So today I stumbled accross this: IBM’s new eX5 architechture/server range.
The main advantage is they seem to have decoupled memory from the Xeon X86 processors, allowing for extra rackmount modules of RAM without needing extra processors or more servers.
It’s all well and good and i’d like to be able to say, this this the birth of MAN’s (Memory Area Networks) to accompany existing SAN and LAN technology, however; I can’t answer that, as there does not seem to be much technical information available!
For example, is the architechture addressable? can a memory module be connected to multiple server nodes and memory used as needed (like a Dynamic SAN environment can for secondary storage). This would allow the system ideally to scale to a full MAN type scenario, with virtualisation provisioning technologies reserving X amount of memory for new VM’s, could even allow memory sharing between server nodes at memory speeds for clustered applications and failover scenario’s.
is it, as I expect, some silicon that extends QPI onto some custom IBM external interface, allowing point to point or limited point to multipoint connection to another tray of ram. I get the feeling this is probably more the case.
I cannot confirm from what I have read so far that it’s either of these, but if it was the former I should imagine there would be more of a fuss…
I guess i’ll have to wait a while for a distributed network technology that is still DDR3-Quick even after additional network addressing overheads.
Still, good to see movement from the norm, if for no other reason than to spark technical discussion.