PSINet and C&W peering

C&W did indeed shutdown their peering connections to PSINet
this weekend. While there are many potential explanations
for their actions, I have no visibility into their decision
process. I am disappointed with their decision to disconnect.
PSINet continues to seek a resolution with C&W to restore normal
connectivity in order to avoid further negative impact to both
companies and the Internet. Their decision is hard to understand
based on the following:

- C&W and PSINet upgraded circuits used for peering between
  the two networks earlier this year. C&W's recent action
  seems inconsistent with the strategy that led to these
  upgrades.
- PSINet's recent addition of direct private peering with several
  of C&W's transit customers relieved the peering connections
  between the networks of a couple hundred Mbps of traffic
  (improving connectivity overall and, undoubtedly, lowering costs
  for those transit customers). This is significant only because
  C&W claims PSINet no longer has sufficient traffic to justify
  the connections according to their published standards. In
  fact, PSINet's overall traffic continues to grow.
- Most of the PSINet traffic previously destined for sites
  behind C&W has alternative paths through other providers.
  While this sounds like a generally good thing, especially given
  the actions C&W has taken, it does make it difficult for those
  that require certain traffic levels to be maintained consistently
  for peering. Specifically, C&W's customers (or C&W itself) could
  alter "natural" traffic flow to favor (or not) various connections
  to meet their published standards (or not). PSINet demonstrated
  to C&W that if naturally less favorable announcements were
  preferred, PSINet could make an almost arbitrarily large (or
  small) amount of traffic flow between the peers. Even so, in
  C&W's opinion, PSINet will not be able to comply with their
  peering policy's traffic standards. It is gratifying to note
  that even without C&W peering, substantially all of the
  traffic previously flowing between PSINet and C&W continues to
  be delivered.
- At this time PSINet has not disabled the C&W peering interfaces
  nor decommissioned any facilities. If C&W chooses to, they can
  re-enable interfaces on their side and bring back the connectivity
  lost between their non-transit customers and PSINet. PSINet
  remains open to discuss with them a new bilateral peering
  agreement if they so choose.

PSINet remains committed to servicing its customers and the Internet
with the best possible infrastructure and policies. PSINet still
maintains hundreds of peering connections with other ISPs throughout
the world. While posting about matters between PSINet and its
peering partners is not typical, the circumstances and questions
arising from C&W's decision required some clarification. Hopefully
this additional clarification helps everyone understand the current
situation.

-Mitch Levinn
PSINet

and you did what to Exodus, Abovenet, and maybe others?

PSI shut off other peering sessions. I know why and maybe you can share
this information with the rest of nanog.

If PSI really cared about their customers and the Internet as you so
write, you would have direct connections to companys like Exodus and
Abovenet.

Also, is there a isp-peering list we can move this to?

Thanks

Christian

oddly enough this looks exactly like the statments made by a few
providers when PSI decided to de-peer.

what goes around comes around.

[Disclaimer: I work for MFN/Above.net, that that for what you will.]

C&W did indeed shutdown their peering connections to PSINet
this weekend. While there are many potential explanations
for their actions, I have no visibility into their decision
process. I am disappointed with their decision to disconnect.
PSINet continues to seek a resolution with C&W to restore normal
connectivity in order to avoid further negative impact to both
companies and the Internet. Their decision is hard to understand
based on the following:

I find this quite ironic, as PSINet decided to terminate peering
with AboveNet in November of 2000. AboveNet was disappointed with
PSI's decision, and had worked hard to avoid the disconnection.
To this day, AboveNet is still unsure why PSI took that action,
with the damage that resulted.

The really interesting part, is the claims made below, which I will
address one by one.

- C&W and PSINet upgraded circuits used for peering between
  the two networks earlier this year. C&W's recent action
  seems inconsistent with the strategy that led to these
  upgrades.

The same was true for PSINet and AboveNet. AboveNet ordered multiple
OC-3 peering circuits to PSI. The first was turned up under our
agreement. It was not until we tried to turn up the second that
PSI decided to reevaluate AboveNet as a peer. We were very much
caught by surprise when PSINet abruptly reverse course.

- PSINet's recent addition of direct private peering with several
  of C&W's transit customers relieved the peering connections
  between the networks of a couple hundred Mbps of traffic
  (improving connectivity overall and, undoubtedly, lowering costs
  for those transit customers). This is significant only because
  C&W claims PSINet no longer has sufficient traffic to justify
  the connections according to their published standards. In
  fact, PSINet's overall traffic continues to grow.

This one one of the reasons mentioned by PSINet in the disconnection
proceedings for AboveNet. Clearly AboveNet could not be an important
peer due to the lack of traffic. AboveNet has maintained since
day one of its existance a very open peering policy, and the low
traffic to PSI was very much in part due to the robust peering that
AboveNet had with many of PSI's downstreams.

