Slashdot Mirror

wut?

Netmask Flaws - I guess you are talking about cidr? Not really sure. You say netmask which implies that you may actually be talking about a broadcast domain; but the rest of your comment speaks to an allocation. eg: If your ISP has a /8 then you will not be using 255.0.0.0 for your netmask - and if you are then run away from them as quickly as possible.

Routing Table Solution/Latency and Hop Count Not Enough - This _IS_ published data. This information is in BGP and you may query many different databases (route-views, bgplay, et al). If your ISP does not have a looking glass then you should stir the waters until you find a clueful network engineer that can hook you up. You may find remote nodes that are within 5 hops by simply looking at the TTL, which is very very simple math.
 
- And, ohhh, come on now. Companies like Internap, and others, have products that do this for policy based routing. These network appliances, such as the FCP, send out hundreds of thousands of packets per second. I do not think that you really want to be doing this. :)

Just mirror the Routing Registries at a central site, create a sqlite database and distribute daily via some sort of distributed method(heh) for a very broad policy, pad on TTL, pad on latency(adjusting over time), attempt to pad on ptr rr(lower metric), do the hokey pokey...

Oh, well, hell.

True, there's no escaping the speed of light. The biggest issue, however, isn't latency itself; it's the fact that TCP scales badly across high-latency connections - the transmitting host will send packets up to the maximum window size, and then spend so much time waiting for the ACK to come back such that big pipes don't get efficiently utilized. There are solutions available, both hardware-based and service-based, however, which proxy the TCP transmission for maximum throughput no matter what the latency is.

There are bigger players on the internet than the telcos. If the telcos try to move to a tiered program, that will only open the doors wider for independent ISPs with pipes running to companies like Level 3 and Internap. The only result of this will be better connectivity. The telcos are looking after their bottom line, and providing good service to Podunk isn't on their agenda (the gains would be a drop in the bucket to them). On the other hand, the opposite is true for a small town ISP with a nice big pipe. People will pay for better service and the telcos will be stuck with laying (and maintaining) copper for it.

On the flip side, they spent the money for the network, they can do whatever the hell they like with it. If you have a problem with that, go to a non-capitalistic country.

This is already out there. Have a look at http://internap.com/

There are ISPs that have no peering arrangements, they purchase all of their bandwidth to avoid situations like these (among other benefits).

Internap is one of them. Since I work for them, I'll end the sales pitch now.

INANE (I'm Not A Network Engineer) but, as I understand it, peering has been a problem for awhile. Even if network providers do peer, as a customer your quality of service suffers. If your packets are sloughed off to some other provider's network and then there are problems downstream, for example, that provider has little incentive to help you since you're not one of their paying customers. Or something like that.

In any event, that's why companies like InterNAP offer multi-homed services.

INANE (I'm Not A Network Engineer) but, as I understand it, peering has been a problem for awhile. Even if network providers do peer, as a customer your quality of service suffers. If your packets are sloughed off to some other provider's network and then there are problems downstream, for example, that provider has little incentive to help you since you're not one of their paying customers. Or something like that.

In any event, that's why companies like InterNAP offer multi-homed services.

I'm surprised to see that Internap's main servers are back up. It's pretty irresponsible to bring up your corporate servers before those of your clients.

Not really. Internap has multiple data centers, so it's likely that the two sites are hosted in two different locations. LJ is based in Seattle, so I assume their servers are hosted there, while Internap's corporate HQ is in Atlanta, where they have another facility.

Compare traceroutes to www.livejournal.com and www.internap.com. They're on opposite ends of the country.

Check out this page on the Iternap site for a real laugh. The flash page is a real hoot too.

Check out this page on the Iternap site for a real laugh. The flash page is a real hoot too.

Every large network provider has a spam problem. Check UUNet, Qwest, etc. Also check a few other clients of Internap. [Disclaimer - I know people at Internap.]

Internap has colo space and excellent connectivity.

I spoke to a sales weasel from Internap and they claim to be able to get around/stop/put an end to/whatever DDoS attacks without that sort of invasion of privacy.

What he said they do is, rather than lease their lines from one backbone provider like Sprint or Genuity/BBN, they lease from 12 major providers. His claim was to be able to shut a customer who was being attacked off from one backbone and re-route all traffic to and through another faster than a DDoS attacker could shift gears.

