Challenges for the Large-scale Deployment of RDMA in Datacenter

PhD Qualifying Examination


Title: "Challenges for the Large-scale Deployment of RDMA in Datacenter"

by

Mr. Zilong WANG


Abstract:

RDMA is an important technology widely deployed in data centers. RDMA provides 
extremely high throughput, ultra-low latency with very low CPU overhead, which 
fits the demand of various distributed systems. However, RDMA’s large-scale 
deployment problem remains unsolved. This well-known scalability issue can be 
classified into two parts, connection scalability issue, and network 
scalability issue. Connection scalability issue means the performance drops 
dramatically when concurrent connections increase. Network scala- bility issue 
indicates that more incidents happen due to PFC when network topology gets 
larger. Much effort has been made to fix these problems. Some work attempts to 
mitigate the risk, while others try to eliminate them. In this survey, we first 
briefly introduce RDMA about its properties and applications in data centers. 
Then we analyze the influences of scalability issues and their root causes, 
i.e. , limited hardware resource and required PFC mechanism. Finally, we will 
describe different approaches and conclude their design philosophy.


Date:			Tuesday, 27 July 2021

Time:                  	2:00pm - 4:00pm

Zoom meeting:
https://hkust.zoom.us/j/99749187323?pwd=UnhNYlM5eU1MUUwvckFVOUh2OTFrUT09

Committee Members:	Dr. Kai Chen (Supervisor)
 			Prof. Bo Li (Chairperson)
 			Dr. Brahim Bensaou
 			Dr. Wei Wang


**** ALL are Welcome ****