RE: My LVS Server is on-line

[Ryan Hulsker <RHulsker@xxxxxxxxxxxxxxxxxxxxxxx>]

>>
>> about 110K of these where scripts which hit an Oracle database.
>
>are any of these people writing to the database? How do you synchronise/update
>the databases that the realservers see?

        Well there is one copy of the database that resides on a Sun E420R, the web servers and the database server are all tied together with a 100Mbit full duplex switch, and the web servers access the database over the network.  Our site is all developed in Perl and we use the Perl-DBI modules to hit the central database over the network.

        You could do the same with a powerful linux box running MySQL or something instead of Oracle.  The company wanted this whole system to be able to scale to 10 times the capacity of the old system, and had the budget for it so we decided to go with the Sun box.  Yesterday, the Sun (4 CPUs) was running all day at a load of .2, so I think we will be ok from a database point of view.

        There are some files that are generated by the web site code that need to be accessible by all of the web servers.  We have taken 2 different approaches to this.

        The first approach is to write these files out to a central NFS mounted directory so that the files are instantly available to all of the web servers.  We do this with data that is not accessed too often, but changes regularly, and needs to be instantly available. Generally client reports that are generated the first time they are accessed in a day, and then they are cached to this central NFS mount.  A client may hit these files 10 times a day max after they are generated.

        The second approach is to copy the files to each web server so that they all have a local copy of it.  Some of our site is generated by our staff using an interface that we built.  These files are small, but are hit any where from 1K to 100K times a day, and they only change every couple of weeks or so.  When a staff member edits these files they are told to expect the changes to take 5 min to propagate to all of the servers.  When they are generated they are written to the shared mount, and then a process on that central box rsyncs them to all of the web servers every 5 min.

        The perl code itself is held in a CVS server, and I have an interface where I can push the code from the CVS server into all of the web servers at once.  The developers have the same interface that pushes the code into a development server.  So they can make edits on their own linux box, check the changes in and push them to the test server.  once they get approval from QA or their team lead he (or I) can update the production server.

Ryan Hulsker
Unix System Admin
Service Intelligence.com