[Spread-users] changing TTL to use IP multicast in WAN
John Lane Schultz
jschultz at spreadconcepts.com
Fri Jun 23 11:26:15 EDT 2006
Igor Sindyakov wrote:
> 1. Speaking of scalability, what numbers (computers, daemons, applications,
> groups, message size, etc.) can Spread handle in practice? "The Spread
> Toolkit: Architecture and Performance" document provides some numbers from a
> laboratory test ( http://www.cnds.jhu.edu/pub/papers/cnds-2004-1.pdf ). What
> about the real-life performance?
For the number of daemons we suggest you limit the total number to somewhere between 25 to 35. Beyond that, the timeout assumptions begin to break down and the ring protocol that the daemons run becomes less stable.
Each daemon is capable of handling up to about one thousand client connections, depending on the load each one occurs. Each daemon can usually support a couple of hundred active clients without trouble.
The number of groups Spread can support is quite large. We've done performance tests on the new 4.0rc2 version with over a million separate groups with little additional overhead. Basically, the main limiting factor becomes your machines' RAM memory.
Spread natively only supports fixed size messages. Out of the box the limit is around 100KB.
> 2. As to the codebase, is any courseware or other technical information
> available for developers who are going to do some enhancements / fixes to
> their copy of the code (with the intention to suggest them back to the
> Spread development)?
Not really, unfortunately. The best resources we have for that are the (architectural) academic papers that discuss Spread, although often the papers end up discussing the experimental version of Spread rather than the Ring protocol (available) version of Spread.
Spread Concepts LLC
Phn: 443 838 2200
Fax: 301 560 8875
More information about the Spread-users