Database Optimization

[nospammmer]

Hello group,

I have a rather general but interesting inquiry that is related to PHP
and I hope this is the appropriate place to post it.

I'm looking for a way to improve dramatically the performance of my PHP
application. The application is getting slow as it is taking more load.
It is performing a very high number of queries to a database, and I
believe that this is taking up most of the ressources.

I'm trying to figure out how I could cache the content of the database
and prevent that many queries to be performed.

What I would like to do is cache all the content of the database in
memory, so that I could access it directly through my PHP application
without querying the database and saving precious ressources.

The database is quite small, 15 - 20 mB and it's size is constant (it
does not get bigger over time). 92% of the queries are SELECT, only 8
percents are UPDATE, DELETE and INSERT.

So, my question is, is it possible and recommandable to place 20mB of
data in shared memory in order to prevent queries to the database? (all
updates, deletes and inserts are performed both in the database as well
as in memory)

Or would I be better to place a copy of the database on a ramdrive?

Other info:
I have a server which runs both the PHP application and the MySQL
database. It has 1GB of RAM. The database receives 250 queries / sec.

Thank you in advance for your kind help

[Erwin Moller]

no********@gmai l.com wrote:

Hello group,

I have a rather general but interesting inquiry that is related to PHP
and I hope this is the appropriate place to post it.

I'm looking for a way to improve dramatically the performance of my PHP
application. The application is getting slow as it is taking more load.
It is performing a very high number of queries to a database, and I
believe that this is taking up most of the ressources.

I'm trying to figure out how I could cache the content of the database
and prevent that many queries to be performed.

What I would like to do is cache all the content of the database in
memory, so that I could access it directly through my PHP application
without querying the database and saving precious ressources.

The database is quite small, 15 - 20 mB and it's size is constant (it
does not get bigger over time). 92% of the queries are SELECT, only 8
percents are UPDATE, DELETE and INSERT.

So, my question is, is it possible and recommandable to place 20mB of
data in shared memory in order to prevent queries to the database? (all
updates, deletes and inserts are performed both in the database as well
as in memory)

Or would I be better to place a copy of the database on a ramdrive?

Other info:
I have a server which runs both the PHP application and the MySQL
database. It has 1GB of RAM. The database receives 250 queries / sec.

Thank you in advance for your kind help

Hi,

I am unsure if placing the database in memory will seriously increase it's
performance: you'll have to test that.

If you database is using its time on scanning tables and joining, and
conversions, etc etc, the time trimmed off could be disappointing.
If the database is non-stop reading files, it could help.
Hard to say.
Which database do you use?
(!)But before you go all that way, did you try some more 'old-fashioned'
optimizations?

Some ideas:
- Try to figure out which tables are scanned a lot, and place indexes on the
relevant column(s).
(If you use Postgresql, try EXPLAIN-command for help)

- Does you DB and your code use Prepared statements?
They can help a lot, especially when the queries are complex.

- If 50 of the 250 queries/sec are the same selects that don't change, you
could try some smart caching.
eg: If a popular query is to get the latest 20 this-or-that, with all kind
of joins on other tables, you could shedule that query every 15 minutes,
and safe the results in a file. Then include the file on the pages where
you need it.
Alternatively: you could just update the file, whenever you know a relevant
table is changed.
(What makes the most sense is up to you to decide of course.)

This kind of optimalization can make huge differences.

In general: Try to figure out which queries are executed a lot, and start
there with prepared statements/indexing/caching-to-file.

Hope this helps.

Good luck!

Regards,
Erwin Moller

[nospammmer]

Thanks for your reply.

I use a MySQL database that is properly optimized. All the indexes are
set correctly and used.

Most of the requests are simple queries using a unique ID and returning
only a single result. There is almost no joins or complex joins.

- If 50 of the 250 queries/sec are the same selects that don't change, >you could try some smart caching.

Unfortunately, most of the the queries are different.

I can give an example:

An user table with around 4000 users. It is possible to consult other
user's information. So a lot of queries are made on single records.

I tested placing a few records in memory with shm functions, and it
was of course, blazingly fast.

But I'm wonderig how the system reacts with higher volume of data, and
what would be the best way to do this.

Thanks

[Darkstar 3D]

Testing it could be easy. Have a link on the page that peeps currently
use asking them to test the effectiveness of the new code. I have
experienced that almost 80% of my users will go in and test the new
stuff, just for curiosity maybe. So my testing on new code is normally
completed within a few days, with no impact on operations.

