Scott:
Ratios are normally applied to either direction, since it is not
totally understood who benefits from what traffic direction. Who benefits:
the eyball or the content provider??? But keep in mind traffic
ratios are only one parameter to establish a mutially equal beneit.
Peter Jansen
C&W
I would say that both, which is why I personally feel that ratios are
misapplied in a lot of instances.
( Some providers feel that it benefits neither the eyeball or the content
provider to peer, they would like both to pay them for transit )
BTW Peter, thanks for killing our peering sessions. I guess our eyeballs
were of no benefit to your content providers? *shrug*
-Chris
Chris I'm with you this to a point. It seems to me that balance is
reasonable to a point but some of these numbers are to close. It seems
more a reasonable to peer to reduce the network distance covered and
provide a better end user experience than to use transit which may cover
greater distances and in the end use more resources. I totally
understand wanting to sell transit however and make money and some
levels this is the most logical and best practice but on others simply
because someone is not with in themagic ratio not peer. It seems to me
that many other concerns are far more important ie trouble resolution
response time, proper route policy and just solid technical ability.
I'd rather peer with someone who won't cause more work for me.
I realize none of these issues are cut and dry but it seemed odd to me.
Date: Tue, 7 May 2002 15:46:29 -0700 (PDT)
From: Scott Granados
It seems more a reasonable to peer to reduce the network
distance covered and provide a better end user experience
than to use transit which may cover greater distances and in
the end use more resources.
But it's much more fun to bill your customers for one direction,
and another network for the other.
*flame suit on*
The discussion is age old, yet opinions still vary...
This makes for some great logic. If you really believe that traffic in
either direction can be equally beneficial, then why require ratios at
all? If on the other hand, you believe that content is less valuable than
eyeballs, wouldn't eyeball providers be the most valuable of peers? Except
in the case of mismanagement (such as a congested peer), a peer benefits
everyone. Why does it matter that a peer benefit both sides in exactly the
same ways?
Yes there are legitimate arguments in the favor of not accepting smaller
peers. If they're all eyeballs and only in one location you have to haul
traffic there that you otherwise would have dropped locally on one of the
bigger peers that they buy transit from. But if they can meet the
locations, I don't see a legitimate argument for ratios. Perhaps what you
need to do is consider distance to the egress point above AS Path length.
Then we comes to those little things that are just there to try and keep
people from qualifying to peer. You can't be serious about requiring 5000
routes can you? Way to encourage aggregation, really.
When it comes down to it, someone on your network has PAID YOU to BRING
them traffic as well as to deliver it. If you can't do that then I'm sure
they can find someone who can. As for the "if they can't peer, they'll buy
transit" argument, I find that equally negated by the "if they won't peer,
why should I buy their transit" argument.
Richard,
I believe you also missed
must operate a US-wide OC48 network.
must exchange at 4 locations over OC3 or above with at least 45Mb traffic
per location
and most friendly of all, you must supply a detailed network topology and
current operational capacities.. why not ask for 5 year business plan and
bank numbers too .. and how about next weeks lottery numbers?
Steve
In a message written on Tue, May 07, 2002 at 05:50:00PM -0400, PETER JANSEN wrote:
the eyball or the content provider??? But keep in mind traffic
ratios are only one parameter to establish a mutially equal beneit.
What I find unfortunate about most of the published peering policies
is that they don't seem to take the spirit of your statement to
heart. Most of the published peering requirements are absolute.
These are the N areas we find important, and our definition of all
of them, please meet them all or you get nothing. What these
requirements are doing are forcing a /business model/, in effect
increasing your own competition.
Imagine two providers. One is a 100% content hosting play, the
other is a 100% end user access play. In terms of ratio it will
be high (10:1?), as all the content flows to the users. Neither
network would be of any value without the other, and I would argue
them peering is a perfect symbiotic relationship.
What a peering ratio like 1.5:1 does is require them to compete.
The hoster must go out and find end users, and the access provider
must find some content to host. They start going after each others
customers, and a price war ensues hurting both companies. The way
providers insure common ratios is to insure they have similar,
often overlapping customer bases.
Note, none of this has anything to do with geography or cost. In
my two provider network you could give either one the nationwide
network, and make the other the small regional guy. The "larger
cost" could fall on either network.
My point is not that ratio shouldn't be used, but that it shouldn't
be used in all cases. Perhaps if you have 10 criteria to evaluate
a peer, it would be more reasonable to require them to meet 9 of
the 10, or 8 of the 10. Allow for the fact that networks are
different. Don't try to make every network look like your own,
you create more competition for yourself in the end.
