1

Our company is hosting a website out of our office, it hosts the site, some downloads, and a ton of training videos. The server itself is a Poweredge1900, running Web Server 2008 64bit, Dual Quad Core, 16GB RAM.

The building we are in provides us with an internet connection as part of the rent. Running a speedtest on any given day usually results in at least 14Mb/s up and down, 5ms ping, so not too shabby. Out building is a sister building of a university, so its half tech businesses and half classrooms, and the bandwidth is provided by the University main building via fiber.

We are having issues with the internet being deathly slow sometimes, and our customers complaining about our training videos being choppy, slow to load, etc.. Running a speedtest during these times still results in fast speed, and low latency. The server also shows no additional load, or indication that is is having issues.

Basically, I'm fairly certain that the issue is not with us, as we have 2 different public IPs coming to us, with our back office network on one physical network, and the website all alone on its own, and performance declines on both simultaneously. How can I gather some data to prove its the connection?

Improve this question

4 Answers 4

Reset to default
4

We had an issue with an ISP upstream "shaping" traffic and causing issues with our VPN traffic between sites.

http://www.measurementlab.net/measurement-lab-tools#diffprobe

We used the ShaperPrope tool to identify the upstream ISP and things seem to have improved once we could identify the ISP involved. They had denied any such manipulation.

At the time, our main ISP suggested some others were shaping video and audio streams

Good luck

Improve this answer
2
  • I just ran the tool, and it detected no traffic shaping.
    – DanBig
    Sep 3, 2009 at 14:29
  • +1 for this answer. Perhaps the traffic shaping only kicks in at under-load periods.
    – tomfanning
    Sep 3, 2009 at 14:33
1

Expanding on my comment on Dave M's answer, if you're dealing with traffic shaping here it may only be being applied at certain times of day, or when the pipe reaches a certain level of traffic for so long.

Another theory - your speedtest result might be being interfered with by an upstream transparent proxy. This theory assumes that the congestion would have to be beyond wherever this theoretical transparent proxy might be.

Improve this answer
1
  • This is what I'm going to look into next, hopefully the admin for the university network can shed some light on this, it could be that there is shaping going on but only while we are streaming content.
    – DanBig
    Sep 3, 2009 at 17:37
1

A nice and simple way to do it in my opinion that is free...

  1. Get a machine outside the network on the internet
  2. Allow it to have a "hole" through the firewall into your server...even if that means a VPN connection...that'd be "ok" (not ideal though)
  3. Get Qcheck free (http://www.ixchariot.com/downloads/qcheck.html) and install on both client and server.
  4. Put http://www.numion.com/stopwatch/index.html on your client's bookmarks.

Now...do a baseline of both Qcheck (throughput tests and response times) and numion's stopwatch loading timer on the website during a normal time. Do it 2 or 3 times to get accurate measurements for a baseline.

Then...when there is a "slowdown" reported run Qcheck and the timer site again and see what the differences are.

THIS ^^^ will at least validate your issues.

For troubleshooting the root cause, you'll need to start breaking up the segments (server to firewall, server to switch, switch to firewall, firewall to ISP, etc.) and running similar tests or checking for packet loss, etc. on those segments to try and narrow down where the culprit lies.

Improve this answer
0

We've found out that the ISP way upstream was doing some throttling/shaping. Thanks for the answers.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .