Re: HTTP -> Messages -> Transport factoring

In message <CB48ADDE-FC3C-423A-ADE6-4C4E5796FF36@tzi.org>, Carsten Bormann writ
es:

>I'm old enough to remember when NFS mostly moved from UDP to TCP.
>Are they moving back?  Where/why not?

The usage-model for NFS is that generally things are read/write,
cannot fail and you move massive amounts of data around.

The move from UDP to TCP more or less coincided with "massive" growing
from below 64k to above 64k, where you'd have to start doing the
hard work with UDP.


The HTTP/UDP use-case is different, in that we are specifically
talking about non-transactional small object retrival, but the
64k limit is of course the same thing.

Once of the attractions of HTTP/UDP is that you can multicast and
whichever server happens to have that icon will send it to you.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.

Received on Thursday, 5 April 2012 20:42:35 UTC