Storagebod's Blog

On the Open Road with FCoE

Intel's announcement of Open FCoE running on their 10 GbE products is yet another manifestation of the 'Infrastructure as Software' trope;  Open FCoE has been officially part of the Linux kernel since November and Intel's announcement brings Windows Server into the fold. 

This announcement means that you can run iSCSI, NAS and FCoE on a single adapter; big deal I hear you cry especially if you work for one of the traditional HBA vendors; you can already do this using a CNA but Intel's announcement is important because Intel is currently the top-selling vendor of 10 GbE cards and this will allow you to use those cards which may already be shipping in your standard server build to run FCoE if you so wish.

Obviously doing FCoE in software does come at cost and that is increased CPU utilisation but this has not really hampered the iSCSI software stacks. And like iSCSI, I suspect that we will find the predominant deployments of FCoE will be utilising a software implementation and I expect that other 10 GbE adapters will come to be supported by the Open FCoE stack.

One place where I will take issue with many of the commentators is this feeling that we will not have a separate Storage/Network infrastructure; it is extremely likely that we will continue to have some kind of separation between Storage and Network infrastructures; this might be physical or it might be logical. But I do expect that in many cases that we will continue to have 'storage ports', 'backup ports' and 'network ports'; this separation will be due to both the political realities of most Infrastructure teams but also simply because it makes sense. You don't necessarily want your back-up traffic or storage traffic stomping all over your transactional traffic or even each other; yes, QoS can help but it just might be simpler to keep them separate.

Yes Ethernet has pretty much won and Open FCoE makes victory even more certain but don't expect storage networks to go away anytime soon and don't expect FC to go away any time soon; 10 GbE ports are still relatively expensive and I'm not seeing a massive appetite for rip and replace at present. The long march to converged network will be a long march indeed but tools like Open FCoE might shorten it slightly.

I guess the big question for many of us is when will VMware see a software implementation of FCoE? 

SRM - Stack Resource Management

2011 will see acquisitions continue to come as all of the major players try to position themselves as owners of the vertical stack; we will see acquisitions to enable the general compute stack but also acquisitions which enable specialised appliance based stacks. Obviously, we will also see the major players continue to announce home-grown stack products as well.

All of these stacks will blow up smog which might well look something akin to a Cloud; the question is whether this will be a toxic cloud or something more beneficent bringing the necessary components for business growth?

One of the biggest challenges will be that we will have different types of stacks; there will be the odd case where a data centre is homogeneous but it will be more usual for a data centre to have at least two vendors. This might be for either technical reasons or for purely commercial reasons; two vendors competing may well be more honest than a single dominant vendors. And of course, there is the ever present public Cloud option to be considered.

The question is going to be one of how to manage these stacks; both from a data centre point of view but also from a rack/server position. How do we manage both the data centre and from application to spindle? 

To this day we struggle to manage storage arrays from different vendors with a single pane of glass; we even struggle to get a common view let alone a common configuration tool. How much more complex will this become when we have several vertical stacks from a variety of vendors to manage?

Yes, we could struggle on and manage at the component level but how is this going to bring the reduced complexity and business benefit that these stacks could bring? We could even find ourselves using shell interfaces/APIs to produce our own consolidated tools; assuming that these shell interfaces/APIs are available. 

Storage Resource Management has been a big #FAIL in general; Stack Resource Management could have a similar future unless we see vendors and other interested parties begin to get a handle on this. I certainly expect to see acquisition in this space as vendors try to steal a march on each other but really, this might a place best served by co-opetition and standards.

Formal standards? Possibly not but two or three of the bigger players could do the market a huge service by beginning to think along the lines of an open standard. And if one emerges; they could do an even bigger service but not taking the approach of 'embrace and extend'.  

 

Cutting Edge

A couple of weeks ago; I was listening to Radio 4's Food Programme, a regular audio accompaniment to our Sunday lunch; they were contrasting the use of knives in the Western kitchen as opposed to the Chinese kitchen. 

