[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [RRG] Why delaying initial packets matters
On 2008-2-17, at 21:31, ext Stig Venaas wrote:
My experience is that some applications don't handle delayed or
dropped
initial packets all that well. This may be okay since it rarely
happens,
but it's a problem if it becomes the norm.
Exactly!
Part of the problem here is
how the application/transport can distinguish between a delay/drop due
to a cache-miss (i.e. mapping not in place) and a packet dropped for
other reasons.
Right. Most transport-related mechanisms assume that loss equals
congestion, and they react by taking load off the (apparently)
congested path.
Wireless networks are one example where loss can be due to corruption,
and transports have struggled to deal with this case. (And, in truth,
it's been addressed because the MAC designers have figured out how to
deal with this, and not because the transport protocols had adapted.)
If the routing system becomes a new source of loss or re-orderings
that aren't due to congestion, the issues will be similar, i.e.,
congestion control will be triggered without congestion being present.
How serious this is depends on how frequently that can happen, exactly
as you state above.
From a single application point of view it may be good
to retransmit the first packet multiple times in a short period of
time,
to make sure that one gets through as soon as possible. I'm not sure
if
it's good if applications did that in general.
It wouldn't be a good idea from a congestion control perspective,
because the initial phase of a connection is exactly when the
transport protocols have the least knowledge about the path
characteristics (bandwidth, delay, etc.), because they have no sampled
path state, so that's when they need to be most conservative.
Anyway, this bring me back to the point I made earlier, that we really
would need to look at some applications. I guess you have some
knowledge
how say BIND behaves? E.g., it looks like the host command will wait
for
5s if the first request is dropped. Is a 5s delay to DNS lookups
acceptable?
DNS applications are a good case to look at, as is web traffic, SIP
stacks, etc.
Lars
--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg