RE: IP QoS issues

["Constantine Protopapas" <wiztech at cmpnetmail.com>]

> -----Original Message-----
> From: Masataka Ohta [mailto:mohta at necom830.hpcl.titech.ac.jp]
> Sent: Tuesday, 7 July 1998 1:36
> To: peter dawson
> Cc: mohta at necom830.hpcl.titech.ac.jp; karl at CaveBear.com;
> wiztech at CMPNETMAIL.COM; peterdd at gto.net.om
> Subject: Re: IP QoS issues
>
>
> Pete;
>
> > > > And instability of routes is going to make it difficult to
> > > maintain QoS
> > > > once a path is established.
> > >
> > > A solved problem.
>
> Here, I said the instability is a solved problem.
>
> >    IMO, there is still much work to be done B4 QoS comes of age.
> >
> > In therory once  a policy function has been configured tomatch
> > router QoS Thresholds, the router must ensure that packets
> > transmitted on a VC do not get dropped by the ATM policy
> > function. This could happen as a result of processing
> > jitter=92s in the router and may result in transmitted
> > cells not to conform to their contract.
> > However there is a lack of mechanisims/specs to map the ATM layer
> >
> > decision agreement and IP layer decision to facilitate a
> > consistent
> > admission policy.
>
> Don't assume ATM, here.
>
> On the Internet, where certain amount (50~30%) of bandwidth must
> be reserved for best effort traffic, average queue length of QoS
> assured traffic will be 2 or 3 that we don't have to worry about
> jitter.
>
> For the time being, a point-to-point giga ether will be the
> major link technology to offer QoS.

>From what I know Gigabit Ethernet has not QoS built in. I can't see how it
can offer QoS with its inherited limitations. 802.3 is not designed with QoS
in mind.
> > Furthermore the  paths of vcs from different routers can overlap
> > in
> > the backbone and this cannot be noticed explicitly. The aggregate
> > numbers
> > work out because  each router can only put so much traffic into
> > the net,
> > Then problems can come up if many routers use the same internal
> > path.
>
> That some ATM switches can not support so many VCs/VPs is a problem
> of the ATM swichies.
>
Depends always on the size of the ATM switch. Most switches that are used on
the backbone (eg. Newbridge 36190 Core Services Switch) goes up to a maximum
2.000.000 VPCs and VCCs per switch. Generally, high-capacity ATM switches
have adequate capacity.