These are part of an interesting (IMO) trend toward using less traditional system architectures, in this case using much less powerful ARM chips (an ARM does a lot less per clock than an equivalent x86_64 CPU) in large numbers to provide basic (i.e. low per-page computation) web services, to more effectively fill a target niche.
Website serving (in most cases) is exactly the type of low-CPU high-I/O task that's perfectly suited to very large numbers of weak CPU cores. Each request can easily run on a different core, but pushing a page (even one generated from *simple* JSP/ASP/PHP) takes very little CPU. Running a lot more cores helps, but running expensive fast cores does not (this is mitigated somewhat by event-driven servers like NginX, but more cores is still better). You will need a ton of network and disk I/O, but things like tcp-splice, tcp-offload, a copy-free TCP stack, memcached, membase, FusionIO cards, etc... are making that a very reasonable proposition for a Linux system like these.
As a side note, that's always been the advantage of Mainframes (which are still used for a LOT of stuff) for similar types of line-of-business tasks (like transaction processing); they don't have a lot of raw CPU, but they can push bits at rates an x86 couldn't dream of matching.
The advantage of these ARM based systems over a distributed or mainframe-type system is that they're a few orders of magnitude less expensive, ideally suited to web serving, and use vastly less space, power, and cooling than a commodity system with the same page-per-second capacity.
I would never even think of using one for something like a database server, but as the web front-end they're pretty close to ideal.
GPU computing (e.g. CUDA or OpenCL) is another example of alternative, more targeted, architecture; this time for extremely parallel high-CPU/low-I/O tasks (the opposite of what these ARM boxes are suited to).
I predict that we will see a continuing diversification of system types as different types of tasks and new computing techniques drive a need for a wider variety of non-traditional architectures. I read an interesting article on the IEEE website today about using memristors as a similarly novel computing architecture to accomplish tasks currently solved by neural processing algorithms because a memristor array acts, in some ways, like a neural array.
I should note here that these ARM systems won't replace x86 systems, just as x86 hasn't replaced mainframes. These new architectures will expand the diversity of systems in use; existing systems will still be used where they make sense, and these new systems will take over in niches where they significantly outperform existing approaches.
Stephen wrote:
> http://www.theregister.co.uk/2010/11/23/zt_systems_arm_server/
>
> so what is plugs thought on this. Aside from pre-shipping with Linux
> (ubuntu) i can see how low power server nodes for those itty bitty
> nibble tasks can make sense, but they are looking to push a ton of
> little cores soon.
>
---------------------------------------------------
PLUG-discuss mailing list -
PLUG-discuss@lists.plug.phoenix.az.us
To subscribe, unsubscribe, or to change your mail settings:
http://lists.PLUG.phoenix.az.us/mailman/listinfo/plug-discuss
(text/plain)
