By: Gabriele Svelto (gabriele.svelto.delete@this.gmail.com), January 31, 2013 5:05 am
Room: Moderated Discussions
I am under the impression that the biggest hurdle that ARM servers have to overcome is their inherent small size (at least for their first generation). This makes them in direct competition with today's virtualization solutions. It's possible to carve a lot of small VMs out of a regular dual-socket x86 box and for certain workloads (e.g. identical OS images) things like Kernel Shared Memory greatly lower the memory requirements too.
If I'd be pricing ARM microservers I'd pit them up against VMs, not against regular x86 servers as it's likely that the workloads they'll run will better much swarm-of-small-VMs deployments rather than fully dedicated servers. The downside of it is that many virtualization solutions are quite expensive software-wise which somewhat lowers their appeal; fully FOSS alternatives are often every bit as good but they require more effort and a talented IT team (which doesn't come cheap either).
We'll see ARM servers carved into VMs as soon as their main memory becomes big enough but without the weight of the major commercial players behind I don't see it taking off very soon. However every downside that applies to x86 virtualization applies even more to ARM where the software ecosystem is still young and commercial solutions practically non-existent. In addition to this I'd expect the IT shops more invested in commercial solutions to be the most conservatives ones and thus less likely to switch to a completely new architecture.
The other aspect I started thinking about is how ARM vendors are trying to leverage custom, integrated interconnects as a selling point against x86. Since these usually include everything and the kitchen sink (storage, network, ILOM management, etc...) they can be a significant differentiator in that they do provide both increased performance and simpler manageability. However those won't be limited to ARM servers - in fact the current solutions are x86-based (AMD/SeaMicro) - and I wouldn't be surprised if Intel would move into that area soon enough if an interesting market for those machines emerges.
If I'd be pricing ARM microservers I'd pit them up against VMs, not against regular x86 servers as it's likely that the workloads they'll run will better much swarm-of-small-VMs deployments rather than fully dedicated servers. The downside of it is that many virtualization solutions are quite expensive software-wise which somewhat lowers their appeal; fully FOSS alternatives are often every bit as good but they require more effort and a talented IT team (which doesn't come cheap either).
We'll see ARM servers carved into VMs as soon as their main memory becomes big enough but without the weight of the major commercial players behind I don't see it taking off very soon. However every downside that applies to x86 virtualization applies even more to ARM where the software ecosystem is still young and commercial solutions practically non-existent. In addition to this I'd expect the IT shops more invested in commercial solutions to be the most conservatives ones and thus less likely to switch to a completely new architecture.
The other aspect I started thinking about is how ARM vendors are trying to leverage custom, integrated interconnects as a selling point against x86. Since these usually include everything and the kitchen sink (storage, network, ILOM management, etc...) they can be a significant differentiator in that they do provide both increased performance and simpler manageability. However those won't be limited to ARM servers - in fact the current solutions are x86-based (AMD/SeaMicro) - and I wouldn't be surprised if Intel would move into that area soon enough if an interesting market for those machines emerges.
