Bad routing/peering. UPC and Cogent.

Chanquete

Hi all

I am having latency issues connecting from my home computer to my place of work. I am located in Dublin North and work is located in Dublin City Center. So around 3km distance.

I have a UPC 100Mb broadband, and my company has a 100Mb line with Cogent. Before they moved to Cogent they were with Magnet and I was able to get great connectivity home-work.

However since they moved to Cogent the latency has been dreadful, causing the connection speed to drop to anything between 1-5Mbps. I did a tracert between home and work and this is what I get:


1 <1 ms <1 ms <1 ms [home computer]
2 9 ms 9 ms 8 ms 46.7.174.1
3 9 ms 9 ms 6 ms 109.255.255.190
4 101 ms 99 ms 100 ms 84.116.238.42
5 99 ms 104 ms 99 ms nl-ams05a-rd2-xe-2-0-1.aorta.net [84.116.134.117]
6 102 ms 100 ms 99 ms us-nyc01b-rd1-gi-2-0-0.aorta.net [84.116.130.26]
7 110 ms 102 ms 100 ms us-nyc01b-ri1-xe-3-1-0.aorta.net [213.46.190.94]
8 99 ms 101 ms 101 ms xe-8-3-0.edge1.NewYork1.Level3.net [4.78.164.221]
9 102 ms 99 ms 99 ms COGENT-COMM.car2.NewYork1.Level3.net [4.68.111.46]
10 101 ms 100 ms 101 ms te0-1-0-5.mpd21.jfk02.atlas.cogentco.com [154.54.3.97]
11 177 ms 176 ms 176 ms te0-1-0-2.ccr21.bos01.atlas.cogentco.com [154.54.44.6]
12 175 ms 174 ms 177 ms te0-0-0-1.ccr21.lpl01.atlas.cogentco.com [154.54.31.170]
13 348 ms 204 ms 411 ms te3-2.ccr01.man01.atlas.cogentco.com [130.117.51.253]
14 183 ms 199 ms 203 ms te1-1.ccr01.dub01.atlas.cogentco.com [130.117.3.213]
15 183 ms 180 ms 179 ms te2-1.ccr01.dub02.atlas.cogentco.com [154.54.56.50]
16 177 ms 178 ms 177 ms [work computer]

So, from the node names, it looks like it does something like DUBLIN <-> AMSTERDAM <-> NEW YORK <-> BOSTON <-> MANCHESTER <-> DUBLIN. Complete madness.

My IT guys tell me there is nothing they can do (although they tried) to get Cogent to improve the situation, and that is most likely a problem with UPC having bad peering and announcing non-optimal routes (I am no expert so I may be misquoting what they told me).

I tried contacting UPC support but they replied with irrelevant questions like "are you using wireless or wired connection" and "have you tried restaring the modem".... I replied nevertheless and asked them to raise it to more advanced support and never heard back from them.

I was wondering if anybody here has had similar issues or has any suggestion or what I can do.

Thank you,
Chanquete

Skittle

I had a problem like that before and I contacted UPC support and send them the tracert output, it took a couple of weeks, but it got sorted in the end.

Bohrio

Chanquete wrote: »

Hi all

I am having latency issues connecting from my home computer to my place of work. I am located in Dublin North and work is located in Dublin City Center. So around 3km distance.

I have a UPC 100Mb broadband, and my company has a 100Mb line with Cogent. Before they moved to Cogent they were with Magnet and I was able to get great connectivity home-work.

However since they moved to Cogent the latency has been dreadful, causing the connection speed to drop to anything between 1-5Mbps. I did a tracert between home and work and this is what I get:


1 <1 ms <1 ms <1 ms [home computer]
2 9 ms 9 ms 8 ms 46.7.174.1
3 9 ms 9 ms 6 ms 109.255.255.190
4 101 ms 99 ms 100 ms 84.116.238.42
5 99 ms 104 ms 99 ms nl-ams05a-rd2-xe-2-0-1.aorta.net [84.116.134.117]
6 102 ms 100 ms 99 ms us-nyc01b-rd1-gi-2-0-0.aorta.net [84.116.130.26]
7 110 ms 102 ms 100 ms us-nyc01b-ri1-xe-3-1-0.aorta.net [213.46.190.94]
8 99 ms 101 ms 101 ms xe-8-3-0.edge1.NewYork1.Level3.net [4.78.164.221]
9 102 ms 99 ms 99 ms COGENT-COMM.car2.NewYork1.Level3.net [4.68.111.46]
10 101 ms 100 ms 101 ms te0-1-0-5.mpd21.jfk02.atlas.cogentco.com [154.54.3.97]
11 177 ms 176 ms 176 ms te0-1-0-2.ccr21.bos01.atlas.cogentco.com [154.54.44.6]
12 175 ms 174 ms 177 ms te0-0-0-1.ccr21.lpl01.atlas.cogentco.com [154.54.31.170]
13 348 ms 204 ms 411 ms te3-2.ccr01.man01.atlas.cogentco.com [130.117.51.253]
14 183 ms 199 ms 203 ms te1-1.ccr01.dub01.atlas.cogentco.com [130.117.3.213]
15 183 ms 180 ms 179 ms te2-1.ccr01.dub02.atlas.cogentco.com [154.54.56.50]
16 177 ms 178 ms 177 ms [work computer]

So, from the node names, it looks like it does something like DUBLIN <-> AMSTERDAM <-> NEW YORK <-> BOSTON <-> MANCHESTER <-> DUBLIN. Complete madness.

