*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Novel and Efficient Techniques for Guaranteeing Routing and Protocol Level Deadlock Freedom in Interconnection Networks
Committee:
Dr. Krishna, Advisor
Dr. Kim, Chair
Dr. Daglis
Dr. Gratz
Abstract:
The objective of the proposed research is to come up with novel techniques that simultaneously provides routing and protocol level deadlock freedom in the interconnection network. Interconnection networks are the communication backbone for any system. They occur at various scales: from on-chip networks between processing cores, to supercomputers between compute nodes, to data centers between high-end servers. One of the most fundamental challenges in an interconnection network is that of deadlocks. Deadlocks can be of two types: routing level deadlocks and protocol level deadlocks. Routing level deadlocks occur because of cyclic dependency between packets trying to acquire buffers, whereas protocol level deadlock occurs because the response message is stuck indefinitely behind the queue of request messages. Both kinds of deadlock render the forward movement of packets impossible leading to complete system failure. Prior work either restricts the path that packets take in the network or provisions an extra set of buffers to resolve routing level deadlocks. For protocol level deadlocks, separate sets of buffers are reserved at every router for each message class. Naturally, proposed solutions either restrict the packet movement resulting in lower performance or require higher area and power.In my work, I propose a new set of efficient techniques for providing both routing and protocol level deadlock freedom. My techniques provide periodic forced movement to packets in the network, which breaks any cyclic dependency of packets. Breaking this cyclic dependency results in resolving routing level deadlocks. Moreover, because of periodic forced movement, the response message is never stuck indefinitely behind the queue of request messages; therefore, my techniques also resolve protocol level deadlocks. I refer to them as 'subactive' class of techniques.
