Skip to:

Identifying and mitigating an unknown performance bottleneck

  • Erlend


    Our and sites are hosted on the same dedicated server. Our .org community site is powered by BuddyPress.

    We frequently experience problems (I’d say at least once a week) with extremely slow loading and sporadic downtimes ranging from a few minutes too several hours at length. Exactly why this happens is beyond us, though we have some theories. For anyone kind enough to spare a moment, I’ve typed up a document detailing our server specs and “slowdown theories”:

    Your help figuring out how we can best avoid these problems we’re experiencing would be greatly appreciated.

Viewing 3 replies - 1 through 3 (of 3 total)
  • Your hardware specs are a little outdated, although in theory 1.5GB ram is sufficient to run a fairly busy box. I would want Xeon or better processors on a production box but failing that core quad or duo. I suspect with that present setup your hard drives may be a bit of a weak link as well.

    You new spec sounds much better, although not sure what you mean by keeping back hard drives for desktop? 5 drives would optimally be configured in a raid 1+5 (never can remember raid sets) mirrored then stripped, cuts the the total space by half then stripes across the mirror; with a fast 1500rpm drive read/write speeds will be very fast.

    Not sure about the DNS question although doubt that this is actually an issue in this scenario but my watchword to a degree is ‘not all eggs in one basket’ so I tend to keep certain things separated and park my domains one place then either IP point them and use the registras DNS as separate from the hosting box or specify a third party dns service such as ZoneEdit which is fast and reliable. If I need to move domains I can do so quickly and I’m also assured on running on dedicated reliable server for DNS purposes.

    Regarding the tweat I would perhaps not may too much heed to that, I tried to ping your site and received 100% packet loss, this may well be for the reason that the network is deliberately not returning icmp echo requests – technically a violation of protocols but I understand that it is somewhat considered now a security aspect and I guess blocking ping requests prevents echo flooding?

    Your mention of failure to connect to DB is an aspect that puzzles me, and think you need to start to examine your server logs in detail when events happen, ultimately this is the purpose of logs and the only way you can ever actually investigate and establish what is happening.

    As for splitting off the DB to another machine, yes it gives great performance benefits but not to the degree that will make a huge difference to your situation it would simply be overkill, unless you were experiencing really high loads, if you can do it it wouldn’t hurt but simply getting one server running at best and all your network connections as fast as possible is going to make the most noticeable difference.



    Thanks a bunch for all your feedback @hnla . With regards to the suspected DNS/Domain issue, do you know how we could go about putting this to the test?

    I’m glad to hear that our hardware is likely be at the root of this issue, because then it’s very manageable.

    Extra note: We might try out some of the techniques covered here:

    Not sure that focusing on DNS resolving is going to make much difference, in theory the root name servers and all the other DNS servers cache lookup results and typically only visit a SOA server once every 24 hours to update so once servers have cached results the registrar or primary dns servers play little part in things This depends on how you are managing things I guess, does your parked domain use the registrars DNS servers or do you reference another DNS server? you might want to try pointing your domain name to a different DNS server, try using ZoneEdit which allow up to 2? free records.

    Yesterday was the first time I visited your site (therefore neither I nor probably my ISP had a cached record of the IP) and my request resolved quickly so somewhere along the route a cached copy of the records were found and used, I also found the site fast and responsive but then again that is not a true test as you have that cache plugin running – out of interest have you disabled that to see whether it makes a difference, it’s not interfering in some manner, I believe that in certain configurations it can cause issues.

Viewing 3 replies - 1 through 3 (of 3 total)
  • The topic ‘Identifying and mitigating an unknown performance bottleneck’ is closed to new replies.
Skip to toolbar