My IT guys tell me there is nothing they can do (although they tried) to get Cogent to improve the situation, and that is most likely a problem with UPC having bad peering and announcing non-optimal routes (I am no expert so I may be misquoting what they told me).

I tried contacting UPC support but they replied with irrelevant questions like "are you using wireless or wired connection" and "have you tried restaring the modem".... I replied nevertheless and asked them to raise it to more advanced support and never heard back from them.

I was wondering if anybody here has had similar issues or has any suggestion or what I can do.

Thank you,
Chanquete

Is your company Irish?

bkehoe

Have had the exact same problems, and posted about it here several times, for over a year.

At some point in Autumn 2010 things went horribly wrong. Up to that point peering happened in London where both UPC and Cogent have a presence and everything worked happily.

Now since the routing changed to go via New York the pings remind me of dialup and speeds over that routing can rarely get above 2MBit.

We have a business UPC connection so at least it was relatively easy to get someone in support who understood the problem and referred it to the relevant department but according to them it is intentional!

We utilise real time audio streams from several companies in Dublin which use Cogent and thus require stability and low latency and since this change has occurred it's been a nightmare but we don't have any other ISP to choose from and for everything else UPC is great.

Chanquete

I will try to raise it again and see if I'm more lucky this time with UPC support.

Bohrio, yes, my company is Irish.

I assume there is no indepent authority / organization that can force/push UPC and Cogent to do the right thing and peer at a closer location? I understand there are cost savings involved but routing Dublin-Dublin traffic through two continents and four different countries is pure madness.

Snaga

UPC are doing the right thing - they are members of INEX which allows them to share traffic with other ISP's that service Ireland (and are also members).

https://www.inex.ie/about/memberlist

Cogent are not members of INEX and I would imagine have little interest in becoming one. (upfront cost to them for very little direct value - at least until their corporate customer base in Ireland increases dramatically).

The quote from bkehoe is interesting...

but according to them it is intentional!

Cogent are renowned for getting into spats when it comes to paying for bandwidth interconnects with others - leaving full interconnects causing major packetloss for people - often requiring creativity by other ISP's to get around the problem.

High latency sucks - but its better than 50% packetloss!!

Sponge Bob

Snaga wrote: »

Cogent are not members of INEX

They are. An Associate member. > https://www.inex.ie/about/memberlist

If you go DIRECT to Cogent and are not an INEX member yourself (or have a INEX member 'hosted' or peered AS number) you can get sent around the houses

OP, tell your COMPANY to join INEX and all will be well.

Snaga

They are an associate member - yes. (I concede I had forgotten this!).

They are not a physically connected full member so UPC can't peer with them at INEX until they physically join (as you know!).

You are right Spongebob - his company could join Inex and try to peer with UPC - not sure the costs will stack up or if UPC's peering team would go for it!! Good thinking though

Sponge Bob

INEX membership is very good value and UPC have no choice ....even if Cogent supply the members own transit outside the INEX

Snaga

Sponge Bob wrote: »

INEX membership is very good value and UPC have no choice ....even if Cogent supply the members own transit outside the INEX

I don't see anything in the MoU that demands you have to peer with everyone that asks! (You need to respond in 5 days to requests, doesn't mean you have to say yes).

That aside - joining INEX is no trivial matter - it is more than a slight pain in the arse for a company with a single connection and likely a handful of IP's belonging to Cogent. (Getting AS, PI space, network migrations etc....).

Simple and more cost effective would be to get a second smaller connection into the office with UPC (or Magnet or someone else already peering with UPC at INEX) and get your IT guys to configure the network to use it for the applications you need.

Then support a local ISP next time your contract is up for renewal! As you have found, the bigger International guys don't care a whole lot about local politics!

bkehoe

Both Cogent and UPC Ireland have a presence in LINX in London which is where traffic used to peer until one of them pulled the plug in 2010.

This is the UPC Business support response received in March 2011:

I have been advised that it is not possible to peer with Cogent in London as they are charging for us sending and receiving traffic if we peer here.

This is the reason that the peering is in the US.

Sponge Bob

I remember when UPC took over NTL and made everybody in Ireland peer in Amsterdam...that was a bloody disaster.

But I suspect Cogent are at fault here...like this example