[nospammmer]

Thanks,

But I would rather stick to profiling. It's much more precise.

Perhaps I should rewrite the exact question: is it possible and
recommandable to load large amount of data (20mB) into shared memory.
If yes, what is a good way to implement it.

Thanks again for your help

[Chung Leong]

nospamm...@gmai l.com wrote:

Hello group,
The database is quite small, 15 - 20 mB and it's size is constant (it
does not get bigger over time). 92% of the queries are SELECT, only 8
percents are UPDATE, DELETE and INSERT.

8% is still a significant amount of writes. I wonder if your database
is running into locking problems. Not an expert in MySQL, but I've
heard that its locking mechanism isn't that great. If a table is
constantly being modified, then queries on it could be often stalled.

[nospammmer]

I do not use any manual locking as I do not need atomic transactions.
So I think locking shouldn't be an issue.

I believe that the source of the problem is the big amount of queries
generated by the mass of users. And that's why i'm looking at shared
memory caching.

[Dikkie Dik]

Maybe my article at http://www.w-p.dds.nl/article/wtrframe.htm
describes something useful. Especially the section on lazy collections
can be interesting.

Best regards.

no********@gmai l.com wrote:

Hello group,

I have a rather general but interesting inquiry that is related to PHP
and I hope this is the appropriate place to post it.

I'm looking for a way to improve dramatically the performance of my PHP
application. The application is getting slow as it is taking more load.
It is performing a very high number of queries to a database, and I
believe that this is taking up most of the ressources.

I'm trying to figure out how I could cache the content of the database
and prevent that many queries to be performed.

What I would like to do is cache all the content of the database in
memory, so that I could access it directly through my PHP application
without querying the database and saving precious ressources.

The database is quite small, 15 - 20 mB and it's size is constant (it
does not get bigger over time). 92% of the queries are SELECT, only 8
percents are UPDATE, DELETE and INSERT.

So, my question is, is it possible and recommandable to place 20mB of
data in shared memory in order to prevent queries to the database? (all
updates, deletes and inserts are performed both in the database as well
as in memory)

Or would I be better to place a copy of the database on a ramdrive?

Other info:
I have a server which runs both the PHP application and the MySQL
database. It has 1GB of RAM. The database receives 250 queries / sec.

Thank you in advance for your kind help

[nospammmer]

That's right. It is interesting. This article touches the kind of
structure I would need for my caching solution.

As you mentionned, I want the class to control the data, no matter if
it comes from the database or from the cache. But I want it to control
all the updates, inserts, and deletes also as my data will never be
modified from outside of the application. The cache will simply be
updates at the same time as the database.

What I would like to discuss, however, is technically how to index,
retrieve, update and store large amount of data in shared memory.

Thanks for your help

[NC]

nospamm...@gmai l.com wrote:

I'm looking for a way to improve dramatically the performance of my PHP
application. The application is getting slow as it is taking more load.
It is performing a very high number of queries to a database, and I
believe that this is taking up most of the ressources.

I'm trying to figure out how I could cache the content of the database
and prevent that many queries to be performed.

What I would like to do is cache all the content of the database in
memory, so that I could access it directly through my PHP application
without querying the database and saving precious ressources.

The database is quite small, 15 - 20 mB and it's size is constant (it
does not get bigger over time). 92% of the queries are SELECT, only 8
percents are UPDATE, DELETE and INSERT.
It sounds like you could improve you performance by using query
caching and/or better indexing... Read "High Performance MySQL"
by Jeremy Zawodny; it should give you some ideas...
So, my question is, is it possible and recommandable to place 20mB of
data in shared memory in order to prevent queries to the database? (all
updates, deletes and inserts are performed both in the database as well
as in memory)
Yes, it is possible. MySQL supports HEAP tables that are stored in
memory. But you still need to figure out a way to save those tables
on the hard drive, because HEAP tables disappear when MySQL server
stops or reboots.
Or would I be better to place a copy of the database on a ramdrive?
Again, you can do that, but you still need to make sure your database
is synchronized to a hard drive somewhere...
Other info:
I have a server which runs both the PHP application and the MySQL
database. It has 1GB of RAM. The database receives 250 queries / sec.

It appears a very manageable load... How many concurrent connections
are you handling?

Cheers,
NC