In the average Chinese kitchen; there are very few knives with the most important of these being the Cleaver. To Western eyes the cleaver is probably a crude and brutish knife used mainly for cutting through bones and other tasks which require a hefty blade but in the hands of the skilled Chinese cook; it is pretty much the only knife needed. It can indeed be used in the way that a Western cook would use it but it can also be used for more refined tasks; indeed every part of the knife is used, from chopping, to sculpting to crushing. It is the only knife a Chinese cook really needs.

Compare that the Western kitchen, a knife for every purpose; you will find a Cleaver but you will find vegetable knives, carving knives, bread knives, cheese knives, filleting knives; a whole array of knives. To the Chinese cook's eyes this probably seems extremely profligate but Western cuisine has evolved in a very different way to that of Chinese due to different environmental requirements.

Neither way is right and neither way is wrong; arguably, you require more skill to wield a Cleaver to cover all the jobs in the Chinese kitchen but there is a counter-argument that you need greater knowledge to make the right choice in a Western kitchen. At the end of the day, the aims are the same and at least the interface is common. 

A pity we can't say the same for the storage industry really.

VNeXt - Converged Confusion

At the core of EMC's mega-announcements yesterday was the long awaited coming together of the Clariion and Celerra lines in the form of the VNX and VNXe. These are firmly targeted at NetApp's core market; dual controller, unified storage; probably still the fastest growing part of the storage market, well certainly within in the external shared storage market. 

The VNX can simply be seen a direct convergence of the Clariion and Celerra but the VNXe takes this convergence and pushes it down into the SMB market; VNXe could be competing with Drodo, Synology, Iomega (yes I know an EMC company) and a whole host of other smaller players. NetApp have also pushed into this space, so we can look for more heated and noisy competition in this space as well.

The VNX range (and I do include the VNXe) further progress the idea that Storage is Software; the heart of every storage appliance is software and the core differentiator is no longer hardware; with almost no exceptions within this market sector, the hardware is Intel-based and pretty much commodity.

EMC, like IBM have massively invested in simplifying their management interface and UniSphere is massive step forward in simplicity. Much-hated by experts, configuration wizards enable the new breed of IT generalists to set up their devices with out becoming experts in storage. Yes, you can still get access to the underlying configuration but hopefully in most cases, it will be good enough. 

I think the VNX is good enough for a great deal of the market; NetApp have ploughed this furrow very well, they often pitch their Filers as good enough and point out that you don't need expensive Symmetrix hardware to service most requirements.

And it is here that EMC have a problem, the VNX is good enough to replace a massive percentage of their Symmetrix footprint but within some territories, the account teams still lead with Symmetrix and come up with a lot of 'good reasons' why a customer should retain this footprint; it is only when the customer makes serious moves towards NetApp that the Clariion and Celerra come into play. This actually adversely impacts both the credibility of EMC but also their products.

The VNX is good enough to consign the Symmetrix range into a mainframe-like plateau and eventual decline; there will be still be workloads and environments (mainframe for one) which are served best by the Symmetrix but the so-called mid-range will be the future for many of us.

Now we have the VNX, I wonder if the next big thing for EMC will be how they federate the VNX range and enable clusters of VNX; federated VNX will yet further push it's market coverage.

And as EMC continue to evolve the VNX architecture; I think we can expect further performance improvements akin to the doubling of VMAX performance. There is probably a lot of scope for improvements as EMC work to converge the Celerra and Clariion code-bases but Unisphere should allow for much of this to be hidden from the customer.

Yet there is still a big question about VNX in general; is it actually the right product for today and more importantly tomorrow? Is it what we actually need and does it progress the storage landscape? It is certainly the right product for EMC, it builds on what they have and is better than what they have but with the Isilon purchase and the still cloudy future of Atmos; time will tell. 

Breaking Records?

There's much to admire in EMC's mega-announcement; there's lots to dispute as well but let's leave it a few days before we all piss on their parade! And I'm sure that there will be much of that to come!

Whilst watching the presentations and the record-breaking; I began to wonder how some of their competitors would go about record breaking!

HP would attempt to jump their whole storage portfolio laid end to end using a rocket bike! It would blow up when one of their OEMs withdrew a key part.