In all fairness, the issue of traffic ratio also came up with PSINet
(so they don't come back and claim they never cared about traffic
levels, just ratio). The ratio is equally funny though, due to
the next point.

- Most of the PSINet traffic previously destined for sites
  behind C&W has alternative paths through other providers.
  While this sounds like a generally good thing, especially given
  the actions C&W has taken, it does make it difficult for those
  that require certain traffic levels to be maintained consistently
  for peering. Specifically, C&W's customers (or C&W itself) could
  alter "natural" traffic flow to favor (or not) various connections
  to meet their published standards (or not). PSINet demonstrated
  to C&W that if naturally less favorable announcements were
  preferred, PSINet could make an almost arbitrarily large (or
  small) amount of traffic flow between the peers. Even so, in
  C&W's opinion, PSINet will not be able to comply with their
  peering policy's traffic standards. It is gratifying to note
  that even without C&W peering, substantially all of the
  traffic previously flowing between PSINet and C&W continues to
  be delivered.

This is true for any provider with a reasonable amount of traffic.
With the number of multi-homed customers behind {Sprint, C&W, UUnet,
AT&T, Genuity, Level 3, Qwest} (and possibly others) it is easy to
move hundreds of megabits of traffic between those providers while
still honoring peering agreements. It is something PSINet ignored,
as well as C&W.

The fact is, most large ISP's can set the ration on any given
peering interface to be whatever they want, of course at the expense
of altering a ratio somewhere else. It's like a baloon, the total
volume (ratio) is fixed, but by squeezing or pulling it you can
make it into limitless shapes.

- At this time PSINet has not disabled the C&W peering interfaces
  nor decommissioned any facilities. If C&W chooses to, they can
  re-enable interfaces on their side and bring back the connectivity
  lost between their non-transit customers and PSINet. PSINet
  remains open to discuss with them a new bilateral peering
  agreement if they so choose.

How lovely. :slight_smile: Let me try from our side "At this time AboveNet
has not disabled PSINet's peering interfaces nor decommissioned
any facilities, if PSI chooses to, they can re-enable interfaces
on their side and bring back the connectivity lost..." AboveNet
is of course always open to discussing peering with PSINet, we hold
no grudges.

PSINet remains committed to servicing its customers and the Internet
with the best possible infrastructure and policies. PSINet still
maintains hundreds of peering connections with other ISPs throughout
the world. While posting about matters between PSINet and its
peering partners is not typical, the circumstances and questions
arising from C&W's decision required some clarification. Hopefully
this additional clarification helps everyone understand the current
situation.

Now this is clearly not true. As much as we had to swallow or
pride to do it, AboveNet took steps to make sure that even though
PSINet terminated peering, AboveNet customers experienced uninterrupted
(if a bit rocky for a short time) connectivity to PSINet. Our
commitment to our customers included making sure they could connect
to PSINet. PSINet clearly doesn't care if their customers can
reach C&W.

I feel no sympathy for PSINet, as I am personally a firm believer
in "connectivity is its own reward", and I think the fairly open
AboveNet peering policy reflects that as well. No one wins when
connectivity is broken like this, I have to believe both C&W and
PSI customers are feeling some real pain right now. I hope this
lesson makes people think twice about terminating such peering
relationships in the future.

That said, the irony here is amazing. PSINet has been put in the
exact same position by C&W that they decided to put AboveNet (and
possibly others, I'll let them comment on their situation) into
several months ago, and isn't even big enough to work around the
problem. "What goes around comes around" is very often true, the
provider that was big yesterday can be small tomorrow, and could
possibly even come back to be big again.

If PSINet would like to start down the road to recovery, I suggest
they start with mending some of the relationships they chose to
sever in the past. If a PSINet representative would like to talk
about the AboveNet situation in particular, please mail me privately
and I'll make sure you get in touch with the right people.

Also, I would like to suggest that if you are a Cable and Wireless
customer, and disagree with their current action of terminating
peering relationships (we have nasa.gov and psi.net confirmed on
Nanog, and rumors of many others in the works) I suggest you let
them know how you feel, first with your words, and if that doesn't
work with your dollars, by taking them elsewhere. I know a lot of
people think ISP's don't listen to their customers, and sometimes
it may seem that they don't, but by and large they do try to keep
their customer base happy.

Now, while I agree that this would provide a better path between 174 and
whoever these multi-hundred Mb/s peers are, I can also understand why C&W
might have gotten a bit miffed. That's a couple of hundred megabytes of
flow that they can no longer bill to those customers. I'd ALWAYS rather
have my customers use our network for transit than have them peer directly
with my peers and bypass the toll booth. That said, if this was their
reasoning for terminating the peering agreement, I don't agree with them
at all. I'd be upset about it if I were them but, I wouldn't be stupid
about the situation.

