Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I am testing my Linode Ubuntu 14 64bit server out, it's the most basic server available from them. I am using Apache Benchmark to test the server, as well as a multithreaded script I wrote in Python, but more on that later. Using AB I noticed I get around 7k requests per second when run locally from the server itself but only around 15 when run from another network/internet. Response time is also about 150 ms for 1000 concurrent connections, locally. Remotely the response time is about 1.5-2.5 seconds for 100 concurrent connections. The network I'm running the remote tests from has plenty of bandwidth and the computer I'm running it from has plenty of ram and processor speed; it's a fast business network. I even tried 2 other networks from around the U.S., on two other computers and the speeds are all about the same.

When running my multi-threaded script I notice it hiccups as soon as I try more than just 100 concurrent requests, this is from external networks. I haven't tried my script locally on the server yet as I either need to upgrade my Python on the server to 3+ or change my script to be 2.7 compatible. I tested this locally and get 150 ms response time when running the script with up to 1000 multithreaded connections, it's simply using urllib2.

I am testing this against nginx directly (a static file), a pywsgi app that's behind nginx, as well as directly to the pywsgi. The pywsgi app has a simple route that replies with a basic response so it should be fast. Not surprisingly nginx->pywsgi provides the best results, probably due to how it buffers the requests. Is there something that is specific with Linode's network that is causing this issue? The orders of magnitude difference between internal and external tests makes me wonder what could be the cause. The only other thing in the way would be iptables firewall, just filtering in http/s and ssh.

dmesg has no information pertaining to my tests.

share|improve this question
2  
You mention that the computer you're launching the test from is beefy, but what does the testing machine's process monitor say about resources used? The first time I did testing, I ran into that classic problem benchmark testing a server: the machine I was running the tests from couldn't handle the responses fast enough. It might explain why testing on the server itself returns quickly, though I still suspect network. See my comment below on Mark's answer. –  msanford Jan 16 '13 at 21:31
add comment

1 Answer

Sounds like there is a network bottleneck somewhere between your desktop machine and your Linode machine. Some suggestions:

  1. Compare ping times from your desktop to your Linode and other locations on the internet, like google.com and yahoo.com.
  2. Use traceroute your-linode-ip from your desktop to see if that helps find the bottle neck.
  3. Try your Linode using http://www.webpagetest.org/ selecting both East Coast and West Coast locations and see how they vary.
share|improve this answer
1  
+1 Indeed this would seem the most obvious place to look. Another thing to check is that the testing computer isn't being mistaken for a DDoS bot and being rate-limited by one of the firewalls (outgoing, inccoming, on the server itself). –  msanford Jan 16 '13 at 21:24
1  
Ddos might be the answer, perhaps linode network limits this.. Hmm, let mw try these ideas, thanks for your responses! –  scape Jan 18 '13 at 0:40
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.