[dba-SQLServer] What happens when

JWColby jwcolby at colbyconsulting.com
Tue Sep 19 11:31:46 CDT 2006


In fact I was trying to run a raid 5 directly on the motherboard raid, using
(4) 320 g hard drives.  LOL, boy does THAT suck.  The write speed for a 4
disk array was about 5 mbytes / sec.  Why bother offering it?  

So I am searching for a raid controller "in my price range".  I have pretty
much narrowed the choice down to the PROMISE SuperTrak EX8350 - 

http://www.newegg.com/product/product.asp?item=N82E16816102076

Or the Areca ARC-1220

http://www.newegg.com/product/product.asp?item=N82E16816131004

Both are some serious money for me - who'd think a raid card would cost
twice or three times the motherboard?  Economies of scale I guess.  

I would bite the bullet and get the Areca which is pretty seriously fast,
particularly on the upper end of the disk load scale, but they apparently
have some compatibility issues with motherboards.  I can't find that they
have ever tested it on the NVIDIA 5x motherboard chip sets.  Promise OTOH
seems to have tested it with that chipset as well as the 750gb version of
the Seagate disk I am using (I'm using the 320gb drives cause they are
seriously cheap right now).

In the meantime the raid 5 speed sucked so bad I went back to two mirrored
320 gb drives and therein came the problem.  Suddenly everything is fast but
no (physical) room for error.

At any rate, I cannot get the SQL Server process to stop.  I have stopped
the server, even rebooted, but when I come back up and go into EM, as soon
as I click the database icon EM locks back up again.  I don't know what to
do here.  8-(  The drive is full, it is never going to write any more info
out there, but apparently it isn't ever going to figure out that it is
locked up either.

Speaking of which... Answer me this...

I built a utility in Access to open the 10 gb raw text files, strip out all
the "padding".  The data was comma delimited, but then inside the quotes,
the data was padded to a fixed width.  So I stripped out all the padding,
then wrote it back out again.  This reduced the files from 10g to ~6.5 gig,
so you can see there was some significant padding going on.

Next I imported all the data.  Having done that I tried to add an autonumber
(4 byte) PK which is where the hang-up came in.  The original data was
~100gb, and the processing of adding a new 4 byte field and indexing it
pushed the data file size up over 300gb.  How is that possible?  This is not
the log file I am talking about, the log file is on another drive.  It seems
that adding an indexed 32 bit field would be a fairly trivial operation,
even given the 65 million records it is dealing with.

John W. Colby
Colby Consulting
www.ColbyConsulting.com

-----Original Message-----
From: dba-sqlserver-bounces at databaseadvisors.com
[mailto:dba-sqlserver-bounces at databaseadvisors.com] On Behalf Of Francisco
Tapia
Sent: Tuesday, September 19, 2006 11:56 AM
To: dba-sqlserver at databaseadvisors.com
Subject: Re: [dba-SQLServer] What happens when

I'd agree, I had one of my staging servers run out of room last weekend and
was lucky to be able to clean out some space.  Because of the data your
pushing you may want to consider better hardware to help keep up with the
demand for the data.

IIRC last time we spoke you were running Raid 0? for these data files?



On 9/19/06, artful at rogers.com <artful at rogers.com> wrote:
> Your system tray should have an icon that looks vaguely like a server with
a green arrow indicating that it's running. Double-click that and see if you
the SQL dialog appears. If not, then I think you have no choice but either
to reboot or kill the process itself and start over.
>
> Given the size of the databases you're working with, Oracle might be a
better choice since it offers tablespaces (which allow you to partitition a
table and place chunks on various drives and servers). But perhaps you can't
go there. I sure wish MS-SQL offered that feature. The last gig I worked on
had 8 databases and the growth rate was expected to be a TB per year. Oracle
would have let us do it all in one DB, with tablespaces pointing to the 8
servers, but since it was an Accenture gig, MS-SQL was the DB. Good thing
too, since my Oracle skills are substantially behind my MS skills.
>
> Anyway, there should be no need to run the process you are talking about.
Just mark the column(s) PK and that's that. SQL will build the index
automatically. You could in theory create a new index containing the same
key(s), but what for? What's wrong with the one SQL creates for you?
>
> Arthur




More information about the dba-SQLServer mailing list