Speaking of techs, Martin Schoffstall left PSI a couple of months ago
to start a company called Epicenter. I thought he was a key technical
asset for the compny, so I was pretty disapointed when I heard this.
However, somebody told me he was a real pain to work with, so his
departure might make things easier for whover they are going to merge
or be bought by.
There was a discussion about PSI's tehnical merits on the network management newsgroup a few months ago. This was the original post:
I am interested in getting opinions on which Internet Service Provider with a
national scope has the best backbone network. I want to know who has the best
designed backbone, if your objective is to minimize latency and maximize the
lowest sustained transfer rate between two corporate sites connected to the
same ISP. Said plainly, I want to find out whose network is the most
reliable.
PSI seemed to originate the idea of running a switched digital backbone that
connects all of the backbone routers by high-speed frame relay switches. I
notice that almost all of the large players (Netcom, UUNET, and Sprint) are
moving to this same architecture.
In studying the PSI network diagrams, I am extremely impressed with the
design. Frame Relay customers attach directly to a Frame Relay switch at the
PSI point of presence (POP), bypassing entirely the Cisco router at the POP.
The design seems to suggest that in many instances you could go from one
corporate site to another corporate site over the backbone with a *single*
router hop (excluding the routers on the customer premises at each corporate
site). This would look like:
SiteA | POP | Backbone | POP | SiteB
Router <-> Switch <-> Router <-> FrameNet <-> Router <-> Switch <-> Router
Is this the reality?
Honestly, I have *not* heard good things about PSI. I
have heard that they have poor customer service, and that they tend to ignore
problems. But I'm trying to separate the customer service issues from the
pure technical issues. Is their design the best one?
Among those vendors who use this design, which ones do not oversubscribe their
networks?
The overall purpose of this question is to try to determine if the latest
generation of ISP designs might support reliable corporate backbones built
over the Internet. Note, I'm not interested in performance of connections
outside of the ISP network. I'm focused solely on corporate site connections
between sites that all connect to the same ISP.
The above design seems very lean, and likely to support
low latency and high throughput, as long as you don't oversubscribe the
backbone. If there is something better, I would like to know about it. If
there are problems with the above design, I would like to know about that too.
This posting evolved into an argument over swiching vs. routing.
It is archived at smurfland.cit.buffalo.edu.
There were also some comments about PSI's customer service; some were
good, some were bad.
............Marc |