Here is a good point. Also something to think about. Recently I read
the argument that ip space is sometimes used up by an isp by attaching
it all to old machines that will answer requests to justify new arin
space. Its a total improper use of the space but it is done to meet
requirements. Someone in this case could create false traffic either
pull or push probably most likely pull and meet the ratio but cause
drastic amounts of useless traffic to be carried on the backbone. This
does create far worse problems. I'd assume such a method could be
determined if not done properly but still it seems that these policies
whether relating to ip allocation or ratio of traffic leads to more harm
than good.
Richard,
I believe you also missedmust operate a US-wide OC48 network.
Personally I would go with "must operate a network with sufficient
capacity to support the traffic being exchanged". Unless I was selling
US-wide OC48s of course.
must exchange at 4 locations over OC3 or above with at least 45Mb traffic
per location
Not an entirely unreasonable goal. But then we come to bizaare ones like:
D. The applicant shall take steps to ensure that its routes are not
announced to Cable & Wireless from another network.
What exactly is this supposed to accomplish?
and most friendly of all, you must supply a detailed network topology and
current operational capacities.. why not ask for 5 year business plan and
bank numbers too .. and how about next weeks lottery numbers?
I don't suppose they'd take too kindly to an ascii diagram which just
happens to resemble a middle finger, would they?
Oh BTW on the subject of peering, has anyone noticed that AOL has cut off
a large number of transit providers and reportedly a number of content
hoster peers (though I havn't seen this first-hand) in recent days. I
guess when you have the largest eyeball population your only remaining
goal is to have the largest content population too. Something to think
about.
Date: Tue, 7 May 2002 17:28:40 -0700 (PDT)
From: Scott Granados
Someone in this case could create false traffic either pull
or push probably most likely pull and meet the ratio but
It almost encourages those with eyeballs to spam, no? Maybe
that's how [names deleted] meet their ratios... adjust how
quickly they sign up and kick spammers. 
cause drastic amounts of useless traffic to be carried on
the backbone. This does create far worse problems. I'd
assume such a method could be determined if not done properly
You mean, like someone binding many IPs across geographically
diverse locations and replaying cache traces? (Pretty sick when
interconnects are reduced to the same level as people cheating
those make-money-with-each-site-you-visit programs...)
but still it seems that these policies whether relating to ip
allocation or ratio of traffic leads to more harm than good.
Yup.
Date: Tue, 7 May 2002 20:42:14 -0400
From: Richard A Steenbergen
D. The applicant shall take steps to ensure that its
routes are not announced to Cable & Wireless from
another network.What exactly is this supposed to accomplish?
IANAL, but that doesn't smell right. Seems sort of... anti-
competitive...
I guess if one is trying to meet the traffic ratio, one could
reroute certain traffic through an intermediary. Maybe some
nets tried this, met the minimum/geo/ratio requirements, and
there just _had_ to be another way to depeer?
Or maybe they want to peer with ASNs who don't purchase transit?
Perhaps they want the ability to quickly *coughpsilastsummer*
depeer *coughexoduspeersrecently* anyone as they please?
(Can't think of anything more serious at the moment.)
I guess people must vote with wallet and local-pref?
I've joked that we'll peer with anyone who hooks me up with a
hot, geeky gal. Now I begin to wonder just how kooky that idea
is.
I don't suppose they'd take too kindly to an ascii diagram
which just happens to resemble a middle finger, would
they?
"Hey, what are all these packets with
.|..
`
in the header?!"
Just wait until IPv6 when people forge source packets with octets
that spell obscene messages when interpretted as ASCII. Or even
as EBCDIC if 5|<r1pt |<1dd135 get interested in history.
Once upon a time, Richard A Steenbergen <ras@e-gerbil.net> said:
D. The applicant shall take steps to ensure that its routes are not
announced to Cable & Wireless from another network.What exactly is this supposed to accomplish?
Doesn't that mean that, if you had a failure peering with C&W, you'd be
cut off (even if you were buying transit from someone else like UUNet
for example)? Seems like that would give C&W extra leverage in a
peering dispute, since, by the peering agreement, your C&W peering
connections would be your _only_ connections to the C&W network. You
can't buy transit that _might_ get you to C&W's network if you want to
peer with them.
Oh BTW on the subject of peering, has anyone noticed that AOL has cut off
a large number of transit providers and reportedly a number of content
hoster peers (though I havn't seen this first-hand) in recent days. I
We recently received an email from AOL with the Subject "AOL email
concerns for hiwaay.net". It had some vague statistics:
Total percentage of messages bounced: 0
Total percentage of bounces accepted: 45%
Total number of AOL member complaints: 0
It then went on to say that if we didn't respond in 24 hours, we may be
blocked from AOL. I responded asking for more information, but never
heard back (and they didn't block us). I didn't see many AOL bounces in
our logs, and I'm not sure what "percentage of bounces accepted" means
(especially when the percentage of messages bounced is zero). Anybody
know what they mean by this?
