pc world+mtu+meltdown

Dan Boehlke dboehlke at mr.net
Thu Nov 5 07:36:45 UTC 1998


Users wouldn't have to do this if Windows had a fully implimented IP
stack.  About a year ago a weekly industry MIS magazine ran a series of
articles, by "networking industry experts," on why the internet or any
scale based entirely on the Windows 95 and NT implimentation of TCP/IP,
winsock.  They saw a protocol that killed the network with retransmissions
when there was congestion.  Didn't adjust for differnt MTU on different
media.  Would break when the server was on a token ring (4k MTU) and the
client was on an ethernet (1.5k MTU.) unless a registry setting was
tweeked.  It turns out that the registry setting clears the don't fragment
bit in the IP header, its on by default.  Its like Van Jacobson was never
born!  Where are this stacks backoff algorithms?  Where is the MTU
discovery?  Where is window adjustment?  The VJ additions to TCP/IP as
well as contributions of many others make it scale.

Yet, because of the marketplace, or monopoly, only one Washington or the
other will decide, there are few options for Windows users.  The market
dried up when Winsock was included, and I see few of the competing vendors
advertising that they have a complete stack that would remedy these
issues.

I won't blame Steve Bass, if the stacks on Windows computers really
worked, his article would never have been necessary.  His mistake is that
he thinks he has discovered something we, the ISPs, are hiding from the
users.  We should let him know that its not us, but the secrets of
Redmond, and the complacentcy of the marketplace when it accepts
non-conformant code that needs to be uncovered.

I have been searching the web for a document that describes the
shortcomings of the Microsoft winsock implimentation of TCP/IP, I cannot
find one. Yet far too often I am asked to resolve problems directly
related to its misbehavior under congestion or lack of MTU discovery. Have
others not run into these?  Why isn't the word out?  Where are the reviews
in the press that would point these shortcommings out to potential users?

As NT servers and workstation continue to grow on the net, these issues
need to be addressed.

They have already told me that NT 5.0 fixes all... I am not holding my
breath, or is it Windows 2000 now.


On Wed, 4 Nov 1998, Jared Mauch wrote:

> 
> incase you see more pps stressing your network, or
> something of the sort, check this out for possible
> reason why:
> 
> http://cnn.com/TECH/computing/9809/29/unclog.idg/
> 
> 	please follow reply-to header.
> 
> -- 
> Jared Mauch  | pgp key available via finger from jared at puck.nether.net
>              | http://puck.nether.net/~jared/
> 

--
Dan Boehlke, Senior Network Engineer                          M R N e t
Internet:  dboehlke at mr.net                       A MEANS Telcom Company
Phone:  612-362-5814                  2829 SE University Ave. Suite 200
WWW: http://www.mr.net/~dboehlke/                Minneapolis, MN  55414




More information about the NANOG mailing list