Re: Thinking about ICP

From: Gregory Maxwell <>
Date: Fri, 1 Aug 1997 23:24:43 -0400 (EDT)

On Sat, 2 Aug 1997, Joao Carlos Mendes Luis wrote:

> Optimizations to this TCP pipe would be:
> 1) Detecting lost links, and reconecting them after some time.
> 2) Detect over-congestion (queue always full, blocking socket I/O) on
> one link and taking some control action (which ?).
> 3) ICP over TCP proxy. (humm...) Since we'll no more have
> multicast support, we could have some kind of exploder
> to run at network mesh nodes, simulating a multicast tree.

 How about adding this feature to ICP:

Squid servers can keep tables of ICP ping times to various hosts (like it
already does).. And can query less often when the responces to slower..
After all, the slower the responce the less useful it is.. Also, an ICP
host should be able to send out 'backoff' packets.. Also the icp table
scheme should sort and group by IP subnets (which usually reflect network

Anyone on this list know anything about lossy internet transfer formats?
I've been toying with the idea of implimenting such a thing between
squids.. (I've already got two squids gzipping all text/html traffic
between them).. Anyone know of any compression methods that a cache could
sanely do in realtime, but are still useful for
internet-lossy-image-transfer (meaning they must be divisable by content
importnace, like wavelet methods).. The reason I thought of this is
because the 'ping' tables I mentioned above are exactly the sort that are
needed for congestion control for lossy image transfer..
Received on Fri Aug 01 1997 - 21:31:13 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:35:55 MST