Data Center Strategies

SSD: Dump the hard disk form factor

Last time I blogged about SSDs the focus was on performance and capacity. Both characteristics are absolutely vital to the attractiveness of SSDs in enterprise class disk arrays. No doubt SSDs are all about performance, but capacity can’t be ignored either. We’ve seen nice strides in SSD capacity with introductions from Intel, Seagate and STEC. Performance also continues to rise now that we are over the initial seismic improvement as compared to our beloved but hopelessly slow hard disks.

All good work on the incremental evolutionary front.

[Angry reader: “INCREMENTAL! WHAT DO YOU MEAN INCREMENTAL? I THOUGHT YOU SAID SSDs WERE AN EVOLUTIONARY LEAP?]

[Calm and detached analyst: “that was way back in 2008, that’s ancient history. remember we live in internet time…]

In September 2008 I commented:

“…take the Humanoids running around an ancient continent. They made slow and deliberate progress using that opposable thumb to hit each other with sticks (remember the opening scene from 2001: A Space Odyssey?), and then one day there’s an evolutionary leap, and we have fire. Bam. New era …..”

Time to make an evolutionary leap: Dump the hard disk form factor.

The gene’s (no pun intended) are in place, and indeed some vendors have explored new possibilities. There’s Fusion-IO with their PCIe based storage adapter, TMS with their similar products and SUN pushing a memory stick form factor as well as their own PCIe adapter.

Why make SSDs look like a hard disk drive anyway?

Well for one, there’s a lot of legacy hardware designs in the world. From servers to mini laptops, physical space, connector and power has long ago been allocated for the hard disk. Likewise, in a midrange or enterprise disk array, disk enclosures are where you put storage devices, right? Isn’t that how the storage chromosomes get built? And don’t we need that to bridge from the high capacity hard disk to the high performance SSD?

Well, yes, at least for a little while.

But consider: If designers removed all the encapsulating hardware and protocols that surround legacy hard drives, disk array systems would get much simpler and cheaper. BAM. I see a new, yet evolved storage system emerging. A system using FLASH memory with no definitive form - and more interestingly - hidden inside the body of a disk array.

Losing the hard disk drive form factor sheds the mechanics necessary for mounting a disk, the wasted space around the mounting hardware, connectors, power supplies, fans, …you get the picture.

With all that extraneous hardware gone, just lay the FLASH chips down on a motherboard with or near the disk array controllers inside a common enclosure.

Some progress has been made on this idea, there’s the  Sun Storage F5100 Flash Array,  TMS’s  RamSan products and Violin’s 1010. All interesting products but not quite what I’m hoping for:

Image integrating FLASH memory into the controller design of a enterprise capable disk array. Resident with the controller hardware and connected onto the back side bus of the controllers via PCIe (or InfiniBand). Forget using SAS or fibre channel to the FLASH memory; part of the game is to drive down latency and that’s where PCIe or infiniBand can help.

NetApp and BlueArc have taken a stab at this approach by partnering with TMS, but the FLASH SSD’s enclosures are not fully integrated into the disk array frame. NetApp also offers a FLASH based cache card which gets close to the concept I’m promoting  but lacks disk heuristics.

IBM get’s closer with its inclusion of SSDs (derived from the quicksilver demonstration product) in its SVC storage virtualization appliance. And Fusion-IO products keep showing up as add in adapters for enterprise servers.

So vendors are making progress but customers need an evolutionary leap from disk array vendors to get them to the next level. That next leap in the evolutionary tree promises higher performance, simplified operation and lower costs. All desirable traits in any species of disk array.

When that day arrives, FLASH memory loses the solid state “disk” tag and just becomes solid state “memory”.

Comments?

posted by Gene Ruth

Just what is an enterprise disk array?

Do you know one when you see it? Is there a reasonable way to define it? Or should we just go with our gut when assigning the label of “enterprise” to a disk array?

Its pretty clear what is not an enterprise disk array. Take the Ctera C200 network attached storage device. It does RAID. It does cloud storage, it can migrate shares and it even does snapshots. Pretty cool. But it’s not a block device (or is that a hidden feature?), it only holds two disk drives and you certainly can’t get someone to wait in the driveway to fix the thing should it die.

So, clearly, features matter, bigness matters and a live-in support team has some value for those needing continuous operations.

But is that enough to hang the “enterprise” sign on a disk storage array?

After all, many mid-range disk arrays are loaded with features. Take a look at Promise’s VessRAID products. Lots of features. Or HP’s MSA disk arrays. Again lots of features but only the most shameless marketing person would call those products enterprise ready.

