Last week’s news reports that COPAN Systems are back from the dead after their purchase by SGI. The newly branded arrays have been tweaked (for example 50% of drives can be spinning on the SGI incarnation) but essentially the concept is the same; store shedloads of inactive data.
HP Blades Day made me do some serious thinking regarding the way blades, virtualisation and storage will mix together. In one of the sessions, Paul Perez from HP discussed memristor technology and how it may be the successor to solid state drives. I posited that it could be possible as drives get larger in capacity we see a reduction in drive speeds. Here’s my thinking.
Virtualisation and data are becoming tightly coupled, to the degree that a subset of the overall data in an application will be manipulated by a virtual machine. Over time that virtual machine becomes a virtual application. The active data remains in storage directly associated with the blade chassis; inactive data is demoted to storage arrays and retrieved as necessary. This is analagous to any virtual storage management; my preferred example would be the mainframe MVS or z/OS operating system where virtual memory pages out unused memory blocks. In the storage world, active data still needs to be persistent, so we’re seeing technologies that offer SSD arrays or other persistent memory (Fusion-IO)
Associating the VM and the active data has a number of advantages; firstly the data replicated/protected need only be the active components. This can be moved around synchronously if required, but essentially a much smaller subset need be configured. Second, the tiering concept can be pushed even further with a storage tier directly integrated into the blade hardware.
For example, it would be possible to integrate the latest arrays from Violin into a slot in a blade chassis. This provides the storage locally; the hypervisor actively moves data in and out of the local area as needed. This doesn’t stop the VM guests talking directly to external storage but improves performance by providing a high speed read/write cache of a significant size within the chassis.
As more workload is virtualised, so the standard disk array becomes a big paging file for inactive data. Perhaps in 10 years the SGI/COPAN proposition could be attractive after all.