Daily Variation


There is a widespread belief that the web pages are slower at some times of the day, or on some days of the week. This seems quite plausible. We were very surprised to find that it isn't so. Last year, we analyzed the results of the beta test of the our monitoring system. We expected to publish "the best times to browse the Web". In fact, we found there was NO significant correlation between day of week or hour of day and page delivery times.

This finding is routinely confirmed by our OnTime Delivery monitoring runs. If you take a long list of URLs and go thru them non stop, you will finish in about the same amount of time (plus or minus 10%) every run. This is regardless of the time of day or day of week. It's what our monitoring station does several times a day.

Even if individual servers have peaks and troughs, for the Web as a whole they balance out. Perhaps routine maintenance at "quiet" hours balances out some of the load. We suspect that most page delays are caused by short term congestion and overloads on the server and its local routers and pipes. Hourly and daily server loads are likely a poor guide to delivery performance.

The ISPs we routinely use for our monitoring aren't causing significant delays. Results at other ISPs may differ. We were originally going to expand our monitoring network rapidly. We conducted some tests in the UK, California and Toronto, and found our local ISPs were above average. Last year, the receiving delays in the UK exceeded those from the sending servers. (Compuserve was abysmal). This year (August), the situation in the UK had much improved.

The delivery rate for all pages in our beta test was 4377 kbytes per hour. A year later, during the Great Web Race, it was 5324 kbytes per hour. This is a 22% increase. While the samples are not entirely comparable, it is encouraging.

If you are seeing consistent delays on all URLs at a particular time of day, look for a local cause. Both receiving and sending ISP will tend to blame "congestion on the Net" for delays, for obvious reasons. While it certainly exists, we feel it is currently overrated. For most servers, there are still considerable improvements to be gained at the sending site.

Return to Results of the Great Web Race Survey.