Atlantic.net got the same thing. I sent a similar "WTF are you talking
about", though not worded quite like that, message in reply. It took them
a week and a half to reply to my reply. Their reply was almost as vague
as the original message. What I got from it is they are looking at
automating spam reporting and want us to setup or simply provide them with
an address they can configure their system to automatically forward
complaints to. Why they couldn't just say that in the first message
rather than sounding all hostile is beyond me. Maybe they've got a new
abuse dept. director who's got a serious drug problem.
They also completely ignored the message I included with my reply
mentioning their netscape.com webmail spam issue and asking what they
could do to stop that abuse.
Has anyone else tried capturing copies of mail with:
:0c:
* ^Received: from .*mx\.aol\.com
* ^From:.*netscape\.net
* ^Received: from netscape\.com .*webmail\.aol\.com
netscapespam
to see what you get? I did it for several days and got >90% spam...50
megabytes of it.
> must exchange at 4 locations over OC3 or above with at least 45Mb traffic
> per locationNot an entirely unreasonable goal. But then we come to bizaare ones like:
D. The applicant shall take steps to ensure that its routes are not
announced to Cable & Wireless from another network.What exactly is this supposed to accomplish?
I can only assume it means that CW consider themselves a Tier1 and they
will only accept routes from other peering only networks ie Tier1.
This always interests me, the kind of unofficial rules that other networks
striving for global superiority all seem to adopt. Surely if a group of
operators agree together that they will run the market and all other
players will buy from them then that cant be legal.. I guess its all
slightly too loose tho...
The opposite rule to this from local providers would be that you set your
BGP route maps to prefer any routes that dont go over a large network (CW)
.. thereby feeding the other more friendly networks (to take another
example UU are arguably the largest network and yet they will peer with
anyone regionally doing a fairly small amount of traffic...)
Whilst doing this may be a little militant you probably dont actually need
to.. not peering must reduce total traffic anyway: on the basis that you
arent a CW customer and chances are the routes you have installed to your
destination dont go over CW's network then they dont get the traffic
anyway.. and the other networks who do peer win.
Steve
UU are arguably the largest network and yet they will peer with
^^^^^^^
> anyone regionally doing a fairly small amount of traffic...)
You misspelled "because".
-Bill
.. altho I missed this from http://www.cw.com/th_11.asp?ID=us_10
"Our backbone isn't just a route to the Internet, it is the Internet."
mine too!
> D. The applicant shall take steps to ensure that its routes are not
> announced to Cable & Wireless from another network.
>
> What exactly is this supposed to accomplish?I can only assume it means that CW consider themselves a Tier1 and they
will only accept routes from other peering only networks ie Tier1.
I'm still trying to figure out if that's what they mean or not. My first
interpretation would be that you just have to have a transit provider who
will let you filter announcements to certain destination. Yeah they would
have to be a tier 1 or have upstreams who would also honor "don't
announce", but thats not particularly hard to accomplish.
If you take their statement at face value, that either don't know how to
use localpref, or they don't like path redundancy. If you try to look for
hidden meaning, you come away thinking that they want you to be a tier 1,
and just couldn't find a better way to state it.
If there is a more reasonable interpretation, I must be missing it, so
perhaps someone can fill me in.
This always interests me, the kind of unofficial rules that other networks
striving for global superiority all seem to adopt. Surely if a group of
operators agree together that they will run the market and all other
players will buy from them then that cant be legal.. I guess its all
slightly too loose tho...
Wasn't the reason all these "big networks" started coming out with actual
peering policies at all to keep the DOJ from going down that exact road?
The opposite rule to this from local providers would be that you set
your BGP route maps to prefer any routes that dont go over a large
network (CW) .. thereby feeding the other more friendly networks (to
take another example UU are arguably the largest network and yet they
will peer with anyone regionally doing a fairly small amount of
traffic...)
Depending on your ratios, this may or may not be a good thing. But
honestly, I'd say that the traffic exchanged with CW by the people who
actually do peer is going down steadily, and it may not be worth your
trouble at all. With a open peering policy to anyone you can reasonably
reach, traffic to 3561 is easily 1/10th of traffic to 701.
I misread this as "minimum/ego/ratio", and suddenly it made a LOT more sense.
Date: Wed, 8 May 2002 11:21:11 -0400
From: Richard A Steenbergen
If you take their statement at face value, that either don't
know how to use localpref, or they don't like path
redundancy. If you try to look for hidden meaning, you come
away thinking that they want you to be a tier 1, and just
couldn't find a better way to state it.
I find it hard to believe that they wouldn't grok local-pref.
Considering all the nice knobs that they provide... I'm sure they
have enough clue to give downstreams higher local-pref than peers
unless there's some special reason for otherwise.