Arvind Krishnamurthy
University of Washington, WA·Computer Science
Data centers run the online services that billions of people use daily, and they are expensive to build and power. One way to cut that cost is disaggregation: taking hardware pieces such as memory or storage out of individual servers and pooling them over a network so they can be shared flexibly.
- Fields
Computer Science
Engineering
- Topics
data centers
disaggregation
disaggregated storage
- Grants
- 2 known grants, through Jul 2029
What they work on
- Design NetFusion, a rack-level pool of programmable network interface cards (SmartNICs) that perform packet and network-function processing for many servers, with safe and fair sharing among applications.
- Develop methods to allocate tasks across SmartNICs with differing resources, handle both long-term traffic and short bursts, and provide fault tolerance and quick reaction to workload changes.
- Build programmable disaggregated storage (PDS), including an elastic, backward-compatible client-side virtual disk and control- and data-plane programming frameworks for expressing application intents.
- Enforce fair resource allocation across different input/output devices and schedule data-plane operations across devices to give predictable storage access latency.
More about the research
These two collaborative projects extend that idea. One asks whether network processing itself can be pooled. The other makes pooled storage programmable so it can adapt to each application.
Why it matters
Better sharing of networking and storage hardware could lower the cost of running data centers. That would make the online services billions of people use cheaper to provide and new public services cheaper to deploy. Cloud providers and enterprises could gain faster, more predictable and fairer storage and networking. Public prototypes would let others build high-performance data center applications.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Paul Ruth
University of North Carolina at Chapel Hill, NC·Computer Science
Deploy next-generation SmartNICs, programmable network cards called Data Processing Units, at four international FABRIC testbed sites so researchers can offload networking, storage and security tasks from CPUs and GPUs.
Peter Alvaro
University of California-Santa Cruz, CA·Computer Science
Design memory protocols that run over network fabrics so that many computers can reliably and quickly reach large pools of remote, disaggregated memory.
Daehyeok Kim
University of Texas at Austin, TX·Computer Science·Early-career
Create a programming model that describes robot manipulation tasks as cooperating control loops with explicit deadlines and recovery from robot and network failures.
The school
University of Washington: 81 labs on PI Pages, $1.7B a year on research, 694 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.