This really is driving me crazy, so I decided to find out if anybody could produce an idea. I've got a small VPS running Centos 5, latest Apache and MySQL and PHP.
About 1% of my customers report very slow page loads, for static HTML pages. Around the order of .1 killerbytes per second. However they do end up the web pages to load.
It is usually exactly the same customers who're stricken, and they're always stricken.
The server isn't under CPU or disk load which is not memory or disk restricted.
Throughput becomes manifest pretty quickly and consistent for the non-stricken customers.
Stricken customers can download files extremely fast using their company VPS servers on a single host. Therefore it does not appear to become the host.
Has other people experienced an problem such as this?
- Tracert from affected customers is common
- Customers claim that they can have attempted various os's and browsers
- Customers originate from a variety of different geographic regions
It may be some craziness with DNS - esp. when the customers have been in a company atmosphere. Usually, DNS searches are cached, but when every request includes a DNS research, that may explain the behavior you identify.
If whatsoever possible, you can request individuals customers to consider their Computers (hopefully laptops) onto another network and then try to connect.
You might request these to install Firebug (a wordpress plugin for that Opera browser) - this provides a very visual indication of what's happening.
- Are these customers accidentally hooking up via
https(file encryption is heavy for any small VPS)
- Are you currently using large TPC/IP packet dimensions (not every hubs love individuals)
- Are you currently using content compression (possibly the rest of the customers have browsers that permit that, and just these customers get it disabled)