Poor Internet connections and peering agreements

[Mr Chunder]

As I understand it, the internet is constructed of many companies carrying traffic who have peering agreements with other such companies i.e. "I'll let your traffic pass across my network if you let mine pass across yours".

This is fine but don't you really notice poor network connections ? When a webpage appears to lock up or your mass data transfer across the Atlantic fails, do you study the traceroute of the different providers between you and your destination and notice it gets stuck with a particular provider on the way ?

Have you noticed that poor performance or timeouts are not necessarily to do with a poor server - it could be poor network connectivity.

I have actually found lots of problem areas on the internet and could name a few key providers who always seem to be responsible for hold-up to my regular sites.

I believe the problem is that because of peering agreements where no money changes hands and therefore, proper SLAs are not really in place.

In this mass network of peers, it seems that you can only hassle the immediate network you are connected and not any one after. So, if you have a problem with a provider two networks down the road, there's nothing you can do unless you phone one network and complain and get them to complain to the next network and so on.

This is a bit of a sore point with me but why can't traffic carriers either negotiate really strict SLAs with each other with opportunities for end users like me to report faults ?

Also, when there is a fault, all traffic arrives at the faulty router and then gets stuck. How come these things aren't dual redundant or on hot standby ?

If a governing body or all providers as a whole could take this on board, it would so much better for the internet industry as a whole.

[Jaiem]

The thing about the net is that it is not as distributed as people think. That is, everyone is talk the net is a generic web of connections. If one part goes down the net just routes around it.

Not always so.

Even though much is distributed there are certain main arteries and pipelines that if they went down or have trouble the whole net boggs down.

[Mr Chunder]

Ok, so what is the industry going to do about it - it is in all our interests (i.e. all online companies) to ensure that we keep pressure on your net connectivity suppliers.

I wonder many people bother doing this ? All too often, we just simply accept poor net connections rather than find the guilty parties and ensure that they are well and truly ticked off for runnning a poor service.

While people do not do this and attribute network failure to a particular site being "down" when in fact, it was to do with a poor net connection, then the networking companies can get away with offering poor services and dubious peering agreements.

We all want to see continued network growth but we do not want it to be known as the "world wide wait" or somewhere where you can do business and leisure 24/7, providing "the site is up". We do not want to see the internet outgrow its infrastructure.

[Jaiem]

The problem is good interconnectivity hardware costs. The further off the main stream of the net connections you go the slower it can get because it's smaller companies that provide the connections. They can't afford upgrades to the levels we'd like to have (not without raising prices to their customers).

And Lord knows we don't need any more government subsidies!

[JTY]

In many cases the net does route around.

A while back one my ISPs peers had an outage and the routers didn't route around so, I wasn't able to connect to large number of servers in California.

[Joolz]

Most routing protocols handle outages. Hell, in the UK if the main backbone goes down there's an emergency ISDN connection (yes you heard me) to the states and all ISPs can go through that.

There is a lot of redunancy. Yes, of course things slow down when things go a little pear. Really, that's the same with any internetworking 'network'. The thing to remember is, remove a backbone, it will ALL still carry on to work.

[Mr Chunder]

Jeez! I don't not believe that there is an ISDN backup between the UK and US of A :eek:

We may as well send morse code or something. :(

Anyway, the fact still remains that not enough effort is placed on making the internet more resilient. I do not know enough about TCPIP to know if it is a routable protocol but I'm off to read up on it. :)

[JTY]

TCP/IP is routable..... and there are a number of protocols that manage the routing data: OSPF, BGP4, and RIP. I am probably missing some others though.

[Mr Chunder]

Ok, so it is routable, then why is it proving to be so difficult for ISPs to route round failed nodes ?

If you visit a fixed place on the web, chances are that you take a fixed route through it i.e. a tracert also shows that I am taking the same route across the internet 95% of the time (there are differences other times). However, when there is a fault on a certain network, traffic is still sent to that same faulty node until someone calls them and complains.

I would expect that an automatic failover would be setup for these cases and that it would be very commonly available. My ISP claim that this isn't case with many providers - all their routes are fixed so they have human intervention when things go wrong.

Not good enough ! :shake:

[Phoenix]

Originally posted by Mr Chunder
Ok, so it is routable, then why is it proving to be so difficult for ISPs to route round failed nodes ?

I would expect that an automatic failover would be setup for these cases and that it would be very commonly available. My ISP claim that this isn't case with many providers - all their routes are fixed so they have human intervention when things go wrong.

Not good enough ! :shake:

You are right, it isn't good enough. Get a new ISP.

Your ISP has not kept up with the times and doesn't know what they are doing-and unfortunately that is the case with many providers. Using fixed routes is cheap and don't require using scary routing technology like OSPF and BGP4, and it doesn't require having a network engineer that knows how to configure and upgrade carrier-grade routers (another ugly expense).

Your ISP's local backbone network should be multi-homed so the traffic is dynamically routed (i. e. not requiring human intervention to route around a problem) over a number of different national backbone networks, as well as public peering points. Your ISP should also have globally routable IP numbers as this is required for being multi-homed.

[Mr Chunder]

Originally posted by Phoenix


You are right, it isn't good enough. Get a new ISP.



Thanks Phoenix, you're right about multi-homing etc. and I would like to get a new ISP but locked in on a year contract :cry:

However, it isn't the immediate ISP who causes the problem, its two ones further on and that my point about peering agreements being useless :mad:

I suspect that my ISP is a lousy outfit anyway and plugged into the internet via the cheapest possible route, and their peers did the same and so on so its crap from here to my destination !

I am trying to use another ISP where they own most of the network between me and my usual destination. I've done some tests and there's much less hops (half the number) and below 100ms all the way (according to my beloved tracert ). This has to be the way forward.

:cool:

[JTY]

Well sometimes route propagation takes time..... and some networks have poorly configured routing tables...