How to choose a transit provider?

Hello there,

I have started writing a blog which I hope it would help buy transit services from providers by doing various due diligences(technical) i wanted to reach out and ask nanog community’s thoughts on this.

What are some of your checklist items ? Price? Their directly peered networks? If they are tier 2,3 who they use as tier 1-2? Are the onnet? I am sure list goes on and on on…

Thanks a lot for your help. I plan to write the blog this month and publish.


Hi Mehmet,

We usually ask the sales director from a neutral datacenter to introduce a sales rep from Tier 1 - 2 ISPs to bargain.

First of all, sign NDA if possible, then ask the following questions:

  1. Price for 100 Mbps on 1 Gbps port to 1 Gbps unmetered or 1 Gbps on 10 Gbps or 10 Gbps unmetered

  2. Contact term, from 12 months to 36 months, or even 60 months.

  3. BGP community or RTBH for blackhole

  4. AS-SET or LOA for BGP filter updating

  5. SLA and network delay (latency) guarantee

  6. Price for NRC and MRC and VAT or tax in some countries

For their peering networks and IX, you can do research on different network using looking glass, mtr, traceroute, etc. And ask your friend if they are already using the service.

Cheapest transit service will not always have good performance, but the most expensive one may not be the best choice. You should choose the suitable provider for your audience.




This depends a lot of who you are and where you are. For example apparently Cogent is better in the USA compared to Europe. This would make them mostly useful in Europe only if you have the traffic to be multi homed, while someone in USA might be able to use them as their only provider.

If you are going to have only one provider, I would recommend to stay away from the so called Tier 1 providers. You want a smaller local provider, which has multiple upstreams and at least some local peering. Sometimes the tier 1 can get you the best quote and their sales people will certainly tell you all about their superior network and how many global connected customers. But more often than not, the interconnect between the various tier 1 providers is not good and you end up with bad connectivity to whoever they are at war with at the moment.

If you have enough traffic to justify multiple upstreams, you can do the tier 1 game. But you still have to be careful to have good local peering. At least if your customers are close to your own physical location. In my country there are several of the big american transit providers. They only have good connectivity to other local companies, that happens to also buy directly from the same transit provider. The tier 1 will refuse to peer with just about anyone and this makes their local connectivity poor.

Also consider the wildcard called They are the opposite to the old tier 1 in that peers with everyone locally. On the other hand, their global network might not be as good (although my experience is that they are pretty good). I am using as an alternative to joining the too expensive local internet exchanges. It is cheaper to get and will be able to get all the peerings that I can’t.

Another interesting player is NL-IX. I know this is an european thing. I believe their concept could spread. They take distributed IX to the next level with a IX network that covers large part of Europe.

If you are an eyeball ISP you also need to consider caches and direct peerings with the big content providers. Akamai, Google, Netflix, Apple, Microsoft etc. If you are hosting provider, those same peerings are completely irrelevant.



Tier 1s are just as succeptible to outages and peering issues as anyone else. Not to say they’re any less, I work for one after all, but one shouldn’t assume they’re always the best for every application. As an example, Hurricane is decidedly not a Tier 1, but have one of the best peered networks out there.

Well peered is a huge plus but that’s hard to measure. Cogent, peers with Google in just a few spots so if you want to get to from Dallas you’re going to go via Atlanta even though they could peer right in TX. That’s a bummer if you’ve hardcoded Google DNS into anything. But how would you know that unless you do a lot of testing with looking glasses?

The choice also depends on what you’re doing with the bandwidth:

If you’re a content provider, for example, you may want to buy transit from AT&T, Comcast, or Charter, not because they’re the best, but because they have better access to the eyeballs. Voice guys may want a “performance optimized” blendwidth for lower latency. Etc.


Why in heaven's name would you *want* to sign an NDA? Aren't you better
off without one?
  - Brian

I've never signed an NDA to receive a quote. Some of my contracts have NDAs in them after the fact but I've never been asked to sign one before I received pricing from a transit provider.


No provider wants you disclose the information. Hmm.... someone posted on LINX that he can get $500 for a 10 Gbps unmetered port from a Tier 1 ISP, do you believe it?

Probably you also have never got the best possible pricing :wink:

Ugh. Requiring an NDA to get best pricing is a business practice
that makes me feel I need to wash my hands after dealing with them.
  - Brian

Agreed. My biggest frustration buying carrier services is the lack of transparency in pricing.

Some points I have not seen so far are:
- how do you connect? local cc in the dc or several other fiber runs
to reach a different dc/city? (affects price, setup time, maintenance
and debugging)
- where is your traffic going to/from? how many intermediate ASs or
long transfers are involved?
- bgp community support to influence routing (including the already
mentioned blackhole)
- flowspec support
- additional services like ddos mitigation
- how many ports/locations per committed bandwidth
- your own experience with the company (sales/support)
- I assume IPv6 support does not need to be mentioned anymore :slight_smile:


I think it’ll depend on your target customer. Residential eyeball? Being on an IX is more important at nearly any size than which transit you choose. Even a good-sized residential eyeball (say 10k and up subs) can be good with Cogent\IX\one other transit.

Hosting and enterprise-focused ISPs will need to diversify their transit providers more.

I would actually venture to say the contrary. An IX should be the last item on your list since it only really makes sense at a certain scale and if you can make use of the providers on it.

Most of the networks you’ll have trouble getting to via transit providers are that way because of how they do business, which also means hardly any of them peer at IXes. I’d say a network should have a least 3 good transits before considering an IX. Even then it’s not so black and white. If after your first transit provider is installed and you set up your flow monitoring, you notice most of your traiffic is going to/coming from ASNs that peer on your local exchanges, then it absolutely makes sense to open a connection right then.

IX links aren’t a whole lot cheaper than transit (sometimes they cost more depending on how hard it is to get to them) and many networks will benefit from a more diverse blend of transits than IX peering regardless of what they’re doing. IXes are extremely important to the internet at large, but they’re not for everyone.


The type of customer on the network is important here.

Most traffic on residential eyeball networks goes to IXes. I know guys pushing 85% of their traffic to IXes. Even small IXes like ours are capturing well over 50% of an ISP’s traffic. Netflix, Google, Akamai, Cloudflare. That’s what, 2/3rds of the traffic an eyeball has?

Now if you’re not predominately serving residential customers, then I agree and briefly stated so before.

Flow monitoring is indeed important.

Usually, DIA (as transit delivered to a customer) is more expensive than transport + transit + small colo (1U\2U stuff) + IX… at least as observed by many of my brethren.

That’s before you get to the fact that a lot of transit is sub-optimal. Most ISPs we’ve hooked to our IXes have seen an immediate increase in network utilization because upstream congestion and whatever latency is gone.