## Thinking About Databases, Efficiency and Technology

From: "andrew cooke" <andrew@...>

Date: Sat, 22 Nov 2008 17:20:53 -0300 (CLST)

The discussion at
think a little.

Some of the comments there seemed a bit confused about the concepts
involved.  The biggest confusion seemed to be between implementation and
logical function.

While my experience is limited to smaller projects, I have found that
managing data efficiently generally depends very much on the particular
circumstances at the time.  The best solution will involve trade-offs that
exploit particular details of the problem at hand.

That doesn't mean that there are not certain useful technologies.  You are
almost always going to use a database, for example.  But those
technologies are going to be complex, and to use the efficiently you will
need to first understand what you are going to do in logical terms, and
then how to do that within a particular tool.

So in the discussion I was reading the "right" answer is not going to be
"use technology X".  It is going to be "do X; it can be achieved with
technology Y in this way or with technology Z in this way".

It is worth thinking about whether there is a problem at all.  One
wonder if the poster had ever considered how a database is implemented,
looked at the obvious correspondence between foreign keys and Java
references, or understood that a database can also cache data in memory.

It seems fairly clear that there is only a problem, then, if there is too
much data to fit into memory.  And even then, it is not clear that
explicit caching is going to improve performance - a database also capable
of caching recent results internally.

Yet it's also obvious explicit caching (eg memcache) works - everyone uses
it.  Why?

There seem to be two answers to that.  First, it's simpler (and more
familiar to many programmers) than configuring a complex database
solution.  I imagine that in many cases you could have careful data
partitioning and denormalisation (incidentally, as far as I can tell, this
is "sharding" - see
http://highscalability.com/unorthodox-approach-database-design-coming-shard),
or master/slave databases doing the same job (although this is going to
depend on the database - there is probably a financial factor here).

Second, explicit caching lets you trade consistency for efficiency.  To
some extent you can get this with databases by ignoring transactions, but
I suspect an explicit cache lets you take the idea much further.

Is that it?

I need to read more at http://highscalability.com/

Andrew