Deploying a mix of datacenter resources in a preconfigured server – compute, storage, network, and memory – in a way that they are fixed, can’t be tuned to a use case, and must be replaced entirely for an upgrade is how the IT industry has been working for years. Each server is an island.
This is an inefficient path when you deploy more than a few servers. That’s why there is an architectural movement in hyperscale datacenters (and it’s sure to be emulated by enterprise in a few years) to “disaggregate” – or, the term I prefer, “pool” – these resources. That allows deployments to be “configured” in the field, creating tailored platforms depending on needs. And it enables more efficient life-cycle management of subsystems. Ultimately this enables more work/$, and that’s almost everyone’s goal. One hyperscale CTO I know told me over dinner he views this pooling and allocating as “hardware-based virtualization,” which is sort of true.
In this AIS interview I talk about the concept, the rational, and show how costly forklift upgrades will be behind us once this small movement becomes common practice.
Keeping up with the flood of global data with network acceleration
It’s hardly a secret that the growth of PC, mobile, and intelligent media devices and their related applications worldwide is exploding. It also no secret that they are driving an equivalent increase in global network traffic. The mystery for many datacenter network managers is how to keep up, as the flood of data traffic saturates networks, drags down network performance, and frustrates users with waiting.
Here, LSI’s Troy Bailey discusses how networks will become smarter to deliver the data you need, in the priority you need, when you need it.