School of Computer Science Assistant Professor Alexandros Daglis has received the Association for Computing Machinery Special Interest Group on Computer Architecture (ACM SIGARCH) and Institute of Electrical and Electronics Engineers Technical Committee on Computer Architecture (IEEE-CS TCCA) Outstanding Dissertation Award Honorable Mention. The honor is given to doctoral work that advances the computer architecture field.

Daglis was selected for his dissertation, Network-Compute Co-Design for Distributed In-Memory Computing, for “contributions to network-centric server architecture for in-memory datacenter services.” The research revisits the fundamentals of how communication-intensive systems function as distributed data retrieval systems grow in size and performance demands. Daglis suggested that co-designing network endpoint resources with central processing units can drastically improve the capabilities of distributed systems.

[Related: Alexandros Daglis Finds New Beginnings in the End of Moore’s Law]

“Traditionally, the network was so much slower than computation that designing the two subsystems completely independently was the right way to go,” Daglis said. “But with recent advancements in networking technology, that performance mismatch is a lot narrower; a datacenter’s network traversal only takes a few of microseconds. I advocate blurring the rigid boundaries between network and computation and pushing semantically richer functionality closer to the network endpoints is the way to go for next-generation distributed systems.”

The industry is starting to take an interest in this direction as well by provisioning extra computer resources on the network endpoints. Smart programmable network interface controllers are re-emerging and being used in production datacenter environments; a prime example is Microsoft’s Catapult project.

Yet Daglis wanted to take it a step further: instead of replicating computation resources on the network endpoint, fuse and co-design the conventionally discrete subsystems of compute and network endpoint. This not only improves efficiency, but also additional types of acceleration that are otherwise unattainable.

“It was truly an honor to have my dissertation work acknowledged by the architecture community,” Daglis said. “It’s definitely an exciting time to be working on network-centric architectures.”

This is not Daglis’s first recognition for his work. The École polytechnique fédérale de Lausanne (EFPL), where Daglis completed his Ph.D., also awarded him a thesis distinction.