Eskimo North


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

          www


          • To: outages-list@eskimo.com
          • Subject: www
          • From: Robert Dinse <nanook@eskimo.com>
          • Date: Tue, 15 Jun 1999 03:48:48 -0700 (PDT)
          • Resent-Date: Tue, 15 Jun 1999 03:48:53 -0700
          • Resent-From: outages-list@eskimo.com
          • Resent-Message-ID: <"A_HLM3.0.mQ4.K-YPt"@mx1>
          • Resent-Sender: outages-list-request@eskimo.com

          
               I wanted to let people know what is going on with www and www2 (the main
          web server and the virtual domain server)...
          
               We have been working on upgrading the kernels and the libraries, first
          on www, and then when we get it stable, www2.
          
               Over time there have been a number of outages caused by NFS problems with
          the existing version of Linux.  An NFS request times out, it gets out of sync
          somehow, and then it can't get any files via NFS until the system is booted. 
          The 2.2.x kernels fix this with a new kernel NFS (rather than being implemented
          as a daemon in user space as with the 2.0.x kernels). 
          
               The glibc fixes an NIS problem allowing us to rely completely on NIS for
          user information which is much quicker than having to look it up in a password
          file, and also means the data will always be current. 
          
               We are running into a number of problems.  Although there are now
          compatibility libraries for UltraLinux, they are not sufficiently effecient to
          work for a I/O intensive application like a web server.  We tried to go this
          route on www, it didn't fly, the server was an absolute dog running this way.
          So it is necessary to recompile everything to use the new libraries. 
          Unfortunately, this is a slow tedious process and there are all sorts of
          interdependencies between various applications and packages that make it a
          rather slow complicated painful process. 
          
               Also, there is a bug in 2.2.9 that makes it somewhat unstable on a
          multi-CPU Sparc platform.  According to the mailing list, this had been fixed
          in 2.2.7, but we found out the hard way that it had not been.  So this is
          causing some instabilities.  We are in contact with David Miller who did the
          UltraSparc port of Linux in an attempt to resolve this issue. 
          
               On the plus side, CPU utilization on www dropped to about 15% of what it
          was under the old kernel, nfs, and libs.  The new kernel NFS does work, we've
          yet to have it die because of the server it's trying to get files in either not
          responding or other freaky problems.  The NIS in glibc also works.  So these
          two things have allowed for much more effecient operation of the web server,
          but unfortunately not without some stability problems.  We are working to get
          those taken care of before we upgrade the virtual domain server to the new
          kernel. 
          
          
          

          • Prev by Date: Eskimo Crash's
          • Next by Date: Seattle2
          • Prev by thread: www
          • Next by thread: www
          • Index(es):
            • Date
            • Thread