Co-Design of Network, Storage and Computation Fabrics for Disaggregated Datacenters

Principal Investigator(s): 
Sylvia Ratnasamy

Traditional datacenters are built using servers, each of which tightly integrates a small amount of CPU, memory and storage onto a single motherboard. The slowdown of Moore's Law has led to surfacing of several fundamental limitations of such server-centric architectures (e.g, the memory-capacity wall making CPU-memory co-location unsustainable). As a result, a new computing paradigm is emerging --- a disaggregated datacenter architecture, where each resource type is built as a standalone "blade" and a network fabric interconnects the resource blades within and across datacenter racks. Several recent studies suggest that such an architecture will allow datacenters to have 10-100x larger CPU and memory capacity and 100x larger bandwidth than server-centric architectures. Unsurprisingly, major cloud service providers are already deploying disaggregated prototypes, e.g., Google, Intel, Microsoft, Facebook, etc.

While beneficial from the computer architecture perspective, decoupling of resources alters many fundamental assumptions that once guided the design and optimization of existing distributed networks, systems, and applications (e.g., compute-storage co-location, high CPU-memory bandwidth, storage hierarchy, data locality, failure models, etc.). Thus, networks and systems will need to be re-architected to operate correctly and efficiently on, and to exploit all the benefits of disaggregated architectures.

In this collaborative project with researchers from Cornell, along with industrial collaborators from Microsoft,  Google, and Intel,  ICSI researchers are working on the necessary infrastructure for this new computing paradigm. The team of researchers includes experts on networking, systems, algorithms and computer architecture, and together are working toward  the ambitious goal of co-designing the entire network, storage and computation fabric for this disaggregated architectures.