[PLUG] PLUG resume database project

Alex Daniloff alex at daniloff.com
Thu Jun 6 16:46:37 UTC 2002


Hello Linux folkz,
One of our PLUGers David and me started to work on PLUG resume
database project for unemployed folkz.

So far we have set up replicated MySQL database to store your resumes.
Pretty soon we are going to start developing resume search and output
interfaces. 

But there are still some questions regarding DB configuration.
Please see the description below and asked questions.
Thank you in advance for any thoughts or sources of information.

If somebody would like to participate in this project please call me
at: 
503-547-2204 during normal business hours. We'll plug you in.  

Data safety and redundancy problem is solved by replicating one
database on another (master - slave relationship).
The server simultaneously runs two MySQL databases on different RAID0 
partitions. 
Both databases provide output (select queries) to the users and client
API 
programs through the load balancing interface. 
But user input (insert and update queries) can be done only on master 
database. That propagates them to slave database with adjustable delay
time 
between slave-master reconnection. In case of one RAID0 failure,
remaining 
database becomes the master and continue to handle load until failed
RAID0 
drive is replaced and array is rebuilt. 
This approach eliminates nessesity in tape drive and periodical backup

procedures.

There are still some questions remain: 
What is better to use for data storage DB datafile on top of existing 
filesystem or raw device partition?  
InnoDB ACID compliant tables in MySQL can use both approaches.
But in case with raw device partition InnoDB can perform unbuffered
IO.
As it said on their web site http://www.innodb.com - using raw device 
partition should improve performance on some UNIXes. 
But what is about Linux?
If it's not true for Linux what filesystem is the best to home InnoDB 
datafile?
Does DB performs better when it has smaller or bigger DB datafile or
raw 
partition? 
What is more practical to assign one big e.g. 10GB datafile / raw
partition 
for DB or start with smaller 100M datafile / raw partition and add
additional 
datafile / raw partition when the first one becomes full?

If you have a better idea for safe data storage, 
integrity and operation, please let us know.

Alex






More information about the PLUG mailing list