Re: Database link performance

From: John P. Higgins <jh33378_at_deere.com>
Date: Thu, 29 Oct 1998 23:04:33 -0600
Message-ID: <363948E1.907701E7@deere.com>

Well, if my 1,000,000 row table on the win95 database had to be joined to a 100 row table on the NUMA-Q, I think I would want to do the join on the 'slow' box. I'd rather ship 100 rows than 1,000,000. And a T1 line is slooooow compared to local access.

By the way, you did finish your first contribution with 'you do the join on the faster database'. Isn't this also a generalization? If the two tables are of comparable size, then the fastes machine is probably a good choice. (Unless the faster machine is also heavily utilized?)

So, okay, we need to think about each distributed join and compute all the factors to know where to route the work.

Ross Mohan wrote:

    John P. Higgins wrote in message <363530F3.135E90DD@deere.com>...
    Don't worry about the faster database. The slowest database around is
like the spped of light compared to the network traffic.
    ||  Does this apply to a win95 single disk 133MHz box over a T1 line to
a 8-way NUMA-Q Sequent box?  Generalizations are generally dangerous. ;-> In
this case, I'd be creating views on the Sequent to force the bulk of work
there....just a thought, no flames please....
    Plan the join to minimize the network traffic.
     Gee, I thought that's what I said.

    Ross Mohan wrote:

        Mostly you want to avoid doing JOINs over the link.  You can force a
join to
        occur on one db or the other by creating appropriate views there.
That way, JOIN
        rows are not shuttled across the network, just the result set.

  Received on Thu Oct 29 1998 - 23:04:33 CST