Ricochet is a low-latency reliable multicast protocol designed for time-critical applications in a clustered environment. Ricochet combines native IP multicast with proactive receiver-based forward error correction, achieving high levels of consistency with stable and tunable overheads. The protocol is designed to minimize delay: a given packet is typically sent just once, delivered immediately upon receipt, and in the event of packet loss will generally be recovered and delivered in milliseconds, long before the loss could even be detected by the application.

Ricochet is a highly scalable protocol and scales extremely well especially in the number of groups. Our evaluation of a Ricochet on a 64-node cluster shows that it scales upto 1024 multicast groups per node while delivering more than 99% of packets at an average of 25ms. This characteristic, we believe, makes Ricochet perfectly suitable for applications hosted in datacenters and clusters where applications join a large number of small sized groups.

An initial alpha release of the Ricochet implementation in Java is available for downloaded. To get the release, you have to download, sign and fax
this license form to Bill Hogan at 607-255-4428.
Stay tuned for updates as we continue to build more functionality and remove bugs. To join the Ricochet mailing list, email Bill Hogan.
Send any comments/questions/requests to either Mahesh Balakrishnan or Amar Phanishayee.
Misc. Presentations


Mahesh Balakrishnan
Professor Ken Birman
Amar Phanishayee
Stefan Pleisch