[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: unmanaged scope comments

>>>>> On Fri, 20 Sep 2002 14:39:27 +0200, 
>>>>> Harald Tveit Alvestrand <harald@alvestrand.no> said:

>> i don't understand what you want to do about it.  will you ping/ping6
>> destination and pick shorter roundtrip?  when multiple A records
>> are returned for www.foo.com, do you probe them to get the optimal
>> performance?  i believe the "quality" factor does not matter, and
>> it cannot be measured from endpoint (at least within short time).

> I have wanted for a long time someone more competent than me to run an 
> experiment where one, after a name lookup, would send SYN packets to *all* 
> the addresses returned, and only complete the three-way handshake on the 
> first one to reply.

> after all, it seems impossible to get anyone to take seriously the job of 
> getting the network to tell me which of two addresses are "best", and the 
> servers in these DOS-attack days have to defend against synfloods anyway.

> just wasting some SYN packets seems to me to be likely to achieve the "low 
> hanging fruit" of *most of the time* picking the best address to use.

> so far, afaik, nobodoy's tried it.

This sounds like an experimental library function I wrote two years ago:

http://orange.kame.net/dev/cvsweb.cgi/kame/kame/kame/libinet6/mconnect.c

(though the library makes a short delay between two consecutive
connect(2)s.)

I used the library with some applications and found it worked well,
but I've not promoted the new library because I didn't know how this
could be widely deployed.

Also, if I understand correctly, there is a web client that has a
similar routine inside the application.

					JINMEI, Tatuya
					Communication Platform Lab.
					Corporate R&D Center, Toshiba Corp.
					jinmei@isl.rdc.toshiba.co.jp