Christian Hopps writes: > In particular why don’t we simply indicate that a lost packet can > induce a delay of the fixed packet interval times the window size - > 1, and so the widow size should be kept to a minimum, and leave it > at that.
As the window size is configured by the adminstrator and adminstrator has no idea how it will affect things. We can't really expect adminstrators to understand that when they have decide that they have hundreds of GBs of memory in their VPN machines, that configuring the reorder buffers and replay windows to 100000 packets long (absurdly large value, but its just few gigabytes of memory for those buffers so who cares about the memory consumption) that will suddenly cause really long delays for all packets if any single packet is ever lost. If the network is as reliable as you assume they always are and you almost never loose packets it might take very long until this happens and when it happens suddenly their whole network freezes until the reorder buffer is filled up and then dumped to the network in one huge burb. This would be really bad behavior and most likely that would cause them to submit bug reports to the implementations and implemetors would blame the adminstrators for putting such stupid values in and adminstrators would blame implementors for allowing such things and not having warnings in the manual of not doing so, and implementors said they did not realize that would happen as RFC did not say so. If we do not explain why window size should be kept mimimum having such text there does not mean anything. And I assume the usable minimum would be the same than for minimum replay window size that must be supported, i.e., 32 packet, which can still cause 32x of normal delay for every lost packet. -- kivi...@iki.fi _______________________________________________ IPsec mailing list IPsec@ietf.org https://www.ietf.org/mailman/listinfo/ipsec
