On 7/1/04 8:14 PM, "Bill Woodcock" <woody@pch.net> wrote:

...

I have a question regarding information on my ISP's peering relationships. Are the speeds of some or all peering relationships public knowledge, and if so, where can I find this? By speed, I mean bandwidth (DS3, OC3, 100Mbps, 1Gbps, etc.). I am trying to transfer large stuff from my AS, through my ISP, through another ISP, to another AS, and I'm wondering how fast the peering point is between the ISPs.

ISPs don't register it or publish it anywhere, generally, and if you ask salescritters, they're likely to say "At the speed of LIGHT!!!" or some such. But the two methods people would generally use to determine this externally would be first to do a bidirectional traceroute and look closely at the in-addr DNS names associated with the router interfaces on the IX or crossconnect, which, if you trust them, may give you some indication of speed. Next, if you have time, you could run pathchar across the link.

-Bill

Of course, the big issue isn't the size of the links - its utilization. Most private peering links today are OC-12 to OC-192. Most big ISPs do this in a half dozen locations - sometimes more, occasionally less. Of course, you'll use the closest exit between ISPs. Ask your ISP what the utilization of that link is - or what the packet loss is, historically. They are much more likely to tell you that. Its funny, you always see people asking about peering link sizes or locations on RFP's, but they never ask about peering utilization or packet loss. The former is both NDA and meaningless - the latter is terribly important. -- Daniel Golding Network and Telecommunications Strategies Burton Group

On Thu, Jul 01, 2004 at 05:14:10PM -0700, Bill Woodcock wrote: > I have a question regarding information on my ISP's peering relationships. > Are the speeds of some or all peering relationships public knowledge, and if > so, where can I find this? By speed, I mean bandwidth (DS3, OC3, 100Mbps, > 1Gbps, etc.). I am trying to transfer large stuff from my AS, through my > ISP, through another ISP, to another AS, and I'm wondering how fast the > peering point is between the ISPs. ISPs don't register it or publish it anywhere, generally, and if you ask salescritters, they're likely to say "At the speed of LIGHT!!!" or some such. But the two methods people would generally use to determine this externally would be first to do a bidirectional traceroute and look closely at the in-addr DNS names associated with the router interfaces on the IX or crossconnect, which, if you trust them, may give you some indication of speed. Next, if you have time, you could run pathchar across the link. -Bill um, pathchar will attempt to do a capacity-per-link over a whole forward path, but it struggles with the complexities of real world infrastructure (alas, who doesn't) after about 1997, when it was last supported. van let caida index the tool at http://www.caida.org/tools/utilities/others/pathchar/ it is 100% van's knightly work. but i think van moved on to other things after concluding that the task of accurately identifying capacities of links many hops away via layer 3 measurement on the global Internet was 'sub-possible.' a few others agree w him. professor constantine dovrolis and his amazing grad students ravi and manish at GA tech have taken on a slightly smaller windmill: measurement of capacity and available bandwidth for a given -path- (which means the narrow and tight links along the path, see http://www.pathrate.org for excellent brief description and source codes). note that neither tool requires root privileges but both require installing software at both ends of the path, which render them less useful for the specific application above. see constantine's IEEE Network paper http://www.cc.gatech.edu/fac/Constantinos.Dovrolis/Papers/NetDov0248.pdf if you want high signal techie survey paper on the field, including a table listing a bunch of tools on page 12. caida has a summer student from UNC this year working on evaluating the accuracy of some of these tools in a controlled lab setting. i personally can't believe you knights put up with not being able to measure capacity and available bandwidth of your peers' paths, but i guess it puts the anycast drama into perspective. :) the bottom line is that the research community has made some progress in the field in the last 5 years, but it's a harder problem than it seems. and it seems pretty bloody hard. as an operator the best thing you can do to advance this particular research ball is to use these tools on paths where you actually know the available bandwidths, and give the tool authors feedback to help make them more accurate or otherwise more conducive to your needs. it's the single largest barrier standing between these tools and these tools being more useful to you. (besides funding and strong-stomached coders of course) </brokenrecord> k