Eskimo North


          [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

          Virtual Domain Server...


          • To: outages-list@eskimo.com
          • Subject: Virtual Domain Server...
          • From: Robert Dinse <nanook@eskimo.com>
          • Date: Thu, 5 Aug 1999 01:06:05 -0700 (PDT)
          • Resent-Date: Thu, 5 Aug 1999 01:06:08 -0700
          • Resent-From: outages-list@eskimo.com
          • Resent-Message-ID: <"Sd0it1.0.zY6.mNKgt"@mx1>
          • Resent-Sender: outages-list-request@eskimo.com

          
           
               I felt a bit more detail was necessary...
          
               We are trying to get the virtual domain server upgraded to glibc and
          also the 2.2.10 kernel.  The primary reason for this is effeciency, the main
          web server went from almost CPU saturated to approximately 10% CPU occupancy.
          
               Some complaints we had about ftp stopped.
          
               The news server went from about 50% occupancy to about 10%, the peak rate
          of news processing (when Sprint is actually in the mood to feed us) just about
          doubled.  The disk occupancy also appeared to drop significantly.
          
               Clearly the newer kernel and libraries are a massive improvement and we
          want to propogate them to the rest of the machines here. 
          
               When upgrading www, we ran into a problem with init, so Aaron attempted to
          avoid this by building a static init and sh first on www2 (the virtual domain
          server).  Something went wrong and the static versions didn't get put in place.
          
               During the copies of the libraries the machine crashed, and it apparently
          crashed while there was a lot of pending disk I/O because fsck on the
          partitions found massive problems.
          
               After the fsck finished, we moved the old libraries back but the machine
          would still not complete the boot-up into multi-user mode.
          
               I ended up having to restore most of the root and /usr partition from
          tape.
          
               We don't want to cause another outage like this, so I have obtained
          another machine and we are just going to basically load that machine from
          scratch with all the newest stuff, and then when we get it operational replace
          the existing box and re-load it and use it to replace another machine, and so
          on until they're all done. 
          
          
          

          • Prev by Date: Virtual Domain Server Crash
          • Next by Date: Voicemail outage early Friday morning
          • Prev by thread: Voicemail outage early Friday morning
          • Next by thread: Virtual Domain Server Crash
          • Index(es):
            • Date
            • Thread