On Wed, Jul 03, 2002 at 01:58:15PM +0100, Matthew Toseland wrote:
> Once a node gets over 100% load, it never comes back down again, and it
> stops doing any more work.

I have observed this too - basically the QR mechanism is failing, since 
its purpose is to reduce the load so that the node can stop QRing - this 
clearly isn't happening.

I think that simply restarting the node is just sweeping the dirt under
the carpet, we need to think about why the QR mechanism is failing to
reduce the load on the node, and address that directly.  Oskar's
suspicion is that the QR mechanism isn't working because it does not
reduce the workload on the network (since the HTL of the QRd message
doesn't drop - thus an overloaded node actually *increases* the workload
on the network as a whole). One possibility, then, would be for a QR to
cause a message's HTL to drop by 1, 2, or maybe even half.  The
rationale would be that an overloaded node is highly undesirable, other
nodes in the network should take both local (avoid routing to the
overloaded node) and global (get more aggressive with reducing HTLs)
action to eliminate the problem.

Ian.

-- 
Ian Clarke                                        [EMAIL PROTECTED]
Founder & Coordinator, The Freenet Project    http://freenetproject.org/
Chief Technology Officer, Uprizer Inc.           http://www.uprizer.com/
Personal Homepage                                       http://locut.us/

Attachment: msg03379/pgp00000.pgp
Description: PGP signature

Reply via email to