Too many MySQL tables?

[Gabby]

I was reading about databases with a ton of tables on Stack Overflow and I found that some people feel there is a performance impact when there is too many MySQL tables on a server. But in the same breath, it caused a huge debate with people who disagree. I'm just wondering if there's a proven answer? Maybe with trials or something?

I know database normalization may result in many tables, which has me so confused. Of course, I know, normalizing all of your data helps the health of your database in the long run, but it also creates more MySQL tables, which some say cause a performance impact, so does anyone have a straight answer to this? XD

[Chevy]

I would say make as few tables as you need, but with a pet site, it will be quite a few. Once you start putting rows in the wrong table (places that don't make sense) or make a table for site wide global fields (which I think is totally unnecessary, some might differ in that opinion) it gets really confusing.

[judda]

The alternative is to have VARCHAR (text-based) lookups on your main tables (instead of INT lookups). The VARCHAR lookups will be way more expensive on a big fact table than any impact of too many tables in a MySQL database. People choose very interesting things to have holy wars over. This being one of them. I'm guessing that the people who are throwing out the facts about how too many tables will cause a performance hit are the same people who say all normalization is bad. Probably sponsored by the "anti-join order".

The fact is, if you are storing the VARCHAR version of the text on each table, you will be duplicating a lot of information (the string value) which costs more (disk space wise, and index wise) than storing an integer value. Example:
VARCHAR(50) => if full will take 51 bytes worth of disk/index space (per row) - actual size of VARCHAR fields is the length of content in the row + 1 byte (i.e. 4 characters long = 5 bytes)
CHAR(50) => will take 50 bytes worth of disk/index space (per row)
INT => 4 bytes worth of disk/index space (per row)

Yes, you acquire a little more overhead from having to do a table lookup against your dimension table, but at the end of the day, you are saving a lot of disk/index space using an INT. Smaller column width means that more of the index can fit on any given page. More indexes per page means quicker querying times.

Also, string comparisons actually take a fair bit more time to actually execute than the INT comparisons (it needs to pull all of the bytes into memory and scan one-by-one).

In short, normalize your tables it'll make your life easier, cleaner, and keep your database as small as it can be.

~judda

[Hituro]

When a database is opened there is overhead that relates to the number of tables, and also the size of the tables, especially with InnoDB, where you are storing all of the tables in one file. But I think that's more to do with total DB size than number of tables. I don't think there is a big downside to lots of small tables, and I think like @judda says, it's better to have more small specialised tables than big generic ones with string keys.

For reference Grophland currently has 95 tables

[Mio]

People choose very interesting things to have holy wars over. This being one of them. I'm guessing that the people who are throwing out the facts about how too many tables will cause a performance hit are the same people who say all normalization is bad. Probably sponsored by the "anti-join order".

LOL anti-join order.

In short, normalize your tables it'll make your life easier, cleaner, and keep your database as small as it can be.

Seconded. The cleanliness has it for me.

[Myztacia]

I agree with @judda on this as well. However, I still say do this within reason. :]

[Gabby]

I agree with @judda on this as well. However, I still say do this within reason. :]

Oh of course, I do everything with moderation (unless you're talking about cheese... or ice cream.. aaaah XD)

[BigThinker]

I have about 20 tables and each table is specific per area or feature. The only table thats big, width wise is the pet table since there's a lot of info per pet, but some of the game tables are ever growing in length but have few columns.

[Gabby]

I have about 20 tables and each table is specific per area or feature. The only table thats big, width wise is the pet table since there's a lot of info per pet, but some of the game tables are ever growing in length but have few columns.

I would personally have two tables. One for all the species information and one for all the necessary data that needs to be saved per pet. I don't know if you combined the two but I'd be careful if you did. A simple INNER JOIN query would be able to join the two tables together if you needed any information from both. A lot of columns could impact performance, it's harder to manage, and it isn't really good normalization practice.

@judda correct me if I'm wrong. ^.^

[Myztacia]

@Gabby That's me with Reese's, ice cream, and avocados! :]
@BigThinker The number of columns isn't necessarily a problem. If you're creating a table just because you want to remove some of the columns from another, that probably isn't the way to go. (I'm not saying you are! I just felt it worth mentioning for future readers of the post.) Keeping them grouped in ways that make sense, such as grouping by area, is definitely a great start. As a programmer becomes more familiar with databases and grouping data, it will be easier to see where one can save on performance by breaking up tables.

Along these lines, it's more important to get the correct datatype rather than worry about how many fields there are. If you have 10 very useful fields but they're all in the wrong datatype, you'll suffer performance later on. In particular, as @judda said, indexes should be in the most efficient way they can be.

One final point I would like to make is that if it is becoming hard to understand where to find data in your database, you are probably breaking the tables down too much. It will be harder for yourself and future programmers to maintain the code.