Conference ulysse::rdb_vms_competition

1040.0. "Sybase - joins and servers" by MJBOOT::WEINBROM (Programming Dinosaur) Sun Dec 15 1991 21:02

    My customer had selected Sybase as their database of choice for their
    VAXes.  When they found out what it would cost them to roll it out to
    all of the plants, Rdb, with its no-additional-charge run-time became
    very attractive and they switched.  Score one for the good guys.

    Anyway, they still think that Sybase is great and every so often hit me
    with features that it has like stored procedures, etc...
    
    There are two about which I was curious:
    
    1. Can Sybase join a table across systems?  
    
    For example, each plant will have it's own database.  Each plant runs this
    application and so each has the same database schema, but with
    different data.  They claim that Sybase could combine the data in the
    tables to that Corporate could look at all the plants as a whole.  
    
    This is not joining two tables over a common key, nor projecting them. 
    This is combining the records from the same table in each database into
    one 'virtual' table like this:
    
    	Sample table at coroprate contains
    		Records (from plant 1's sample table)
    		Records (from plant 2's sample table)
    		Records (from plant 3's sample table)
    		Records (from plant 4's sample table)	etc...
    
    What really piques my curiosity is that the notes here seem to
    indicate that Sybase doesn't yet provide 2-PC.  You have to roll your
    own and so the scenario above seem unlikely.
    
    2. Can someone explain the Sybase virtual server in general terms and
    why it is special?  (I'll save someone the trouble "Besides the fact 
    that it sounds good in advertisements...")
    
    Having done lots of system programming on a VAX, I understand how one
    might implement a server running in a single virtual address space
    using asynchronous I/O and ASTs, but I can't understand how one could
    port that across multiple platforms without significant effort, assuming
    that such facilities even exist in other O/S's and without asynchronous
    I/O which does not exist in the language of portable products, C.
    (Without which I haven't a clue as to how you'd get good performance 
    if you did port it)
    
    Are there other benefits besides reduced memory utilization?
    Are there gotchas? (Like extra context switches, bottlenecks, etc...)
    
    Inquiring minds want to know!
    
    Steve