Databases Reference
In-Depth Information
CHAPTER 3
Optimization Tips
When you're programming an application, you generally want the database to respond
instantly to anything you do. To maximize its ability to do this, it's important to know
what takes up time.
Tip #21: Minimize disk access
Accessing data from RAM is fast and accessing data from disk is slow. Therefore, most
optimization techniques are basically fancy ways of minimizing the amount of disk
accesses.
Fuzzy Math
Reading from disk is (about) a million times slower than reading from memory.
Most spinning disk drives can access data in, say, 10 milliseconds, whereas memory
returns data in 10 nanoseconds. (This depends a lot on what kind of hard drive you
have and what kind of RAM you have, but we'll do a very broad generalization that is
roughly accurate for most people.) This means that the ratio of disk time to RAM time
is 1 millisecond to 1 nanosecond. One millisecond is equal to one million nanoseconds,
so accessing disk takes (roughly) a million times longer than accessing RAM.
Thus, reading off of disk takes a
really
long time in computing terms.
On Linux, you can measure sequential disk access on your machine by
running
sudo hdparm -t /dev/hd
whatever
. This doesn't give you an exact
measure, as MongoDB will be doing non-sequential reads and writes,
but it's interesting to see what your machine can do.
So, what can be done about this? There are a couple “easy” solutions:
