5

I've read through other threads on performance differences between RHEL 6 and RHEL 5, but none seem a tight match to mine.

My issue manifests itself in slightly slower average response time (20ms) per request. I have about 10/10 servers of the same hardware spec with CentOS 6.1 and CentOS 5.6. The issue is consistent across the group.

I am running Ruby on Rails with Passenger.

  • Apache config is identical (checked out from the same SVN repo)
  • Ruby and Passenger are identical builds.
  • Application is identical and being served traffic round robin.
  • mod_worker

An interesting clue from server-status: The CentOS 6.1 servers have a steady 20-40 threads in the "Reading Request" state while the CentOS 5.6 servers have around 1. I'm graphing this so I can see it trend over time.

I also have a bunch of much newer machines that are significantly faster and are running CentOS 6.1. They dust all the older machines in response time, but I can see they also have a steady 20-40 threads in the "Reading Request" state. This makes me believe I can get their response time down, if I can figure out what is holding up these requests.

My gut is telling me that I need to tune some network setting in sysctl, but I haven't figured it out yet.

9
  • ooh. interesting one. There are bonkers loads of metrics and settings that you could pull here to find out.
    – Tom
    May 9, 2012 at 0:06
  • How many requests per second are you serving avg? whats the current cache and buffers?
    – Tom
    May 9, 2012 at 0:14
  • each individual server gets to around 40/sec
    – daniel
    May 9, 2012 at 0:35
  • there is a good amount of free memory, if that's what you are asking CentOS6: MemTotal: 8061920 kB MemFree: 1898468 kB Buffers: 92748 kB Cached: 2422552 kB CentOS 5.6: MemTotal: 8174968 kB MemFree: 1107736 kB Buffers: 147204 kB Cached: 4750620 kB
    – daniel
    May 9, 2012 at 0:37
  • 1
    Apache versions? May 9, 2012 at 2:13

2 Answers 2

0

One FAQ in this area is that memory mgmt in RHEL6 is different and mulit-threaded apps can suffer. Try running apache under MALLOC_ARENA_MAX=1 and see if it makes a difference.

0

Wild shot #1: Any chance you're logging hostnames instead of IP addresses? If yes is that on both? If yes is one of them running nscd?

Wild shot #2: Is LDAP involved in any way there? I.e. is the apache or any other component of the request running as an ldap user?

You must log in to answer this question.

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