Just as it looks like the hardware side of cloud infrastructures is close to becoming a 'done deal' built on commodity Intel x86-based servers, there are interesting signs that a new server `war’ might be on the verge of starting. Not only that, further out a new and radically different approach – probability computing – could fundamentally change the way we utilise servers and develop cloud architectures.
In the here and now there is already a groundswell of both interest and development going on around alternative processors. There are already growing expectations for the ARM processor, commonly found at the heart of mobile phones and low power client systems.
The company is said to be now very close to introducing an updated processor design, codenamed Eagle, which will support virtualisation and large physical addresses. Both of there are very important to the server marketplace. It is said that all the vendors of virtualisation hypervisors, including market leader, VMware, are already well on the way to creating new versions of their hypervisors to support the new processor.
The target market is low-end servers, which in practice are the ideal type of system for the role of commodity cloud server. The trend now is towards building hybrid processors and architectures with a few, high performance, high complexity specialist devices, supported by a large number of low power, low complexity commodity devices. This can be seen in plans for future multicore processors as well as in complete datacentres.
Low power consumption has always been a strong suit of the ARM processor family, and that will certainly prove attractive to datacentre service providers and businesses with large on-premise infrastructures. It is also why ARM has been so successful in the mobile marketplace.
That itself adds an important factor into the mix because if the same processor architecture can be utilised in both servers and end user clients that level of commonality will only enhance the delivery and reliability of service provision. And with the coming of the next generations of Smartphones – Blackberry maker, RIM, has a new tablet system called BlackPad now in production in Taiwan and expected to be launched in November – having a common, coherent end-to-end platform for delivering and consuming cloud services will make a great deal of sense. There is a lot to play for here.
That is, of course, why Intel is buying security tools maker, McAfee. Embedding the same anti-virus and security-oriented technologies into both Xeon server processors and Atom mobile client processors is likely to create a good deal of interest, if only because it both ticks an important service management `box’ for both service providers and users, and makes that important problem one that is then owned by Intel rather than the service providers.
McAfee also has console-based security management tools like ePolicy Orchestrator that would obviously be directly integrated with as much of the cloud infrastructure as Intel processors could find a role in.
There is, of course, a side issue to the Intel/McAfee story, and that concerns the likely response of alternative x86 processor vendor, AMD. So far there has been no hint of any action by the company and with Intel snaffling the security market Number Two, AMD would either have to take on Symantec – probably too big a fish to swallow – or one of the smaller, niche security tools vendors.
Though a side issue it is an important one, for in a commodity market there is always a strong need for a second source of supply. So Intel actually has a vested interest in seeing AMD offer something similar, otherwise server vendors are likely to see the trap doors of a supplier lock-in looming up on them.
That could in fact make the ARM business model look significantly more attractive. The company’s forte is processor design, and it licences other vendors to make them or incorporate the designs in their own offerings. That means there are a number of alternative sources, and a number of alternative derivatives of the design from which vendors can choose.
An important component in this will also be the applications and utilities that form the basis of most of the services being delivered. It is easy to assume that ARM-based servers would trip up over the need to run existing server-based software, most of which is written for Intel’s x86 architecture and/or the RISC architectures such as IBM’s Power and Sun’s Sparc families.
This may well be true, but also then begs the question as to whether those traditional applications will, as user expectations of cloud services rise, be able to meet the need. There is every chance that new approaches to software will develop, many of which will be platform agnostic and abstracted from directly interfacing with the technology. Here, for example, service modelling based on developments in Java are a likely alternative approach.
By far the most interesting new approach, though still at a stage where it could easily crash and burn, is the development of Probability Computing. The company developing this approach is Lyric Semiconductor and is based on the PhD work of CEO, Ben Vigoda when he was at MIT.
Those with some knowledge of analogue computing will probably identify most readily with the Probability Computing model, for instead of using the digital computing approach of working with `0’s and `1’s Lyric’s GP5 processor design can work with all points between `0’ and `1’. The Lyric design is based on the use of `probability bits’ (pbits) that represent the intermediate states between `0’ and `1’.
GP5 stands for General Purpose Programmable Probability Processing Platform, and a production device is still some years away. The first samples are not expected till 2013.
The potential importance of the probability design is that it is likely to be far better equipped to meet the complex tasks and problems that cloud infrastructures will allow, and indeed encourage, users to work with. While digital technology is well suited for the tightly defined applications of today, such as databases where a specific answer to a specific question can be extracted from a known and defined dataset, its capabilities at effectively managing the multi-faceted and unclear tasks that service-based processes will generate, are becoming suspect.
Here, the task is to identify the most likely result of a complex interaction between multiple and unspecified datasets, and while that may sound like a rare and obscure requirement it does include tasks such as web searching. It also encompasses increasingly mainstream business processes such as financial modelling and spam filtering, where the ability to find the intersection of data can be performed on multiple variables in parallel.
Its potential at speech recognition could be the basis of a whole interface model within cloud architectures.
The company’s expectation is that the technology could increase computational capabilities by two or three orders of magnitude compared to today’s server processors. It also maps well on to the potential of parallel computing in cloud architectures using multicore processors. One expectation the company has is that GP5 devices will be mixed with one or more conventional digital processors.
This actually maps well on to the expected approach to parallel adoption in the cloud, where hybrid systems architectures and manycore processors are expected to appear. And these will exploit the low-cost, low-power capabilities of the ARM and Atom processors to provide the majority of the general purpose processing resource, coupled to specialist processor engines, which could now certainly include the Lyric GP5.
It will, of course, require a new programming language, which Lyric has developed. Known as Probability Sythnesizes to Bayesian Logic (PSBL), this uses a rules-based approach built around the specification of the constraints of a problem, rather than trying to define how to solve it. Version 1.0 should be available by the end of this year, and while it will only run slowly on simulators on digital computers, it is hoped that it will start the development of a GP5 ecosystem by the time the hardware is available.
Post new Comment