[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: ZRTP comments part 1: congestion
I wanted to comment on the "packets lost at the beginning of the
flows" topic. I've talked to quite a few people about this including
the transport ADs, some IAB folks, and some other IESG folks - there
were all very doubtful that this is actually happening. You can test
it yourself by trying to load a web page. If the first SYN packet
was lost, it would be 3 seconds before the first retransmission and
the delay of opening the page would be very noticeable. When trying
to trace back where this rumor came from, it seems that some times
the first packet that a device sends on a LAN may be lost due to
issues in how devices are processing ARPs. However, we aren't talking
about the first packet sent from the interfaces of a device here, we
are looking at problems on new flows for a device that is already
done previous network transactions.
When Phil first told me about this, I asked him to send me some flows
or point me to a network where I could try it out since it seemed
surprising to me. I'm still waiting for any more information about
where this occurs. I can't find any extensive discussion in AVT but
the discussions I can find is back in Nov 1 on the thread "ZRTP
Initial packet loss". On that thread, Dan asked you for some details
for a sort of test set up where we could duplicate or observe this.
Still waiting to see any reply to that too. Actually the one reply to
that was from another implementor of ZRTP saying they are not seeing
this problem.
Cullen <with my individual hat on>
On Mar 17, 2007, at 5:33 PM, Alan Johnston wrote:
Actually, this has been extensively discussed in AVT - I would
refer you to the mail archives meeting minutes for the full text.
To summarize, it is a practical observation that packet loss at the
very start of a media session is often very high - perhaps 50% for
the first half dozen packets. Today, things like early media are
affected by this, and sometimes the start of a Hello. If ZRTP (or
DTLS as you will discover with implementation) happens prior to
media flow, you will need to deal with this packet loss or the key
exchange time will extend much longer than you would predict on paper.