Christian Nielsen wrote:

PSI shut off other peering sessions. I know why and maybe you can share
this information with the rest of nanog.

If PSI really cared about their customers and the Internet as you so
write, you would have direct connections to companys like Exodus and
Abovenet.

Also, is there a isp-peering list we can move this to?

What, you think this isn't on-topic for NANOG? Seems operational to me.
Besides, I want to hear why PSI shut off the other peering sessions.

Here are some points to think about:

1) Some peers dont upgrade peering links quick enough.

If this is the case, companies go after the largest src/dest for IP.

2) Some IP providers charge for IP

If this is the case, companies go after the largest src/dest to reduce the
amount they might have to pay.

3) Some IP providers wont peer.

If this is the case, companies go after the largest src/dest first

4) and lastly, some companies have very open peering policys

And if this is the case... those companies get asked to peer all the time.

Personally, it would make much easier to peer with fewer larger players
than many little players.

but heh, this is the internet right???

BTW, to solve these types of problems, Smaller companies need to ban
together and start peering with each other in the local metro area.

Christian

Date: Tue, 5 Jun 2001 10:26:29 -0700 (PDT)
From: Christian Nielsen <cnielsen@nielsen.net>

[ snip ]

BTW, to solve these types of problems, Smaller companies need to band
together and start peering with each other in the local metro area.

Until the smaller companies get bought out by OneM^H^H^H^H a national
provider who has no interest in peering.

Quite frankly, it seems that most smaller companies have no clue what the
heck peering is or why it's good. I've approached a few, and none seem to
get it.

One (cable company, mind you) wanted to *charge more for peering than for
transit*. They thought that we would be getting some sort of "priority
access" to their network, and wanted to charge a hefty price... and this
was talking to a net admin, not a droid. Jeez.

Eddy

i have been involved with a couple small exchanges, and the way it worked for
us was "build it and they will come".

initially TorIX was only a few ISP's who happened to be in the same building.

then a few other ISP's pulled in circuits and peered.

then i bumped into an @home tech, and they jacked in.

it took 4+ years to get the momentum going, but now we have a pretty good
peering thing going on.

so, if you can at least get one other ISP to peer, pick a good place, and
drive a stake in the ground.

others may migrate towards it.

(some of the selling points are sharing usenet news feeds and/or http caches)

(broken routes to www.torix.net)

yeah, well, we are network techies, not web designers.

the www.torix.net site has been slated for overhaul for a while, but
nobody has really gotten around to it.

interesting info can be had from:
http://torix.golden.net/~torix/

> From: Christian Nielsen <cnielsen@nielsen.net>
> BTW, to solve these types of problems, Smaller companies need to band
> together and start peering with each other in the local metro area.

Hear hear. The local exchanges that are not corrupted by
their sponsors' "upsell" agenda have somehow sustained
themselves, because they solve a problem for their
participants.

Local exchanges (regional exchange points, community
internet exchanges) are becoming more critical parts of the
infrastructure. With the convergence of the metro network
(look at the growth rate of investment in the metro network
compared to the national network--that's not just for
national-network connectivity), voice moving onto data
networks (80% of voice traffic stays within the local
exchange) and the migration of new applications to the
Internet making the Internet Matter, traffic within the
metro area will become a significant, if not dominant,
segment of the overall traffic.

Until the smaller companies get bought out by a national
provider who has no interest in peering.

Quite frankly, it seems that most smaller companies have
no clue what the heck peering is or why it's good.
I've approached a few, and none seem to get it.

I've talked to a lot of companies about a lot of different
kinds of technology. I found that if you find a problem they
have that they really care about, or that you can educate
them to care about, typically they will appreciate the value
of the solution, and be motivated to get it.

Talking to a small company about peering (which they would
be unlikely to know about, since I was probably the first to
approach them) resulted in blank stares and a quick
dismissal. Talking to a company about saving money on their
transit, improving performance and reliability, diversifying
routing paths, adding value to their customers,
differentiating themselves from competition, and doing so in
a manner catered to their skill-set, resulted in most
companies at least acknowledging that the solution (peering)
was right for them, and many of them moving forward. Fitting
it into their business and financial models was a different
hurdle, but much more easily overcome when they were
helping.

As you might suspect, there are different methods to sell to
the networking guy vs. the business manager. Since the
networking guy rarely has a company check-book, I found the
most effective approach was to convince both of them, or at
a minimum the business guy.

One (cable company, mind you) wanted to *charge more for
peering than for transit*. They thought that we would
be getting some sort of "priority access" to their
network, and wanted to charge a hefty price... and this
was talking to a net admin, not a droid. Jeez.

