I've been a fan of the network at Corexchange. Incero uses Corexchange's blend plus adds to it their own Cogent (Corexchange has Cogent in their blend already) and nLayer. nLayer is kind of hit or miss for me elsewhere and blah. The Corexchange blend is heavy on Level3, Abovenet and NTT. I'd take that network over most networks in the US. Incero's usage seems to be 46% Cogent, 38% Corexchange and 15% nLayer, so very different that Corexchange house blend.
Oh dear, let me help here.
If you get your information from bgp.he.net (as I assume you did) then you need to understand how that site works. It's basically a route collector. That is, it's looking up the routes from their location to our location. As Cogent is the most interconnected network in the world (most AS numbers connected) the cogent path INBOUND to us will appear more often across the internet when looking at possible routes.
HOWEVER, the number of routes and route options inbound to a provider has nothing to do with which links the provider actually uses to send the data out to it's customers.
For example, here is a screenshot of our current uplinks, the top graph for each link is a daily graph, the bottom graph is a weekly graph (
BLUE IS OUT FROM US,
GREEN IS IN TO US):
So we can see, on the weekly basis:
Corex 10gig uplink outbound 6gbit/second
Cogent 10gig uplink outbound 0gbit/second (NO TRAFFIC SENT OUT, used as failover)
Nlayer 20gig uplink outbound 13.2gbit/second
=
40GBIT/SECOND CAPACITY, 19GBIT/SECOND USED, 21GBIT/SECOND SPARE, 0GBIT COGENT USED
So if you love Corex bandwidth, and you love Nlayer bandwidth, then we have the best of both worlds (plus some added redundancy). Our 20gig nlayer link is made up of 2x 10gig links going out diverse sides of the building, and out corex and cogent links also go out diverse sides from each other.
Our 40gig (2x 20gig via diverse sides and paths around Dallas) ZAYO/ABOVE.NET link is currently deploying and scheduled to turn up for customers on Sept 1st, bringing us to 80GIG. Then we will without a doubt have the best network offering out of most every provider in Dallas.
--------
As for shovehosts problems. I apologize for not being able to help him.
He was trying to scp (encrpytion) a file to another host with X latency, SCP will typically cap out over any non 5ms link at a crazy low rate. The correct way to test his link would be to use iperf.
Giving us a download speedtest from softlayer showing good speeds, then an upload to them showing slow speeds just reminds me that they have those old ass pata drives in there writing at 10MB/second. If he had used the frostgaming speed test server or pushed/pulled to cachefly he would have reached full speeds.
Ultimately we gave him an awesome deal, he said he was going to TEST THE SERVER FIRST (no clients) and then he became a version of R****t C***ke, but we're an unmanaged host for expert users only so I refunded him and cut him loose. Since then I've hired two additional staff to handle support from our datacenter office and will defer such tickets to them.
At some point hopefully we can setup an automated/self-service webpage that schedules an iperf from our network to someone's intended target, so they can test our network outbound without actually having to setup a server.
All customers can reach their 1gbit speeds outbound at any time.
----------
xoxo
Gordon