IBM would jump a tank of sharks obviously! However later most of the sharks would be revealed to be Basking Sharks and toothless!

NetApp claim to jump 41 arrays deduped into a single array but the other actual arrays would also be lined up to cover WAFL, aggregate reserve, snap reserve and rightsizing overheads.

HDS would jump a single VSP but claim that they had a huge number of arrays virtualised behind it.

Oracle would jump 41 yachts, who gives a crap about storage anyway!

Dell would jump 41 disks; HP bought the arrays that they wanted.

Microsoft would wave their arms about and make motorbike noises whilst jumping buses on their xbox.

 

 

Bod's Annoyances: Server Virtualisation

Server Virtualisation? Is 'Bod simply being obtuse; yes it is entirely possible but before you write this off as lunacy, I do have my reasons. 

Firstly, server virtualisation is not actually a bad thing; it's been around for many years but the landscape today is actually more than a little concerning and I wonder if it will come to bite us really rather nastily over the next decade. The sheer number of virtual servers being built leads me to wonder about the long term sustainability of the current infrastructures and with the current virtualistion==cloud lazy thinking; I am really rather worried.

So how did we come to such a position? The seeds for this were sown in the early eighties when IBM launched the personal computer; prior to this (and yes I am ignoring the early adventures in personal computing by Apple, Commodore, Radio-shack and the likes), computers were large, complex and extremely expensive beasts managed by the computing department; access was very restricted and not many people had access. But with the personal computing revolution; our relationship with computers changed, indeed we actually started to develop a relationship with computers. 

One person mapped to one PC; that's the very definition of personal computing but as these devices became more powerful, it became pretty obvious that they were capable of so much more but the big, multi-user, multi-tasking operating systems which ran on the large corporate mainframes were not available for these personal computers. So we ended up running larger and more complex applications on these devices but using the operating systems available on the desktop; these were single-user operating systems with limited multi-tasking capabilities. So an application per device paradigm became the norm and even when multi-tasking became available and more common; the paradigm had become firmly embedded. 

And the lack of multi-user capablities really hampered the development of a centralised management and secure environment; so it became even more embedded. Add into that, a growth in CPU power far outstripping our ability to use it for a single application; we ended up in the position that we had rampant physical growth but with ridiculously low utilisation figures.

So in order to use this power and to manage it effectively; we have looked to server virtualisation to fix this problem allowing us to build virtual servers which more effectively use the available CPU or more to the point allow us to run multiple single application servers on a single physical server. It also allows concurrent management of these applications to be more easily achieved. Even the rise of the Intel-based Unix systems has not done little to break the paradigm of one application per server but instead of a physical server, we now talk about virtual servers.

It doesn't have to be this way and we only have to look at the way that virtualisation is used in the mainframe environment; you don't talk about running 1000s of server instances, you are more clever about how you virtualise. You may virtualise an environment but it will still only run a few actual virtual machine instances.You may group related applications into a single machine instance but not have those applications running as separate server-instances.

We could do this in the Intel-space; we should move away from the default position of one application per server instance, we should move to the position where we understand why we do so. There are applications which certainly work better when run in massively distributed and parallel environments but this is not every application. A database server can run more than a single database instance; a web-server can run more than a single web-server.

Okay, it's not server virtualisation itself which annoys me but the unthinking use of it and the traps that we are busy building for ourselves.

Turning desktop operating systems into powerful server systems probably was not the wisest move that the IT industry made but the short-term gains were probably too attractive, can we hope that we don't do the same thing with mobile device operating system? Actually, I am not that hopeful really; history does unfortunately have a habit of repeating itself.

Bod's Annoyances: Enterprise License Agreements

Over the next few weeks, I thought might share some of my annoyances with life in Corporate IT; some of them are vendor driven and some are the result of wrong-headed thinking in the end-users! I'm sure everyone has their own annoyances and some of my readers could probably write several volumes on theirs.

So first up is 'Enterprise License Agreements' or what they mean and how end-users tend to look at them; this is often a case of the end-user walking into a nicely vendor-dug trap and smiling as they fall into it. 

