Detailed parsing of the Dynamo storage engine

First, suppose a scenario where your site wants to store the IP that the user logs on to. How to solve this problem? The traditional method is to use the database. The database provides the convenient operation interface, the complex query ability as well as the thing guarantee.

OK, now suppose everyone loves your site, and more people are visiting. A database has not been processed. So you installed 3 database hosts, divided the user into three categories (men, women, it people; there is always some way to divide the user into roughly the same number of parts).

Each time you visit, you first look at which category the user belongs to, and then directly access the database that stores that type of user data. So the processing capacity increased three times times. This time you have implemented a distributed storage engine, Dynamo is a similar thing. It's just a little bit more reliable, usability and so on. Let's look at what is inconvenient about that simple distributed storage system, and how the Dynamo is resolved.

Possible problems of cloud storage in simple distributed system

Let's start by enumerating the possible problems of a simple distributed system:

1 It is difficult to expand: if the current business development quickly, 3 hosts can not support, need to add to 5 host, then how to deal with it? First, you change the classification method, divide the user into 5 categories, and then migrate the existing data again. You have to put a note on the site, "system Maintenance", and then began a great migration project, until finally the migration completed, found that in fact, 3 units also need not, users have gone.

2 data reliability can not be guaranteed: one day, found that a database server's hard drive is broken, this trouble comes, the original site does not make money, no high-end machines, only a regular incremental backup. After a day of complex recovery work, you have to say to some users, trouble you to do things again ah.

3 single Point problem: responsible for the user classification, and then decide which data server to use the host is the lifeblood of the site Ah, it if the downtime, all the data can not be accessed, it if full load, increase the data server will not be helpful to the overall performance. I seem to see a PC server full of Exorcism spells.

These problems, seemingly small, it is not easy to solve it. Especially when you think about your site, you might someday have as many users as Google (probably because you're a genius or Google is going to fail). Now let's see how the Dynnamo is solved.