Anyone have any experience with this company? Was this cat just blowing his own horn?

Their AUP is pretty vague and restrictive. For example: "Customer shall not use, nor shall it permit others to use, the Services for any ... immoral ... purpose." So porn sites cannot use Internap? ISPs cannot use InterNAP beacuse their users may surf porn?

Also: "Customer shall not use, nor shall it permit others to use, the Services ... to alter ... [or] disable any security or encryption of any computer file, database or network. Um, huh? Even if it is legal? Bewarned sysadmins: if you use InterNAP do not ssh over to your remote data center and run crack on one of your machine's passwords files to check security. Even using tcpdump to troubleshoot network snafus may be a violation of the AUP.

I'm sure all providers have such lame policies but that doesn't mean they have to have one too, especially since they seem to claim to be such and enlightened Linux-luvin' company.

I'm kinda backwoods in terms of routing distance from anywhere, and that site's always responsive, it seems.

"Utilizing our proprietary network architecture and advanced routing technologies, we are able to route, or transfer, data to and from businesses that are connected to one or more of our P-NAPs, in a manner that minimizes the use of congested public exchanges and private peering points. This optimal routing of data traffic over the multiplicity of networks that comprise the Internet enables higher transmission speeds, lower instances of data packet loss and greater quality of service."

Having a look at their web site this page mentions that "* The P-NAP, and its ASsimilator technology are proprietary and patent-pending " - so we really don't want these guys to be the future of the internet. No matter how good their technology is, if they are moving such basic internet standards to proprietry forms, they are not a company I feel people should support.

tangent - art and creation are a higher purpose

I appreciate the insightful commentary for once...

However, I would like to point out something: I'm not saying that anyone is doing something they shouldn't, I am saying that the public peering model is fundamentally broken, and that what we currently have results in asymetric routing and high packet loss at peak times. Why should traffic hit these bottlenecks if it isn't necessary? Why not use private peerings with multiple backbones? Extra hops are introduced, and eficciency is lost. That's why I think what they are doing over at InterNAP is so cool: they want to re-architect the system so that all participants receive maximum benefit. I'm not a nutcase, and I know I keep bringing them into this discussion, but if you ever hear their CTO talk, you'll get chills at the ideas they work with....

These large backbone providers are essentially advertising (Qwest, for example) "We have the largest, fastest backbone in the world!", but who cares? If you get dumped at the competitors network immediately, you spend all your time in transit as a freeloader on someone else's network. BB prov's don't pay each other for this transit, so it gets lower priority on everyone else's network. That's not a good system, because participants have competing agendas. ...

The Public Exchanges? MAE-East and MAE-West (metropolitan Area Exchange) handle 70-80% of all traffic that goes from one backbone to another., and are paid for by the government. It's the one place where everyone can peer for free, so it is where the smaller backbone providers peer at, so they don't have to pay exhorbitant rates to UUNET or the other fatties...

InterNAP is getting it right, and is beginning to get noticed around here and elsewhere. Check out their web page for a good explanation. They provide to some of the biggest blue-chips in the country, including M$ and Amazon. They have a different way of doing this stuff that solves the public peering problem.

Now hold on a second, there.

There's a very underreported problem with the way that peering is set up right now on the network. "Peering", as it has been, is only going to become worse and worse, because of the kludges we have had to provide to keep things just as they are aren't enough anymore. 90% of all routed traffic going from one backbone to another goes through one of the MAE exchanges, or in other words, through a public access node that is paid for with tax dollars alone. This being the case, those facilities are massively insufficient for the load they bear, which is why on peak times, you can get 20-30% packet loss. This is absolutely unaceptable. And the worse of it is, the more loss, the more traffic, as the TCP protocol simply has hosts re-send what didn't get through...

Some people in the high up networking world are getting a clue, and have actually solved the problem. For more information, a better and more thourough explanation, and more technical details on the solution, go to this site .

To continue, the tier1 providers refuse to peer with smaller ones, as the article suggests. UUNET gets pretty much whatever it wants because they own 30% of all the destination ip-space. What the big guys do is dump your traffic off at the local public exchange points at the soonest opportunity, making their competitors carry the load. But of course, since they all do this, everyone gets a worse QOS unless they just happen to be going to someone located on their own backbone.

Thats not good, and that is what I think this article is (rightfully) pointing out.