My research interests lie in distributed systems, with particular focus
on overlay networks, application level multicast systems and peer-to-peer
protocols. My supervisor is
Robbert van Renesse. I'm a member of
the Spinglass group, which does research in reliable distributed systems and
networking protocols.
In early work, together with Robbert van Renesse and Dan Dumitriu, I built and
evaluated the performance of SelectCast, an application level multicast protocol. SelectCast goes beyond
previous work on Application Level Multicast Routing
Protocols (ALMRPs) in the following two ways:
- SelectCast offers users great flexibility in how to
select routers. For example,
routers can be selected to minimize latency or maximize throughput. The
default selects routers based on longevity, thereby attempting to optimize
robustness. The selection may be changed at run-time.
- SelectCast allows senders to specify the set of
intended destination hosts through the use of a SQL condition on selected
attributes of such hosts. Note that
this is different from publish/subscribe systems in which the subscribers
specify what messages they are interested in receiving, either by topic or by a
predicate on messages. Our strategy
is strictly more powerful than traditional topic and content-based
publish/subscribe mechanisms.
I am currently developing FastCast, a routing
Infrastructure for IP multicast–based WAN applications. FastCast allows members of WAN applications to
send, and receive messages using the standard IP-multicast interface. Usually,
the members of WAN applications reside in different Internet sites that exploit
their choices of hardware multicast. Using the FastCast infrastructure, a WAN
application multicasts messages to an IP multicast address. FastCast sends the
application’s messages between sites using a tree of TCP connections built
using control information received from an ALMRP component. Each dissemination tree has assigned a Flow ID. The
ALMRP may be chosen based on the application’s choices, and the number of
sites used by the application. When several applications use the same AMLRP
between the same sites, FastCast reduces ALMRP overhead by disseminating
messages through the same tree of TCP connections. Moreover, in order to be
TCP-friendly, FastCast uses at most one TCP connection between any pair of
sites, and may run a credit-based flow control on each TCP pipe.
By using the FastCast infrastructure, site
members receive original IP multicast packets regardless of where senders
reside. Sites can use their choices of IP multicast protocols without changing
their software. The FastCast infrastructure is able to change the ALMRP assigned
to an application on-the-fly and experiences less “churn” by adding and
dropping IP multicast addresses at the site level instead of the host level. The
FastCast infrastructure separates the data path from the control path for any
choice of ALMRP. Such separation improves forwarding performance and updates
ALMRPs on-the-fly.
Publications
- Adrian Bozdog, Robbert van Renesse. FastCast: A routing infrastructure
for Ip-multicast based WAN applications. Work in progress.
- Robbert van Renesse, Adrian Bozdog. Willow: DHT, Aggregation, and
Publish/Subscribe in One Protocol. Proc. of the International Workshop on
Peer-to-Peer Systems (IPTPS '04). February 2004, San Diego, CA.
- Robbert van Renesse, Kenneth P. Birman, Adrian Bozdog, Dan Dumitriu,
Manpreet Singh, Werner Vogels. Heterogeneity-Aware
Peer-to-peer Multicast. Proceedings of the17th
International Symposium on Distributed Computing (DISC 2003).
October 2003. Sorrento, Italy.
- Adrian Bozdog, Robbert van Renesse, Dan Dumitriu. SelectCast
-- A Scalable and Self-Repairing Multicast Overlay Routing Facility.
Proceedings of the First ACM Workshop on Survivable and Self-Regenerative
Systems. October 31, 2003. Fairfax, VA.
- Adrian Bozdog, Robbert van Renesse, Dan Dumitriu. The Performance of SelectCast
-- Scalable and Self-Repairing Multicast Overlay Routing. In submission .