I’ve recently found myself busy plumbing the depths of data center performance, in order to recommend the best options to our clients. One finding I want to share addresses something called connection latency: the time a request takes from my computer to hit server, and for the server to respond with some data. I’ll walk you through my method.
This came up because Sametz serves clients across the nation, but their target markets may or may not be more regional. Because of this, it often makes the most sense to locate their data center closer to their target markets.
In evaluating acceptable load times, I cite Jakob Nielsen, because of his experience in UX design. In the following article, he notes the levels of responsiveness that can make or break a user interaction. While there is no target minimum load time, we can aim for a much more manageable three seconds. Most of what I have seen suggests that this is pretty fast, and quite possible without busting the budget for speed.
In the article below, John Matherly also purports to map the concentration of internet-connected devices in the world. Globally, the highest concentrations seem to be in the Eastern US, and Europe. With this information in mind I set out to group web hosts by their proximity to a motivated web browser, hoping to get the most out of response times, search ranking, and the other intangibles associated with a fast web application.
I took the list of hosts from this report detailing the most popular providers among the top million web sites, and limited to those centered in the US. In classifying data centers by geography, I assume that their server farm sits reasonably close to their mailing address, unless they make some other information available. We can reasonably assume that the site they use to sell their services represents the best they have to offer, so I verified my estimation of their latency by pinging their commercial domain.
I also have done no investigation into the associated costs, or features that these hosts provide. That is a separate discussion, but definitely worth having to narrow them down to which ones deserve an endorsement.
A good East Coast data center seems to range from 20ms to 30ms, which is reasonable given that we live on the East Coast. This is also a major center of internet connectivity for the world, so it makes sense to favor East Coast hosts for most purposes.
From the East Coast, a Midwestern data center seems to average ping at 50ms. Most US activity seems to center on the coasts, but it never hurts to have a local option for a Midwestern customer.
From our office, a West Coast data center pings at around 95ms. That lag can fairly illustrate the point of network latency: it takes twice as much wire time to get any thing out of California as Chicago, and three times as long as Virginia. The West Coast is certainly a center of activity in its own right though, so it should not be discounted.