But let’s digress a little, maybe one of the first things to do is define what “enterprise” means. Certainly it implies “big” and “complex”. Or at least one might jump to that conclusion. In fact an enterprise could be a Mom&Pop bicycle shop. Most folks, however, are more likely (at least in the IT business) to think of a large international company when throwing around the word “enterprise”. So for this blog’s purposes, let’s assume that “enterprise” implies a company size that needs one or more dedicated data centers to service its IT needs. A datacenter with many hundreds of Terabytes and thousands of users.

So with that definition for enterprise, we are clearly discussing disk arrays that can service the largest of IT environments. For international companies that means global capabilities, with IT infrastructure to go with it.

Back to features. A globally distributed IT infrastructure requires a way to sync data. So at a minimum, an enterprise disk array (EDA) needs to efficiently move data across the globe. Replication is the answer and efficient usage of bandwidth goes hand in hand to keep costs down. So the feature set grows.

Likewise, as EDA’s multiply across data centers, management becomes a challenge. An operator will become quickly frustrated if the management paradigm for an EDA doesn't aggregate EDA’s and make them work in unison. For example, setting up replication across the globe can be akin to hanging wall paper - lots of moving parts and high criticism when perfection is not achieved. The EDA better have a decent management scheme.

And so the story goes ad nauseum.

Issues  such as security, efficiency, operating system support, environmentals, and especially scalability all factor into whether a disk array is an “enterprise” disk array. There is no one-size-fits-all criterion but there are clear measures that can be applied.

Some of this is in the eye of the beholder but in general it boils down to this: more scalability, more specialized features, more reliability, more of everything, characterizes an enterprise ready disk array.

But let’s get specific. To be in the enterprise game, to be at the top of the disk array food chain, a disk array must at least:

· Scale to greater than 500 TByte: In a large datacenter, an EDA must meet current needs as well as expand easily along with the growth of data. What’s magic about 500 TBytes? Nothing specific other than it sets the bar for size and scalability.

· Offer extreme availability: High availability is a cornerstone of many data centers that provide continuous operating environments. The disk array, through redundant design, maintenance procedures, and proactive self-healing should be capable of offering the best available uptime (up to 5 nines) whether as a single unit or as part of a larger disaster recovery and business continuity implementation.

· Integrate with the IT ecosystem: An EDA is not a standalone product. An EDA should support business process implementations by integrating into enterprise management applications and IT operational processes as a key component in an IT infrastructure.

· Provide holistic service and support: The supplying vendor must offer not only maintenance contracts to keep a system running smoothly on a global scale, but also options for onsite operations, training and 24/7 service.

· Offer multipurpose operation: The EDA cannot be a one trick pony. The system must be capable of being the focus of all data services including data protection and archiving within an integrated multi-datacenter ecosystem. An EDA vendor must deliver the highest performance capability for demanding databases and transactional applications as well as less demanding but still vital file serving or server-virtualization data capability.

That’s a start, the above gets a vendor into the game, but we still need some criteria to compare enterprise disk arrays. What would that be? Is a spec sheet comparison interesting or is there a more insightful way to compare products?

Whose to say an EMC VMax is better than a IBM DS 8300? 

IT organizations are. With their purchase orders.

And their Burton Group analyst friends stand ready to help them with research into this very subject.

And back to the Ctera C200. Cool product but Ctera: do you think you could juice it up to compete against a HDS USP? Just kidding ;-}

Comments?

Posted by Gene Ruth

Application delivery and the evolution of endpoints

Anyone who hasn't been hiding under a rock for the last few weeks won't have been able to avoid the brouhaha concerning the highly anticipated launch of Apple's tablet. Not quite as exciting to some, but certainly more real was Steve Ballmer's keynote at CES last week, where he showed off Windows based tablets from HP, Archos and Pegatron. At the same time there has been an explosion of announcements on Android tablets with manufacturers such as Archos (again), Compal, Dell, HP (again), and Motorola, offering tablets ranging in size from 5 to 10 inches. While in the UK, T-Mobile has announced its own monster 15 inch Vega tablet that will also run Android. If this wasn't enough Intel is pushing Moblin their own Linux-based OS at Netbook manufacturers and have just launched the beta of AppUp Center Intel's take on the single platform application storefront. And let's not forget Walt Mossberg’s interview with Google's VP of Engineering Andy Rubin strongly hinting that Google has RIM in its sights by announcing plans for an enterprise version of the Nexus One smartphone complete with physical keyboard and longer battery life.

A couple of years ago enterprise IT might have seen all this as an interesting diversion and little more. But now with the iPhone, and more importantly the iTunes App Store, expectations on application delivery to the endpoint have changed. Enterprise users now expect to be able to access both personal and business apps where ever they are.  So while many of these new devices are clearly aimed at the consumer market, it would be shortsighted not to expect some of them to see service as enterprise application delivery endpoints in the near future. 

