HydraNet -- Network Support for Very-Large Scale Services
The rapid diffusion of the Internet and the explosive increase of the Web's popularity have put the networking infrastructure (routers, links, protocols) and the server hosts under severe stress. The question of how to scale the Internet, in particular its services, has become an important one. With HydraNet we are investigating network-level mechanisms that provide solutions to this problem. In particular, we propose the replication of transport-level service-access points to allow for transparent distribution and/or replication of IP services. We have proposed a set of mechanisms for IP that allow server processes to run on remote machines, while remaining addressable with the home machine's IP address. These conceptually simple mechanisms were realized on a local testbed and turned out to add very little overhead, while being extremely versatile. As a first application we deployed HydraWeb, an infrastructure for intelligent push-based Web caching.
We have integrated support for very-high availability into HydraNet. We use HydraNet-FT as a basis to provide truly fault-tolerant TCP service, in which failures of servers, and recovery, is fully invisible to TCP clients, to the TCP stack at the client side.
G. Shenoy, S.K. Satapati, R. Bettati,
"HydraNet-FT: Network Support for Dependable Services."
Proceedings of the 20th IEEE International
Conference on Distributed Computing Systems.
Taipei, April 2000.