Fog Creek Software
Discussion Board




comparing connection speed of different webhosts

My company is changing the webhosting service and planning to shift to a dedicated managed server solution. There are a hell of a lot of companies who've contacted us for providing the above service.

But the question is which company to choose. Hardware configuration , support, pricing etc can be compared but how to compare the connection speed. How to make sure that if we host our website with company XYZ, our website will load faster than if we host with ABC and this is the biggest concern of my company.

I know loading speed of web pages is a complex function of website traffic, timing of the day, point of access etc etc. so the commands like 'traecrt' and 'ping' are useless to use as a benchmark. But there must be some way which can determine that on an average website A loads faster than website B.

Could any of you suggest some way to compare the connection speed of different web host services ?

Sach
Saturday, May 10, 2003

You can hire a company like Keynote or Gomez to measure it for you. That's what they specialize in, and they'll have computers around the world measure the performance around the clock.

Joel Spolsky
Saturday, May 10, 2003

Web site loading speed is mostly influenced by the load on your server. Make sure your server got the muscle to handle the task.

ISP connectivity is important too. Go for one for the biggest providers with multiple backbones. That way data takes the shortest, thus fastest, route and they got backup connections if something some of the other connectivity goes down.

If you want to believe my blue eyes here's what you should do. For a managed dedicated server and money is not a problem go with
http://www.rackspace.com

If you are on a budget and are able to manage your servers yourself go with
http://www.rackshack.net

Both are huge with multiple gigabit backbones and have excellent reputations in the dedicated server market. Everything else you ever wanted to know about hosting can be found on:
http://www.webhostingtalk.com

Jan Derk
Saturday, May 10, 2003

You can try http://www.cycletraders.com/ but don't expect too much.  Still, it's free.

J. D. Trollinger
Saturday, May 10, 2003

An additional small point, possibly a bit esoteric, though not in contradiction to anything I've seen the previous posters write, is that the loading speed, as perceived by the end user, may mask a number of the other factors cited. Where true, this by itself doesn't necessarily obviate the need to have a server that would respond better if hosted at XYZ than ABC, when measured, say, right at the firewall. However, it may be a useful bit of info to consider in assessing how urgent the situation may be for you.

There may not be any good data for your particular site, but let's say that via demographic or marketing surveys, for instance, you do happen to know something about your user demographics such as whether only a tiny percentage are on broadband, or if a large proportion are using older, less powerful PCs. Under those circumstances, the 'loading speed' as perceived by the user may be more a function of the users' connection characteristics and client machine performance than what your server can pump out.

Admitedly, my statements are based on a set of pretty special conditions, such as having such data in the first place, but something to consider if those conditions apply to you. Perhaps it may lend perspective to how urgent any user-perceived performance problem may be.

When we do performance testing, for instance, one of the things we consider is page responsiveness as measured by a fairly modest box over a dialup 56kb ppp connection.  Obviously, if the users' connectivity to your system is the bottleneck in the system, it actually keeps the heat off your server to some degree.

Joel's comments about Keynote are correct. This is one of the types of evaluations I believe Keynote offers -- perceived responsiveness via dialup connections (vs broadband/T-1), with the significant advantage of geographic dispersion of measurement points. Note this is distinct from actually loading your server.

May not apply in your case, just something else to keep in mind as you evaluate options.

anonQAguy
Monday, May 12, 2003

*  Recent Topics

*  Fog Creek Home