2

I realized that sometimes one of our servers monitored by Munin had a strange graphic of "Interrupts and context switches".

The server have a Xeon 5560 (8 cores) with 8gb of RAM.

At rush hour, the load average does not exceed 5 and 20% CPU usage.

Yet at the same time we record a very high level of Interrupts (the max : 100k). Munin seems at this point stop drawing the graph.

https://i.stack.imgur.com/ctg0Z.png

https://i.stack.imgur.com/3Ncmb.png

Is this a real problem of achieving the 100k?

What happens when this happens?

This can impact the performance of applications running on that server?

Thanks for your help !

ccarloss
  • 21
  • 1
  • What is this server? A web server running apache? A Java application server? A file server? Does this spike occur only during your rush hours? How many interrupts / context switches is your normal level? Please update your question :) – Janne Pikkarainen Sep 20 '11 at 13:18
  • Sorry ! Two apps on the server : Lighttpd and SmartFoxServer. Yes, this spike occur only during the rush hours (17h -> 22h). I put two munin's graphic into the original message which respond to your last question (i think) ! Thanks for your help ! ;) – ccarloss Sep 20 '11 at 14:57

2 Answers2

1

Inexplicably, the interrupts plugin explicitly limits the graphable value to 100K:

# munin-run interrupts config | grep '\.max'
intr.max 100000
ctx.max 100000

The script can be modified so as not to emit these options. Ideally, the plugin would be modified upstream so that a future version of munin benefits.

  • This bug was fixed upstream in 2016/2017, cf. https://github.com/munin-monitoring/munin/commit/c15cb55af392ce6d1b2160776a7c4171ae14c80a – Josip Rodin Sep 30 '21 at 19:31
0

This is a very wild guess: I've seen both context switches and interrupts heavily spiking when a server connected to a 100 Mbps network had negotiated a 100Mbps half-duplex connection instead of 100 Mbps full-duplex.

I have no idea what is SmartFox Server. If it's a Java-application, those can also under some circumtances cause context switch storm.

What OS you have? Is it fully patched? Is your server firmware up to date?

Janne Pikkarainen
  • 31,852
  • 4
  • 58
  • 81
  • I will check the first point, but i don't think it's that. SmartFoxServer is a Java application management multiplayer games. The OS is Debian 6.0 with the basic setup. what do you mean by fully patched?! The server's are new (put into production two months agos). Thanks :) – ccarloss Sep 20 '11 at 15:42