Should I ask to move datacentres?
I know the optimal solution is to host locally, but it's a lot more expensive to get hosting in the UK. But in going from west coast to east coast, I can cut out a big chunk of the geographic distance.
I guess the best comparison is to compare a traceroute from my linode in Fremont with one between somewhere this side of the atlantic and a linode in the Atlanta datacentre. Anyone care to help me out by providing me with either an IP I can traceroute, or suitably edited data if we're being paranoid?
3 Replies
My Atlanta Linode is on 64.22.124.184, if you want to see what times to it are like.
Fremont
traceroute to technomancer.co.uk (64.71.152.131), 30 hops max, 40 byte packets
1 192.168.2.1 (192.168.2.1) 2.136 ms 2.057 ms 3.910 ms
2 10.231.212.1 (10.231.212.1) 10.720 ms 11.122 ms 11.351 ms
3 leed-t2cam1-a-ge91.inet.ntl.com (80.0.54.153) 13.495 ms 13.717 ms 17.568 ms
4 seac-dpim1-1-coc-1-gw.service.virginmedia.net (80.0.50.146) 19.983 ms 18.537 ms 18.193 ms
5 lee-bb-a-so-230-0.inet.ntl.com (213.105.175.65) 18.754 ms 18.802 ms 20.590 ms
6 nth-bb-b-so-100-0.inet.ntl.com (62.253.185.101) 20.654 ms 16.769 ms 16.860 ms
7 tele-ic-1-as0-0.inet.ntl.com (62.253.184.2) 16.696 ms 17.199 ms 17.356 ms
8 gsr12012.lon.he.net (195.66.224.21) 16.839 ms 17.402 ms 22.035 ms
9 10gigabitethernet2-3.core1.nyc4.he.net (72.52.92.77) 89.328 ms 89.513 ms 83.532 ms
10 10gigabitethernet3-1.core1.sjc2.he.net (72.52.92.25) 163.581 ms 166.892 ms 167.296 ms
11 10gigabitethernet1-1.core1.fmt1.he.net (72.52.92.109) 167.610 ms 168.902 ms 169.097 ms
12 li14-131.members.linode.com (64.71.152.131) 167.329 ms 164.543 ms 164.795 ms
Atlanta
traceroute to 64.22.124.184 (64.22.124.184), 30 hops max, 40 byte packets
1 192.168.2.1 (192.168.2.1) 2.304 ms 2.106 ms 4.386 ms
2 10.231.212.1 (10.231.212.1) 40.085 ms 40.393 ms 40.665 ms
3 leed-t2cam1-b-ge91.inet.ntl.com (80.0.55.153) 40.861 ms 41.057 ms 41.341 ms
4 leed-t3core-1b-ge-115-0.inet.ntl.com (195.182.179.241) 41.437 ms 42.833 ms 43.329 ms
5 lee-bb-b-so-020-0.inet.ntl.com (213.105.175.69) 43.525 ms 43.724 ms 43.904 ms
6 nth-bb-a-so-030-0.inet.ntl.com (213.105.175.133) 46.485 ms 21.535 ms 13.163 ms
7 212.43.162.230 (212.43.162.230) 37.590 ms 37.919 ms 38.192 ms
8 cr02.frf02.pccwbtn.net (80.81.192.50) 38.386 ms 38.572 ms 38.803 ms
9 gnax.ge2-13.br01.atl01.pccwbtn.net (63.216.31.130) 132.427 ms 132.812 ms 132.877 ms
10 atl-core-a-g3-1.gnax.net (63.247.69.177) 131.460 ms 135.460 ms 135.800 ms
11 l3-atl-18.gnax.net (209.51.131.74) 149.957 ms 133.698 ms 133.834 ms
12 splinter.brokenbottle.net (64.22.124.184) 133.824 ms 129.441 ms 143.790 ms
It's the same number of hops despite being routed completely differently. Lets ping a little.
--- technomancer.co.uk ping statistics ---
40 packets transmitted, 40 received, 0% packet loss, time 38999ms
rtt min/avg/max/mdev = 161.370/167.056/192.970/7.243 ms
Atlanta
--- 64.22.124.184 ping statistics ---
40 packets transmitted, 40 received, 0% packet loss, time 39004ms
rtt min/avg/max/mdev = 126.157/128.749/148.592/4.488 ms
So we're looking at 38ms +/- 10ms better off. Not huge but is it noticeable? I suppose I should work out the average page load time on hot and cold cache samples. But off the top of my head, a page on leedsrocksoc.co.uk (drupal+lots of avatars) might be >40 GET requests on a cold cache on a busy page and maybe 4 on a very warm cache, but there may be parallel connections going on. Best guestimate maybe between 0.1 and 1 second per page. That's comparable to the difference between default configs for lighttpd and apache (although a well tuned apache is nearly as quick for me but that's off topic).
I think I'm going to ask the rest of the site teams this one. I think we're going to have some scheduled downtime during the summer when the sites are pretty much dead and there are major site upgrades to be rolled out.