[ale] Building a Linux/Mysql Database server.

[Mike Harrison]

On Tue, 24 May 2011, The Don Lachlan wrote:
> Disk I/O is a common bottleneck, especially in databases. SSD is likely
> overkill; SCSI or SAS (Serial-Attached-SCSI) in RAID10 should be fine but I
> would avoid standard SATA.
>
> I'm guessing (hoping!) that expanding your DB from 45G to 600G will also
> include some tuning of the database and application. As such, performance
> will change and your current bottleneck(s) may be resolved.
>
> Without more information about your applications and what the current
> performance is or what the _expected_ performance will be, it's impossible
> to be specific. Is your application focused on reads from the DB? Writes?
> 50/50%? How much data? How quickly? Are constantly you re-reading the same
> data? How many CPU cycles for a standard query? Can't answer your questions
> without more information. Give us some numbers, man!

Yeah. What Don said.

Point blank: Buy as much ram as you can stuff 
in.. and as fast of a drive system as you can. Hardware is cheap.

Then you'll spend 172 man hours finding the 3 queries that need to be
rethought, and another 200 to find the slow join because the key 
fields are different  formats (one is varchar(20) in UTF8 and the other is 
a bigint, the third is char(20) in ASCII, all containing a number {for 
example}).

And lastly, the 30 minutes beating yourself over the head because you just found what looks 
like a piece of your code that seems to thev been copied directly from
thedailywtf.com (originally from experts-exchange.com).

Oh, and don't forget that the foo.java module opens up 30 database 
connections, for each web page refresh, with cursors and locks, that is 
persistent for up to 3 minutes.

My point is: Your "hardware" performance issues might not be.