Re: Sybase/Oracle table size limitation / Largest production database

From: Pablo Sanchez <pablo_at_ppp-mew.corp.sgi.com>
Date: 1995/11/02
Message-ID: <47at01$rkt_at_murrow.corp.sgi.com>#1/1


In article <47altq$1kg_at_empire.texas.net>, smsi_at_texas.net writes:
> I wasn't clear enough. Hold two variables constant, vary the third
> and measure response time. So for instance, what happens to
> response time for an insert for ten simulaneous clients requests
> for 500K/1M/10M/100M rows of 500 bytes each? In my example the
> choice for number of clients was arbitrary, but reasonable.

Inserts should be pretty trivial. Updates may be where you can be wooly. What you need to ensure is that the update is following an index.

In a small table (say it's only 20MB or something), if an update turns into a table scan, then it's not that bad.

When you are dealing with monster tables (500MB and greater... that's what I consider monster...) then a table scan is obviously more time consuming.

How much so? Easy to compute: figure out rows per data page for the entire table. Consider each data page is an I/O. On our current disks we get about 100 I/O's per second so from there you can figure it out... dunno what type of hardware you have but you can get those numbers from your vendor.

I guess the short answer to your question is:

  If your queries are indexed, then you're not going to notice much   of anything. If they are not indexed... :-)

> How often are you dropping and recreating your index???

About once a quarter... as long as six months some times...

> And how
> static/dynamic is your data, i.e. % of rows changed per day and
> whether add/delete/change.

Yesterday 9,260 rows got added: .09% I don't have visibility into how many got updated. I'm pretty sure that none get deleted.

> Thanks!

Yur welcome.

> I'm trying to get a feel for what people are finding "reasonable"
> in terms of production performance and maintenance.

If your queries are optimized (using indexes) then you should expect sub-second response. Of course if you are bring back tons of record then it's dependent on that...

It's very critical to specify your clustered index for something huge.

Pablo Sanchez              | Ph # (415) 933.3812        Fax # (415) 390.2821
pablo_at_sgi.com              | Pg # (800) 930.5635  -or-  pablo_p_at_corp.sgi.com
===============================================================================
"I am accountable for my actions."
  • pablo
Received on Thu Nov 02 1995 - 00:00:00 CET

Original text of this message