Thread: UK-based Dedicated Server Needed

Its not going to make any differnece really All hosts will be within reason as fast as each other.

I've already posted the MX's, so just dac.nic.uk to go...

[root@typhon root]# traceroute dac.nic.uk
traceroute to ls3.nic.uk (195.66.240.114), 30 hops max, 38 byte packets
1 rb-sov-edge.netrino.co.uk (84.234.19.2) 0.427 ms 0.403 ms 0.294 ms
2 cr1-fe40.rbs.uk.euroconnex.net (217.112.80.49) 0.884 ms 0.636 ms 1.601 ms
3 collector.linx.net (195.66.225.254) 1.123 ms 0.849 ms 1.353 ms
4 fa3-0.tr2.tfm7.thn.linx.net (195.66.248.232) 1.153 ms 1.225 ms 1.269 ms
5 nic-gw.linx.net (195.66.248.90) 1.248 ms 1.189 ms 1.237 ms
6 195.66.240.114 (195.66.240.114) 1.285 ms 1.131 ms 1.233 ms

See for yourself at http://www.netrino.co.uk/tools/network.php

Originally Posted by neilrbowen

Its not going to make any differnece really All hosts will be within reason as fast as each other.

That's not strictly true is it Neil. I would expect a reasonable difference to be made if a host saw the network via a peer rather than over a transit connection.

Chris

Originally Posted by neilrbowen

Its not going to make any differnece really All hosts will be within reason as fast as each other.

Originally Posted by NetrinoUK

That's not strictly true is it Neil. I would expect a reasonable difference to be made if a host saw the network via a peer rather than over a transit connection.

"fast" ... "reasonable difference" - all subjective terms.

I think it all really depends on exactly what we're talking about, and what the application is. In this case latency, is a script sending mail going to really notice an extra 5ms in TCP connection set-up times? I seriously doubt it. Is a gaming user going to notice an extra 2-3ms, I don't think so, 10ms maybe.

If you are measuring using tools and actually generating comparative stats then you can notice these things, in real world applications, I don't really think so.

In terms of peering and throughput, you do win more noticeably, since you have more control over the throughput between yourself and the other peer, being a direct connection.

Just my personal view, of course.

A lot of gamers will complain about number of hops... which goes to show how much they know..

Hi,
I know I am demanding too much but you already got my critical point. Now,

This is my existing server's traceroute to mx3.nominet.org.uk

root@ds899# traceroute mx3.nominet.org.uk
traceroute to mx3.nominet.org.uk (213.248.199.23), 30 hops max, 40 byte packets
1 81.21.78.2 (81.21.78.2) 0.555 ms 0.484 ms 0.420 ms
2 g4-0-0-258.ld-cr31.cix.gxn.net (62.105.126.18) 0.715 ms 0.507 ms 0.455 ms
3 g0-2.cr03.hx2.bb.pipex.net (62.72.156.110) 6.065 ms 5.877 ms 5.879 ms
4 g1-1-5.cr01.hx2.bb.pipex.net (62.72.142.1) 6.619 ms 6.727 ms 6.203 ms
5 fe1-2-nom-col1.nominet.org.uk (217.79.160.112) 7.650 ms 6.900 ms 6.678 ms
6 host-213-248-206-10.nominet.org.uk (213.248.206.10) 6.779 ms 6.582 ms 6.762 ms
7 host-213-248-206-1.nominet.org.uk (213.248.206.1) 10.497 ms 10.813 ms 10.182 ms

I have two questions:

1. What is the difference between hop 6 and hop 7 here. Which one is final point at which nominet receives the email.

2. Do any one of you have a server offer which can give me better than NetrinouK's offer which is,

host-213-248-206-10.nominet.org.uk 0% 5 5 0.81 1.00 1.19
host-213-248-206-1.nominet.org.uk 0% 5 5 3.94 4.14 4.41

BTW these are round-trip times, aren't they

Here is the problem with relying on traceroute too heavily, there are often too many things going on with the way the traceroute test packets are processed, to always be really meaningful - unless you know how the protocols work (I doubt many people here have read Stevens Vol 1).

For example, routers and hosts can de-prioritise certain traffic types if they are busy (processing the kind of packets used in traceroute tests), rate limit ICMP responses (Solaris) and firewalls will often drop them completely.

Relating to point #1 this is the problem you are seeing, hop 7 is dropping the traceroute traffic so may be an indication of the higher round trip time for that hop. No way of knowing how many hops further on than that until you reach their mail server. (Yes, those are RTTs).

as far as i can see, none of those are mail servers - they are just routers, with the actual mail server possibly being at a hop 8 or 9 - something is dropping ICMP traffic after hop 7. possibly a firewall

tbh, i think you're placing far too importance on these latency values. you should only start worrying when you see the pings getting near 200ms. SMTP (mail) protocol doesnt need particuarly low latency. if i were you, i'd concentrate on finding a provider with decent customer support and uptime, rather than deciding over a couple of ms difference in latency

The difference between hop 6 and 7 is most likely the backhaul connection from London to Oxford.

The latency between hops 2 and 3 will be the backhaul from the datacentre you are in (hosteurope or whatever they were called, perchance?) and London.

I've dropped you with a PM containing full traces, and for sake of argument a trace past nominet's ICMP echo filtering firewall/router.

The main point i think is, if your worried about mail, are you going to notice even a 500ms latency

Of course i would worry if my host had a latency like that, but i think support, and uptime are of more importance for this.

Originally Posted by philb

The difference between hop 6 and 7 is most likely the backhaul connection from London to Oxford.

I had been assuming that all these hosts were actually in London DCs rather than the Nominet office, but you could be right there.

[root@s58 ~]# traceroute dac.nic.uk
traceroute to ls3.nic.uk (195.66.240.114), 30 hops max, 38 byte packets
1 195.242.213.254 (195.242.213.254) 0.163 ms 0.139 ms 0.119 ms
2 84-45-252-145.no-dns-yet.enta.net (84.45.252.145) 0.521 ms 0.488 ms 0.524 ms
3 telehouse-east2.bdr.enta.net (84.45.193.194) 0.562 ms 0.614 ms 0.569 ms
4 pr2.tcm.linx.net (195.66.227.254) 1.111 ms 1.471 ms 1.104 ms
5 g0-0.tr3.tcm.linx.net (195.66.248.126) 1.166 ms 1.307 ms 1.784 ms
6 fa1-0.tr2.tfm7.thn.linx.net (195.66.248.254) 191.100 ms 271.299 ms 308.442 ms
7 nic-gw.linx.net (195.66.248.90) 1.298 ms 1.101 ms 1.221 ms
8 195.66.240.114 (195.66.240.114) 1.269 ms 1.281 ms 1.157 ms

traceroute to mx3.nominet.org.uk (213.248.199.23), 30 hops max, 38 byte packets
1 195.242.213.254 (195.242.213.254) 0.168 ms 0.126 ms 0.113 ms
2 84-45-252-145.no-dns-yet.enta.net (84.45.252.145) 0.467 ms 0.430 ms 0.415 ms
3 fe1-2-nom-col1.nominet.org.uk (217.79.160.112) 0.534 ms 0.515 ms 0.525 ms
4 host-213-248-206-10.nominet.org.uk (213.248.206.10) 0.860 ms 0.728 ms 0.692 ms
5 host-213-248-206-1.nominet.org.uk (213.248.206.1) 3.643 ms 3.524 ms 3.726 ms
6 * * *


Not bad, seams even better than a direct peer, which goes to show you its not all about peering, more about the route. A route over 1000 Mbit transit into nominet, can be better than a 100 Mbit peer and even congested peering connections.