Over the last year or so, there’s been a tremendous amount of hype surrounding blade servers with numerous companies suggesting that blade servers are the key to the future.
For its part, analyst community expects significant growth in blade server deployment: IDC, for example, predicts that that there will be a compound annual growth rate of 40.4% in x86-based machines devices. This would mean that such systems would account for 8.9% of the total server market in 2005, rising to 25.4% by 2009. IDC sees blades gaining popularity at the expense of rack-mounted systems to such an extent that sales will jump from an expected $539.6 million in Western Europe for 2005 to $1.74 billion four years later.
In the midst of all this optimism there are certain trends that need to be looked at. One striking things is that the vast majority of sales of blade servers will inevitably be to large enterprises and not small to medium sized businesses (SMBs).
Gary Barnett, research director at Ovum, explains why this may happen: “At this point in time, the greatest focus is on the big company Intel consolidation market [as] for understandable reasons this is the largest medium-term opportunity. If you’ve got hundreds of Intel servers across your estate with locally attached storage and you’re spending millions of pounds trying to administer them, blade servers can offer you huge cost savings.”
When assessing just how such savings can be made, you really just need to think of the underlying technology attached to blades which are individual custom-built motherboards, each of which includes one or more processors, memory, storage and network connections. These blades slot into a chassis that provides a common power supply and air-cooling system and they are usually hot-swappable.
Given such a configuration, you can place a high number of blades into a server rack chassis to create systems that are not only more compact and powerful than traditional equivalent rack-mounted servers, but also much cheaper and easier to manage.
Stewart Hayward, commercial director at reseller WStore, explains: “Because all of the servers are in one chassis, you reduce the cost of running and managing the network. You can also reduce your cabling requirements and consolidate the amount of power you use. And you get a single console to manage everything rather than having to deal with different system images. It’s all silly things like that that make a headache for IT managers.”
Savings can be leveraged throughout the lifetime of the blades. Once a chassis has been purchased, it becomes cheaper to buy and install subsequent blades than to acquire and deploy a whole new server from scratch each time.
However, you should be aware that blade servers are suitable for running every type of application. Indeed, even though they come into their own with commodity or modular packages and particularly ones that undertake lots of number-crunching such as Microsoft’s Exchange messaging applications, what they are not suitable for is large, heavily transaction-based monolithic applications. Data processing on, for example a SAP R/3 would be an example.
Clive Longbottom, a service director at Quocirca, gives his opinion to this debate. “There’s been a lot of hype where people have said ‘blades are the answer, but what’s the question?’ So customers have started at the commodity end to see how they work”.
In effect, blades have tended to be used less for big mission-critical systems and more to run peripheral functionality and new services comprised of compound applications.
Yet, one of the key selling points of blades is the relatively small amount of room that they take up – an often overlooked factor that can save enterprises a lot of money in terms of renting expensive office space for data centres.
As a result, you should be aware that the key vendor messaging to you for blades has related to system consolidation and rationalisation.
“[The emphasis] has been about saying you can put your messaging system in an environment that takes up little space. It’ll be highly scalable, but you can just put it in a corner. The same is true of areas such as portals, which have little intrinsic value, but are central to how people work. So you can get the system solid and then forget about it and focus on areas that you’re worried about such as CRM,” Longbottom says.
Unsurprisingly, therefore, blade systems have so far proved most popular with financial services and retail organisations. But according to Hayward, they are also being taken up in a big way in data centres or “anywhere they can synthesise a lot of processing power and organisations need the power of centralised applications”. This includes companies with large e-commerce sites, hosting firms and application service providers.
Small- and medium–sized firms have, to date, not shown the same level of interest in blades. This may be down to a sense of how appropriate hey are but also it could be how they are sold. Explains Stewart: “Normally we wouldn’t attempt to sell to a company with less than 250 users and with a minimum of 15 to 20 servers to consolidate, at which point it’s cost effective to transition.”
Greg Carlow, managing director at reseller Repton, says blade systems tend to be deployed in two key ways: either as a consolidation tool to tackle “server sprawl” or, as what is part of “a smaller but faster growing trend”, to replace UNIX boxes with Linux when the time comes to undertake a large operating system upgrade.