This represents a major challenge to enterprise IT. Not only is there a rapidly growing array of devices and operating systems to consider, but with the growth of single platform application storefronts like iTunes and AppUp Center,corporate IT has lost control of the delivery channel. If enterprise IT organizations are to stand any chance of being ready to react to this change they will have to act quickly to consider how they meet growing business expectations and deliver applications across such a broad range of platforms.

We are already seeing indications that Citrix and VMware are positioning themselves to be able to offer solutions to this question, but with each organization’s technology background leading them in radically different directions.

VMware's approach, somewhat predictably for a company specializing in virtualization, is to look towards abstracting mobile phone operating systems from the underlying hardware platform by developing a mobile hypervisor platform. To this end VMware have already successfully demonstrated a smartphone dual booting Android and Windows Mobile, with the ultimate goal being to be able to run multiple Smartphone operating systems concurrently on the same device. While VMware's stated goal for this technology is to provide a means of isolating personal and business applications on the same Smartphone hardware, it doesn't take much imagination to see that the same technology would enable application developers to write exclusively for whichever preferred Smartphone platform, leaving the hard work of delivering the application on other hardware platforms to VMware and their mobile hypervisor. Assuming that VMware is successful in getting mobile carriers to pressure Smartphone manufacturers to adopt this approach, there is still one significant issue that is unlikely to be resolved.  It's hard to imagine Apple showing any enthusiasm for any proposal that would relax it's current level of control over the iPhone.  So while a mobile hypervisor could reduce the number of platforms that a developer would have to support, it would not enable compatibility across every platform.

Perhaps equally predictably, Citrix is looking to exploit server hosted application virtualization (AKA XenApp) as a means of allowing developers to address endpoint diversity. The message that Citrix is promoting appears to be ' There is no need to rewrite your applications from the ground up for every smart phone or tablet operating system and form factor. Instead why not just adjust the application's individual window panes to match the size screen at the endpoint and leave the rest to us’.  Backing this message up Citrix has incorporated new swipe gestures into v 2.0 of the Citrix Receiver for iPhone (their iPhone client)  to provide a better mobile experience while navigating between window panes. Citrix is calling this category of apps that can be delivered to a desktop, tablet or smartphone, "Mobile-Friendly" applications

As a side note, it looks like the open source Xen community is following the same path as VMware with their Xen ARM project which is aimed at delivering an open source hypervisor capable of running on smartphone platforms. Citrix has also hedged its bets in the mobile space by investing in OK Labs an established embedded hypervisor vendor already deployed in many mobile devices.

Right now it looks as though the short term advantage lies with Mobile-Friendly applications. Citrix has had a client for Windows Mobile devices for a number of years now, and has recently shipped v 2.0 of their iPhone client and released a tech preview of their Android client, while VMware is not expected to ship their smartphone hypervisor until sometime in 2012. At the same time, if an ISV follows the Citrix path to deliver the application on a small form factor non-Windows endpoint all that they need to do is to rework the GUI to fit the display size and present the application on XenApp or XenDesktop (I already know of two heathcare ISVs that are giving this proposal serious consideration). Conversely, if they wish to follow the path that VMware appears to be promoting their applications will require a comprehensive rewrite. 

In the long-term however, the situation may not be as clear. VMware's approach, if successful, would offer the advantage of disconnected operation, although it is difficult to assess how much of an advantage this would be given that almost every application requires some level of connectivity today, and as time moves on we should expect ever more ubiquitous connectivity. There's one other possibility that may overtake both these options and that's HTML 5.  This might offer the best long term possibility for a single common development platform for mobile applications and would also allow some degree of disconnected operation without the cost or complexity of either an application hosting or virtualization based solution. It is too early to tell which solution (if any) will win out, but I expect to see significant development in this space over the coming year. We should also expect to see Microsoft take action to protect its market share by coming in on the side of Citrix if it looks Windows Mobile is heading towards failure. 

Gartner Acquires Burton Group

By now everyone has heard the news.  So here's our take on this:  It's very positive.  Here's the top reasons
-Gartner is going to retain the whole Burton Group analyst and consulting teams.
-Gartner is going to maintain the Burton Group brand and IT1 product.
-Burton Group's product gets a world wide sales force behind it

In other words, this is not another Meta Group acquisition. 

So some of you not familiar with the details of the two products may be wondering "what's the difference between the products?"  Actually, there is not much overlap at all.  We sold to two different markets.  Gartner focuses on the executive level - the strategic question of "what to do."  Burton Group focuses on the IT leaders and implementers - the question of "how to do."  So both products compliment each other and focus on different audiences within the IT organization.

The question of vendor independence has also come up.  Gartner has a strict policy of no vendor sponsored white papers, just like Burton Group does.  While the perception may be that Gartner is swayed by vendors, their policies all force vendor independence.

So this is an opportunity for Burton Group research to grow its reach into the market.

Comments are welcome.
[Posted by Drue Reeves and Richard Jones]