ADSM-L

Re: ATM: SVC vs. PVC

1998-01-20 18:03:24
Subject: Re: ATM: SVC vs. PVC
From: Nick Clayton <clayton AT GLOBALNET.CO DOT UK>
Date: Tue, 20 Jan 1998 23:03:24 -0000
There is no difference in the possible throughput between using either PVC's
and SVC's. They are just two different methods of obtaining an ATM
connection when running TCP/IP over ATM. (Necessary as ATM is switched
whereas TCP/IP is connectionless.)

Permanent Virtual Circuits are set up manually whereas Switched Virtual
Circuits are established automatically by an arp server.

There are advantages and disadvantages to both. The only (on AIX anyway)
type of connection supported by IP over ATM emulations is UBR (Unspecified
bit rate) when using SVC's which will use what amounts to a round robin
method (actually choice of shortest and widest path choosing the path with
least connections down if multiple routes of equal length/width are
available) of establishing circuits if multiple paths are available.
This may cause problems if you have some connections with higher traffic
rates than others as some links may become saturated while others have very
little traffic as there is no distinction between a connection with a ping
running down it and a 20GB/hr backup.

PVC's will let you control the path of each connection but if you haven't
set up a PVC then no connection and they will not use even the primitive
load balancing above so much thought may be required. Also if you have
multiple paths you will still lose the connection if the one with your PVC
on it goes down.

We use SVC's at our shop and have encountered problems due to having single
connections taking up approximately half the bandwidth of a 155Mb pipe and
hence very little granularity in data traffic. As a result we are planning
to go to 622Mb for our switch to switch links as this traffic will be a
smaller percentage of the bandwidth available over a single link in the new
configuration.

Also LANE emulation (as opposed to classic IP) can create many to one
circuits (looking like one connection from one end any two+ from the other)
which can cause further imbalances in circuit distribution.


    Nick Clayton
(Vertex Data Science)




<Prev in Thread] Current Thread [Next in Thread>