Ubuntu demand –interestingly to me – I would have thought RHEL would have been the way-out front source of Linux demand) Specifically, ScaleIO 2.0 supports Ubuntu OS 14.04 LTS. What’s nice is that it polls the Ubuntu repository for kernel upgrades and upgrade automatically.
- Provide a tighter integration with the OpenStack ecosystem. How cool is this – ScaleIO 2.0’s Cinder Driver is fully contributed and integrated with OpenStack Liberty release. Customers can just download OpenStack and get ScaleIO drivers as well. Mirantis Fuel plugins developed by EMC Code enables easy deployment of ScaleIO and OpenStack, and I’ve sent a nice shiny new VxRack System 1000 to Mirantis so they can beat on it. With Ubuntu support we can move forward with JUJU charms enabling simplified deployments with Canonical Ubuntu OpenStack (not GA yet – but coming soon!)
And when I say that ScaleIO is unparalleled in how it performs, and how it scales – we’re willing to put our money where our mouth is.
We sent a VxRack Flex Node (which uses ScaleIO as the SDS) to StorageReview. If you visualize the VxRack System as the engineered system, it gets populated with nodes (and there are variety of nodes for a variety of use cases). A VxRack Flex node is ScaleIO + validated hardware. BTW – with StorageReview there was no “don’t publish” safety net – the facts would be what they would be.
It’s interesting to me - people assume that EMC would sometimes that we ask people to pull their testing, or blogs – I can say with confidence that I don’t know of a SINGLE case where we’ve done that, and I would always fight for more transparency, not less. I will fight for people’s ability to publish, share, and post. Even when something gets posted I don’t like (or wish I could retract), I will always strive to ADD versus censor. I don’t understand why startups (with less to lose, frankly) would do otherwise – and man, they seem to censor a lot – but I guess they have their reasons.
What did StorageReview find?
Well – they found that ScaleIO is simple, very fast, scales like a no transactional stack that they’ve ever seen (or I have for that matter!) – and is very, very efficient. This last one (resource consumption/efficiency) is very important in terms of CPU/Memory resource use – very important for SDSes that are pressed into hyper-converged use like VSAN and ScaleIO. I would be embarrassed to have a transactional SDS that needed multiple vCPUs and GB of memory – even more embarrassed if that was needed multiple times over in a cluster. I frankly wouldn’t be able to keep a straight face if that was needed on EVERY node in the cluster…. And that’s even before you get into the debate of kernel-mode vs. guest/user space.
Don’t take my word for it.
Read the first article.
“Some people might call the enterprise storage market a dull space, but the EMC ScaleIO solution has easily piqued our interest in a way few solutions have.”)
Then read the second article.
“We can't help but to be shocked by the level of performance offered. ScaleIO managed to be one of the few platforms to smash it out of the ballpark in all areas of our scaled MySQL test… First, throughput was phenomenal, breaking records by an incredibly wide margin... even at near full capacity…. Second, application latency remained nearly flat through an ever-increasing testing environment…. Third, even under rising application loads ScaleIO managed to keep peak latency in check, which is very important in a web-scale environment where swings in demand could cause other applications to suffer if response times creep too high… Sure, it's easy to say that the ScaleIO nodes did so well because they're all-flash. As the numbers show though, the system easily coped with the workload at full capacity, something very few flash arrays can do while keeping latency in check at the same time.”
Then read the third article.
“As the VxRack Node works through our testing regimen, we're again talking about the system setting records.”
And no, we didn’t pay them :-)
Now, I want to make something clear – we embrace all forms of SDS packaging (and there are 3).
- Software-only.
- Software + bundled hardware.
- Integrated appliances/engineered systems.
The delta between an integrated appliance and an engineered system is simple:
- Appliances “start small, scale easily”. They DO NOT make you think about “how would I scale to hundreds of nodes” up front, but instead say “just use whatever switch”. This is most popular in SMB/SME customers and Enterprise ROBO situations. As much as people claim “web scale”, it’s important to realize that this is the largest single segment using Hyper-converged and SDS models today.
- Engineered systems contemplated very large scale, and the corresponding network domain up front – even at entry scales. That’s why no hyper-scale player actually uses hyper-converged appliances. Rather what they hyper-scale folks do is they use SDS models with rack-scale systems they consume or design themselves. This is also the domain of the Enterprise Datacenter, where “scale easily and in small bits” is good, but solutions must also “scale big”. In the Enterprise Datacenter, SDS, SDN and Hyper-converged Rack-scale engineered systems are in the early days of cannibalizing both the external storage market and the Converged market. Again, people shouldn’t buy into the marketing hyperbole – it’s early days. If you add up the whole SDS/Hyper-Converged market – it’s a small fraction of the more classic system architectures. But – the benefits around operational simplicity are pronounced, and SDS/Hyper-Converged are out-growing the classic system design model, so marginalizing them is a bad idea – and embracing is the way to go.
To make this distinction clear, VxRail and VxRack momentum is HOT, so I will use da to help further clarify, data and perspective that perhaps I can uniquely help provide.
- Looking across the customer base, the small, simple VxRail has an ASP of around $93K. They want simplicity and low entry point more than anything else, and that market is nearly 100% vSphere. Yup – there are customers who are way below the $60K list price we touted at launch. Those ones are offset by customers buying 2 appliances, and a smaller number buying 4 and an even smaller number buying 8 or more. But – the average = $93K.
- VxRack Flex Nodes start around $120K list, and the VxRack System starts at around $350K. Even with a single VxRack Flex node – so why the delta between a standalone VxRack node and a VxRack system? Well – a VxRack system is designed to scale, which means snapping in nodes, and racks – which means the smallest config still includes ToR switches, management networks, racks, cabling, and the software that manages and monitors the system integrates the network domain. Also – some of those customers are very vSphere centric, but the demands for heterogeneity are much stronger here. The ASP (not entry price) on VxRack Systems on average currently hovers around $4M. Yup – there are some customers at $350K, and there are some customers who are deploying 500 nodes at a time (who pull the average way above $4M). Imagine how you would need to design and operate for that kind of scale – network (and management/support of said network) simply cannot be an afterthought – that would be ignorant or insane.
Ignore what I say, or what competitors say for a moment – the above are facts, and reflect the fact that hyper-converged appliances and hyper-converged rack-scale systems are not the same (even if Gartner doesn’t – yet – recognize the distinction). Ultimately – growth in the market is the ultimate judge, so we’ll see if I’m right or not!
Summarizing this topic of SDS and packaging….
The starting point is “are you a customer who wants the most integrated vSphere stack in the world”, or “are you a customer who wants an open heterogenous stack, inclusive of vSphere”? That simple question then results in simple options:
For Virtual Geek readers that are more focused on the vSphere ecosystem, or have a world-view that things that at that layer homogeneity is the way to go, it’s simple:
- VSAN = SDS Software only.
- VSAN Ready Node = SDS Software + Bundled Hardware.
- VxRail = Integrated Appliance, and VxRack System with SDDC Nodes = Turnkey Engineered System.
This is the “vSphere-only universe” and it’s a great universe.
For Virtual Geek readers that are more focused on a heterogeneous universe at that abstraction layer (inclusive of, but not limited to vSphere) here’s the similar simple list:
- ScaleIO = SDS Software only.
- VxRack Flex Node = SDS Software + Bundled Hardware
- VxRack System with Flex Nodes = Turnkey Engineered System
This is the “Heterogeneous universe” and it’s a great universe.
ScaleIO 2.0 is a huge release – please join me in congratulating the engineering team, and the best way to do that? Download the bits! Try! Share!