This doesn't seem too far off from the state of peering in
general today. Isn't that what ratios and traffic levels and
geography policies are all about, enforcing that peers are
somehow "equal" and deserve to share traffic freely, and the
rest of us can pay for "priority access"?

Pete.
pete@commix.org

[ Support your neighborhood and community network projects. ]
[ Create a community Internet exchange. Ask me how. ]

we lucked out, in that 151 Front Street in Toronto is sorta like 60 hudson,
25 broadway, ??? wilshire.

so many ISP's had bandwidth going in that direction anyways.

most of the peers are connected with a simple cat5 running back to their
own closet/room in the building.

there has been some talk about putting sister switches in other buildings,etc
but that starts to get into real money, and then you have to work out how to
split it and maintain it and SLA's, etc, etc, etc.

right now the switch(es) are on loan/donated/sponsored? by cisco.
extreme and other vendors have offered their hardware as well.

the location of the first switch was picked up by one of the founding members.

we have since made an arrangement with RACO in 151 Front whereby RACO will give
the switches a place to live.

works out pretty good.

the operating expenses can be substantially reduced if the parameters are
defined appropriately.

torix is physically managed on a best-effort basis by a core group of techs
from the founding members. techs from newer members are welcome to help out.

not that there are that many techical issues.

TorIX is a switch (well, hopefully soon, two switches).

there is not alot of management, or failure points there.

the interface is cat5, 10 or 100mbit ethernet. there is some talk about
getting gigE ports added, but that doesn't make things more complicated.

the location of the switch is what is most important. it needs to be in
a reasonably neutral place, and one that lots of people can drag circuits to.

in the case of TorIX, this is in the RACO facilities at 151 front.

cat5 can be dragged from anywhere in the building (at the members expense)
or external circuits pulled in (again at the members expense).

the location of the switch is currently in a members space, but a second switch
is being installed in a RACO common area to provide more neutrality.
we are also open to putting other switches in other places. no one member
"owns" TorIX.

TorIX is not intended to be a full-bore MAE-North. at a recent meeting
some people were interested in making it more "professional" complete with
SLA's and paid-staff, etc, etc.

my comment was that they were welcome to start TorIX-II, but there was no
reason to disband or mutate the existing facilities.

for a small exchange, especially if you are looking to start with small
or medium ISP's, you need to keep the monthly costs as low as possible.

with TorIX, those monthly fees are zero. but each member bears their own costs
of getting to the switch.

OK. I'll bite.

On your 5-min aggregate graphs, you show the following numbers:

Max In: 184.8 Mb/s (6.6%) Average In: 41.9 Mb/s (1.5%) Current In: 44.1 Mb/s (1.6%)
Max Out: 84.7 Mb/s (3.0%) Average Out: 56.5 Mb/s (2.0%) Current Out: 57.2 Mb/s (2.0%)

How is it that in an exchange fabric, you're showing 13.1Mb/s difference
between what it put into the fabric and what is going out of the fabric
for this poll and 100.1Mb/s over the past 32 hours? Perhaps I'm missing
something but, it is my understanding that exchange points provide a
fabric for peers to exchange traffic. By that simple definition, when one
peer sends traffic out it's exchange point interface, it's destined for
another peer on the fabric and the numbers for IN/OUT of the fabric will
be REAL close to identical. (I understand that with 5-min polling, they
don't always jive perfectly but, I've never seen them be that far out of
whack!)

OK. I'll bite.

On your 5-min aggregate graphs, you show the following numbers:

Max In: 184.8 Mb/s (6.6%) Average In: 41.9 Mb/s (1.5%)
Current In: 44.1 Mb/s (1.6%)
Max Out: 84.7 Mb/s (3.0%) Average Out: 56.5 Mb/s (2.0%)
Current Out: 57.2 Mb/s (2.0%)

How is it that in an exchange fabric, you're showing 13.1Mb/s difference
between what it put into the fabric and what is going out of the fabric
for this poll and 100.1Mb/s over the past 32 hours? Perhaps I'm missing
something but, it is my understanding that exchange points provide a
fabric for peers to exchange traffic. By that simple definition, when one
peer sends traffic out it's exchange point interface, it's destined for
another peer on the fabric and the numbers for IN/OUT of the fabric will
be REAL close to identical. (I understand that with 5-min polling, they
don't always jive perfectly but, I've never seen them be that far out of
whack!)

I guess I can better answer that. One of the members ports wasn't being
added for the graph which has now been corrected. If you look at the
historical data, before the middle of week 21, you will see what you were
expecting. The switch and interface numbers changed during week 21 which
caused a little disruption in the graphing.

Joel