Re: Databases and Enterprise Storage

In article <952549672.29272.0.nnrp-09.9e984b29_at_news.demon.co.uk>,   "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote:
>
> One of the big reasons for buying the 'Enterprise Storage'
> solution is that it's all a nice big black box and you don't
> have to worry about what's going on inside.
>
> Most DBAs seem to spend a lot of their time trying
> to find out what is going on inside the database.
>
> I sympathise with your common sense DBA perspective,
> I once spent several weeks trying to explain to management
> that the problem was clearly inside the big black box and
> they should try a produce a map of which bits of wire
> were sending data to which discs.
>
> One of my rules of thumb - if it's an Oracle database avoid
> any discs bigger than 9Gb: and I bet your Enterprise
> Storage boxes are talking about 36Gb+.
>
> --
>
> Jonathan Lewis
> Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk
>

True - I hate black boxes. I want to know not only how it works, but why. Also more spindles generally equal better response time.

One experience with partially shared dasd involved a 198 hour process reduced to 12 hours by doing two things: Changing the dasd configuration from RAID-5 to Raid-1 and moving around datafiles (about 80 hours saving) (Raid 0+1 would be even faster, but EMC does not really support this) and tuning Oracle memory and init parms (about 100 hours). No sql or database layout changes were made!! (long story as to why) Proving the DASD was even a problem involved over 2 man months of effort!!!

When the DASD was dedicated, i/o times where 8-15ms for writes. Acceptable. Now, as only a few additional servers have been thrown on the frame, they average 100+ and are growing as the EMC's prefetch - caching algorthim orients intslf toward file and print server type of operations rather than the random I/O patterns of a normal DB and contention is encurred.

Of couse cost, ownership, flexibility, scalability, ability to troubleshoot, monitor and performance tune are all significantly negatively impacted. And for this we get the benefit of..........

I like the concept and the benefits of big dasd, IF it is dedicated to a server or group of servers meeting a specific business need. But this shared stuff has me really worried.

I say if we wanted to do this stuff on a mainframe, we should have just started there. My kingdom for someone who can answer this question - What are the benefits of client server? How do we preserve its potential benefits while reducing its inherence drawbacks? (not the reverse?)

I turn the soapbox over to others now...... :-)

--
Doug Coan
Oracle Certified Professional DBA
AEGON USA
dcoan_at_aegonusa.com


Sent via Deja.com http://www.deja.com/
Before you buy.