Oliver Thylmann’s Thoughts

Greetings everyone. On and off I seem to be having problems with my POP access to Gmail and I now seem to have found the reason for all of this, just if some others also have the problem. The problem itself is that the pop request just times out. The problem seems to be with Google or something on the route to some of the server pools.

It was working fine and I ran a traceroute to pop.gmail.com and ended up going over:

11 po13-0.chicr2.chicago.opentransit.net (193.251.240.137) 154.259 ms * *

12 66.249.95.253 (66.249.95.253) 158.005 ms 158.562 ms *

[…]

15 64.233.175.97 (64.233.175.97) 244.343 ms 239.228 ms 243.426 ms

[…]

17 72.14.205.109 (72.14.205.109) 240.442 ms 72.14.236.142 (72.14.236.142) 251.769 ms 66.249.94.73 (66.249.94.73) 237.330 ms

It worked just fine. I seem to be somewhere in Chicago there.

Next it stops working again and where am I?

7 br1.frankfurt1.iphh.net (80.231.66.38) 56.323 ms 56.652 ms 56.785 ms

8 216.239.46.48 (216.239.46.48) 66.315 ms 57.540 ms 56.848 ms

9 64.233.175.248 (64.233.175.248) 63.733 ms 63.763 ms 64.019 ms

10 216.239.43.84 (216.239.43.84) 66.259 ms 75.122 ms 216.239.43.89 (216.239.43.89) 64.718 ms

11 216.239.43.88 (216.239.43.88) 64.565 ms 64.047 ms 63.464 ms

12 216.239.43.84 (216.239.43.84) 66.127 ms 67.032 ms 64.233.175.246 (64.233.175.246) 66.150 ms

13 216.239.43.42 (216.239.43.42) 67.479 ms 72.14.236.150 (72.14.236.150) 78.471 ms 216.239.43.34 (216.239.43.34) 79.919 ms

14 64.233.183.109 (64.233.183.109) 67.929 ms 66.769 ms 216.239.43.30 (216.239.43.30) 72.721 ms

The hop back and forth seems a bit weird there. At the next try when it is not working I am in Frankfurt again.

7 br1.frankfurt1.iphh.net (80.231.66.38) 56.530 ms 58.030 ms 56.345 ms

8 216.239.46.48 (216.239.46.48) 57.547 ms 57.390 ms 56.825 ms

9 64.233.175.248 (64.233.175.248) 63.578 ms 64.133 ms 63.929 ms

10 216.239.43.80 (216.239.43.80) 79.355 ms 67.291 ms 216.239.43.89 (216.239.43.89) 63.445 ms

11 216.239.43.88 (216.239.43.88) 63.757 ms 83.265 ms 63.496 ms

12 64.233.175.246 (64.233.175.246) 66.718 ms 70.648 ms 66.321 ms

13 216.239.43.26 (216.239.43.26) 68.140 ms 216.239.43.30 (216.239.43.30) 66.530 ms 72.14.236.146 (72.14.236.146) 66.717 ms

14 216.239.43.30 (216.239.43.30) 66.663 ms 72.14.236.146 (72.14.236.146) 66.918 ms 64.233.183.109 (64.233.183.109) 66.586 ms

If anyone from Google is listening, please look into it, and I am always happy to provide more answers to any questions. The problems are both on a Mac with Mail and on an XP machine with Outlook.

Greetings from Germany, where I wish I get my mail from Chicago again.

Mena and Ben have posted a note on the recent TypePad performance problems entitled The Ups & Downs of a Successful Service. I can fully relate to how hard it is to get a system to be really scalable and get the power you need. It’s not easy to run a fast growing web site and when running the IT side of one of the biggest performance based advertising networks in Germany (now Ligatus part of OnVista Group) I was fortunate enough to have an existing infrastructure to base everything on as OnVista was already running onvista.de, the biggest bank indepedent financial portal in Germany.

Looking for a hosting facility is a fun business, especially if you need it scalable and if you then start thinking about the immense power that Blade Servers need you start to sweat. IBM for example will only allow you to put stuff into their racks or racks that pull in cold air in the front and push it back out through the back of the racks. Some hosting centers push air into the racks from below the floor and pull it back out at the top but the fans, or rather turbines that are in the blade centers pull from the front and push out in the back and as they are so big, the airflow could get problematic if it needs to be diverted to the top. You also need A LOT of power. A fully equipped blade center needs at least 2 lines of 16A power, but for failover, you need 4. Now if you put 3 blade centers in one rack, you kind of need 8 lines of power (sharing two between two centers) and that’s already well above the standard setting that man centers provide.

As recently posted by Cringely the climate control systems need 3 times the power than a rack pulls in, so you kind of need 8 + 24 lines of power (ok, not really, but still a lot) and if the blade center boots up, it sucks all the power it can get. It actually sounds like an airplane starting. Very funny.

At the 240mbps that Ben quotes, and presuming you pull 5mbps on each server, you need 48 servers already. That’s without factoring in that you need failovers and database servers in the backend, and load balancing and and and.

I was fortunate enough to be able to offload a huge deal of the traffic to Akamai, as the ads can be cached, pulling load and traffic of the backend and making the entire system immensely scalable in terms of advertisements delivered.

So people of TypePad, I feel with you. Your task is a lot bigger than mine was. Hang in there, better times will come. I for one will stay loyal.

Omid is the business guy behind Google and John did an interview with him which you can find here: John Battelle’s Searchblog: Titans Column: Omid Kordestani.

One part of it actually rang a bell:

The measurability of online advertising will extend broadly to all areas of media. You have companies spending billions of dollars on television. As more and more consumers adopt technologies like TiVo, I think you’ll be able to have much more useful forms of advertising — more targeted, more measurable, and with new pricing models. Just imagine if we made it possible for our advertisers to quickly publish relevant ads that could range from the local plumber on one end to Super Bowl commercials on the other.

I absolutely agree. A lot of things will open up for performance advertising. We will just need to exploit technology to allow us to do that. Did you know for example that easyJet prints different phone numbers on posters everywhere to be able to track which ones and where work best? Did you know that Jamba seems to be booking their ring tone ads on a performance basis on TV channels? And this is just the start. Just imagine if you can set up an API to let your ad be automatically queried from your servers just seconds before it runs on TV.

Long time I didn’t post here, but that is, as already reported, because I am done, and I now also have my MBA certificate which looks mighty fine I might add. I just received note from the OU though that they are looking for donations in a system called Case for Support. If you did study, or are studying, I suggest you give. It’s a great university with some great projects.

Ok, Google Base has almost landed. As Reported on several spots, like here, Google is about to launch Google Base and I have now gotten one step further than an error message. I now get a login prompt, but it does not accept my password.

Let’s see what happens next.