Choose your server city by ping, not by the map

The cheapest server in the list might be in the wrong city for you. Before you buy, it is worth ten minutes to work out where your users actually are and how the network gets to them.

Distance is a starting point

Light in fiber covers about 200 km every millisecond, and the round trip doubles that. Every router on the way adds a little more. So as a rule of thumb, every 100 km between you and the server adds roughly a millisecond of ping.

That makes a server in Dallas about 30 to 40 ms from either coast of the US, and a server in Frankfurt under 30 ms from most of Europe.

But cables do not follow the map

Networks follow the cables, and the cables follow money. A user in Phoenix might reach Los Angeles faster than Denver even though both are a similar distance away. In parts of Latin America, Miami can beat a server in the next country over, because the cables between them run through Florida.

The only way to know is to test.

How to test before you buy

Every city has a test IP. Ping it from your own computer, or ask a few users to do the same:

ping 167.88.62.166 (the Dallas test IP)

Then run a traceroute from the data center back to you with the looking glass. It shows each hop and where the delay comes from. If one city is consistently 20 ms better than another for your users, that matters more than a $5 difference in price.

What ping matters for

For a website, the difference between 20 ms and 80 ms is hard to notice, because the page takes far longer to render than the network takes to deliver it. For game servers, voice chat, remote desktops and trading tools, it is the difference between smooth and sluggish.

Some quick picks

• Players all over North America: Dallas or Chicago.
• Mostly East Coast: Ashburn or New York.
• West Coast and Asia: Los Angeles, Silicon Valley or Seattle.
• All of Europe: Frankfurt or Amsterdam.
• The US and Europe from one server: London or New York.

Once you know which two or three cities work, compare their prices. Stock and markdowns differ a lot between neighbors, and the same machine is often cheaper one city over.