Research

Research Interest
Human Computer Interaction and Data Center Networking.

Earlier Work

Continuing research on the paper "A Scalable, Commodity, Data Center Network Architecture, Mohammad Al-Fares, Alex Loukissas, and Amin Vahdat, Proceedings of the ACM SIGCOMM Conference, Seattle, WA, August 2008".
Devised a fiber optic solution to reduce the cabling cost and complexity of a new data center architecture presented in the paper above.
By using commodity switches, a data center supporting k3/4 (k=48) hosts with the cost of less than 2 million dollars can be built. This is regardless of the fact that the cost and cabling complexity of 3k3/4 cat6 cables is a serious issue that can dramatically affects the total estimated cost of the data center. To solve this problem, two solutions were presented. First one was to use MMF fiber instead of cat6 cables and keep the same data center architecture. The second one was to remove the core switches and have an optical ring across all pods.

Implementation of the 16 host data center demo with 20 netfpgas.
Netfpga is a network board containing four 1 Gbps Ethernet (GigE) interfaces and a user programmable FPGA which is built by Stanford University. We used 20 of them to implement the demo of 16-host data center pretended in the paper above. We pre-populated the routing tables of each netfpga based on two-level lookup algorithm and validated the concept described in this paper.

Development and running Openflow on netfpga for the new data center demo.
Instead of loading the netfpga with reference router project (the default packages that make the netfpga function as a router), we used the Openflow package. Openflow is based on an Ethernet switch, with an internal flow-table, and a standardized interface to add and remove flow entries. In order to perform flow classification and flow scheduling, identifying large flows by a controller based on flow statistics is necessary.

 

© 2009 Pardis Miri