Intel’s Disaggregated Server Rack Moor （英特尔的分解服务器机架沼泽）.pdf

Intel’s Disaggregated Server Rack Does “Disaggregation” Really Mean Anything? There’s been a lot of discussion about “disaggregated” servers, racks, and datacenters since Facebook and Intel promoted their vision for the phrase at the Open Compute Summit at the start of this year. Haven’t we spent the last few decades disaggregating datacenter architecture? And if so, what does disaggregation mean now? Is it something different? Executive Summary The proposed Facebook + Intel disaggregated server rack is an extension of current system architectures introduced by AMD, Calxeda, and HP. While scale is an ambitious differentiator, it is a bit of useful misdirection for Intel with respect to their target markets and workloads. The core differentiator for this interpretation of disaggregation Strictly speaking, is “atomicity” – separation of components at a functional level “disaggregation” to enable a range of related workloads that require some means dividing an scale as a baseline. Complete disaggregation is not a integrated whole thing technology aimed at all datacenters and every workload, at (an “aggregate”) into least not in this decade and perhaps not ever. its component parts. Facebook and Intel are designing a specialized, completely disaggregated architecture optimized for large address space in-memory databases and analytics. This architecture is not intended to take over the entire datacenter. But it will be relevant for many classes of Big Data analytics, and it just might (finally) displace mainframes. IBM is reacting to this potential threat with OpenPOWER, a wildcard competitor for ARM’s server licensees. Hyperscale State-of-the-Art Over the last decade, hyperscale datacenters have developed specialized architectures that partition workloads so that each runs on more optimal hardware. Specialized architectures