For a single payment; you can consume as much of a vendor's product as you possibly can over a period of time; so what's not to like?

Well after the contracted period, you face a license true-up; so you've merrily gone on consuming as much as you can eat and now you have an elasticated waist-band on your servers and now you must pay for all those licenses. 

Of course that payment will probably take the form of another 'Enterprise License Agreement' and so it goes on but you've got so many installs of the software; you have effectively become an addict and there's no way out.

Time after time; I hear project managers tell me 'But it's free and it doesn't cost my project anything!'. You try to explain that it's not free and there might be a better fit for their problem but it all falls on deaf-ears. After all, the project manager gets it for free and there's no annoying budgetary considerations for them to take on.

And then there is the situation where an inordinate amount is spent on an ELA and yet at the end of the period when you do the calculation, your cost per license is no better than if you'd gone for a PAYG-type model.

ELAs, one of the industry's great cons or at least, one of Bod's annoyances!

 

Information Technology

2011 is the year of 'Big Data' or I so hear; organisations are collecting ever more data which may hold some value. The problem for many of these organisations is unlocking the value and turning this data into useable information. But I am not going to talk about how this is done; that is topic for another blog and perhaps for someone who has more expertise in that area; I am more interested on how we access the information and the tools that we use.

I wonder if we could see a change in how corporate IT is consumed; the explosion in mobile, non-traditional computing devices has led many to posit a future where much of IT is consumed in the form of apps; small specialised applications which do one or two things very well and this might very well be true but before these apps become truly useful to the corporate 'Knowledge Worker', there are other changes which need to happen. 

The IT department needs to enable the access to the information and at times to the raw 'Big Data' to allow these workers to move beyond the superficial; it is the area of curation and publication which could well be the growth area for IT departments. Building information stores with standard APIs to allow the publication of information which these apps can access; the IT department may not necessarily control the apps and presentation but they will control the access to the data. They may publish reference clients in much the same way that Twitter publishes a reference client but does not mandate its use. 

Technologies such as Object Storage for example will come to the fore as enabling technologies as will the already more established interfaces for exposing data and information but instead of trying to restrict the use of these to IT applications; more innovative uses will be encouraged. 

IT moves away from supporting individual devices but simply provides interfaces to Corporate Information and at that point becomes truly the 'Information Technology' department and not the 'PC Support department'. 

I think we are some years away from this happening and there will always be a place for traditional IT as it does support many essential functions but it's probably a more interesting future for many of us than the current status quo. 

 

 

Chuck D was right

So we are less than a week into 2011 and the Hype has already started; I was hoping that this was the year that we moved beyond superficial marketing and into something more substantial to reflect the market but instead EMC are mailing out broken records hyping some announcement that they intending to make later on this month. This is also accompanied with an on-line campaign with a whole host of EMCers singing from the corporate hymn-sheet having spent the festive season learning the words and consuming the kool-aid.

I'm sure that they will announce a new product which will surpass everyone else's current product; yet another manifestation of the commonly seen phenomena that 'my new product is better than your old product'; this will be inevitably followed in the next six months by someone announcing a product which surpasses the announcements made by EMC. The whole thing is reminiscent of the old DB2 vs Oracle or the Sun vs IBM vs HP benchmarketing wars of yore. Do we really want a year of boy-racer announcements from the storage vendors? 

I really hope that I am wrong and that EMC's focus of their announcement is not going to be purely performance based and how fast it is? Or even worse how it is the fastest array when combined with VMware and UCS but only in perfect weather conditions and being driven at a steady 60 miles per hour? 

I actually think that most of the storage vendors have some interesting things to talk about and there are some big themes to be tackled but it's less than a week into the New Year and I'm already irritated! And if this announcement does merely turn out to being all about how fast EMC arrays are, I hope that NetApp et al show some maturity and do not respond with similar hyperbolic and meaningless announcements.

Until then, can I recommend track 3 on Public Enemy's 'It Takes a Nation of Millions to Hold Us Back'? A record worth holding and keeping.