But one of the key procurement issues lower down the food chain, Hayward believes, is that blade servers are currently “an all or nothing proposition” in terms of replacing existing infrastructure.
“Most [companies] are fairly happy with their IT infrastructure so if you say ‘why don’t you replace your 10 regular servers with blades?’, they don’t see why they should junk what they’ve already got. They’ll say ‘we plan to replace six servers in three year’s time and four in two years, but why throw it all away now because it works?’” Hayward explains.
But another important issue for you to consider relates to the initial upfront cost of such systems, which can be as much as 20% higher than regular servers. “If we could find a gradual way to introduce blades, it would work, but you’re saying with your first purchase, you need to buy a chassis, blades, additional power units, switches and the like. Once the initial purchase is there, it becomes cheaper each time, but the initial acquisition costs put customers off.
“We need longer-term messages around ease of management because [customers] won’t make the initial purchase unless they’re educated to see it as a longer-term investment. But it’s difficult because it’s a total cost of ownership argument, which [smaller firms in particular] are notorious for not listening to,” Hayward explains.
Nonetheless, while the time may not be right just now, Barnett does see “a huge potential opportunity” for blades among all firms over the next couple of years, although he too recognises that this requires the blade vendors “to think differently and come up with a completely different proposition”.
For big companies, part of the appeal of blade technology, he believes, is that it makes it easier for them to move to more of a utility model of computing “using virtualisation to manage blades as if they’re one virtual computer so they can do some pretty sophisticated workload management”.
“But most SMBs don’t have that type of problem. Depending on size, they’ll have a limited budget and very limited space in terms of a computer room. So their telecoms and networking equipment will be in a cupboard and one of their two servers will sit under the finance directors desk with [someone] in accounts being relied on to change the storage tapes that will be left on top of the server,” Barnett says.
This means that there is potentially huge pent-up demand for a “wine cooler with the glass front ripped out” to house all of these different functions in one place, with a different blade catering to each of them and administered via a single easy-to-use management console.
“It’s about collapsing infrastructure into as small a place as possible and making it easier to manage and run,” Barnett points out.
Another downside to the blade proposition is that at the moment, blades from one vendor do not work well in the chassis of another. There is also currently a dearth of blades that can perform specific functions such as storage or networking or tackle specific workloads.
Having access to specialised blades would enable you to create highly optimised units targeted at specific functions. This indeed is a work in progress on the part of many of the vendors.
As for the interoperability issue, some suppliers have already made various ultimately self-serving moves in this direction. Hewlett-Packard, for example, launched its OpenBlade initiative in an attempt to drive interoperable, multi-vendor blade server standards in February 2002. Because these amounted to a renaming of the CompactPCI bus, however, most systems vendors refused to sign up to the scheme as they used different technology and it has since become a programme for non-rival partners.
This has led IBM, in turn, to try and grasp the nettle. In September, the vendor got together with Intel to release most of its BladeCenter specifications - except for those relating to the core processor blades and the chassis - in a bid to encourage third party vendors to make the technology a de facto industry standard by developing offerings based around it. It has also announced but so far failed to launch its blade.org scheme, which is intended to build a more formalised community around the concept.
But one of the most important issues relates to cooling and the problems brought about by cramming a lot of systems into a small space.
Carlow explains: “Blades are massively power-hungry right now and this can cause major problems. One banking customer had to stop a deployment because it couldn’t get the heat out of the room and it had less than half the blades running that it hoped to.”
This means that it is quite common to see chassis “with a lot of blades at the top and bottom and progressively less in the middle because they haven’t got an adequate air cooling system”.
Traditional systems, he says, are cooled from bottom to top, while blades are cooled from front to back, which means that most air conditioning units are not set up properly to cope with them. This is starting to be tackled by improvements to fan design, however, and will become less of a challenge as dual core processors become more widespread.
But warns Barnett, even once these creases have been ironed out, the prospects for blades are not totally dependent on technology: “The key is that, with systems like this, the channel should not be selling technology but specific outcomes such as infrastructure simplification or lower management costs,” he says.