How Reliable is the Trans-Atlantic Link?

bdamm asks: "We are looking at doing a Web application that will be rolled out internationally. Currently we are looking at the feasibility of centralising our application, or building a larger system that can allows for multiple hubs separated by slow connections, such as the Trans-Atlantic link. My question is, does anyone know just how good the Trans-Atlantic link and other major non-U.S. backbones are? Is it fast enough to allow everything to be in one place, or just good enough to have HTML generation occur centrally with satellite image servers?"

Comments

Depends

[bluGill]

I live in Minneapolis From work my connection to most well connected sites in sweden is better then my connection to any site outside of the Minneapolis area. (My company might have a leased line to Sweden for internal use - I don't know our internal topography)

One of the tenets of IP is you don't know the path a packet takes. It normally happens that successive packets take the same route, but there is nothing innheirant in IP that says they have to. (routing protocols are just a lot easier to design when they send them the same way) It used to happen that packets that only physically need to go 70 miles, from the UK to France would travel over a thousand, through the US. This may or may not have changed.

You can be sure that the optimal routing will change in the future. If you build a big data center anywhere don't be surprized if someone else provides cheaper bandwidth, and everyone switches to them but you are left hanging dry in a physically bad location for that cheaper serivce.

Network Topology & stuff.

[AliasTheRoot]

transatlantically there is something called gemini (2 cables, geddit?) which drops off at two locations in the uk. Gemini is jointly owned between a bunch of telco's (wcom, c&w...) & some isps (easynet...). Each of the cables, carries *many* fibres, each cable carries (i think) 622mbit. transatlantic connections tend to originate at washington or new york

wcom / uunet and most of the other telco's (level 3, colt...) operate large ring networks, with similar capacity (or greater) than the transatlantic capacity, these networks typically pass through the central business hubs of the eu (london, berlin, frankfurt, amsterdam & paris). They also tend to have connections over to stockholm, barcelona, milan, roma etc.

peering arrangements are typically made at internet exchanges, for example Linx. All of the major players peer here (often not with each other... :), and depending on your isp, you may or may not be directly peered (otherwise your traffic is routed transatlantically).

due to economics, transatlantic bandwidth tends to be cheaper than european, although this *is* changing, level 3 for instance charges less if traffic is just going across europe - probably to encourage take up of capacity.

finally, the uk is not the only drop off point in europe, i think there are similar connections in place in sweden, holland, germany & france.

if you want hard figures and network maps, suggest you look at:

www.uu.net
www.level3.com
www.colt.com

and also track down the various internet exchanges at key points around europe.

I know less about asia pac, connectivity tends to come in to hong kong, australia and japan, and i think those countries are directly connected to each other.

1 point to note, in many countries you will have to build datacentres locally / have some kind of operations in place. It's a legal not technical requirement. We're based in the uk, but when we launch in germany we'll have to deploy a seperate datacentre there.

Something interesting about European access...

[trims]

Ok, I don't have independent verification of this, but from my talks with several of the big bandwidth providers, and others, there is an interesting problem about providing access for European consumers....

Logically, it would seem to make sense to open a UK (or German, or French) farm to serve EC people, but it turns out this isn't necessarily the case.

The problem seems to be the peering arrangements in the EC are bad-to-non-existant. As far as response time goes, this seems to be the priority of EC network providers:

1. traffic to their own customers
2. traffic back to the US backbone folks
3. traffic to other providers in the EC.

I've heard that the last is considerably below the other two. So the predicament becomes that opening a datacenter in the EC really only buys you better connectivity to people who are directly connected to that datacenter. Oops.

What alot of people have recommended to me is this: Open a datacenter first on the US West coast, then on the US East Coast. The Westie will serve Asia and most of the US, while the Eastie will cover the Eastern seaboard and the E.C. Only after you've done that should you look at putting a datacenter into the E.C., and then consider which PROVIDER (not which Country) has the most demand on it.

Stuff is funky, isn't it?

-Erik