Relational DB in-memory?

问题:

I have a simpleton question on Redis. If the key to it's performance is that it's in-memory, whey can't that be done on a regular SQL db?

回答1:

Any DBMS can be run "in memory". Consider the use of a ramdisk. However, most DBMSs (those with SQL) are not designed to run entirely in memory and put alot of effort to minimize disk IO and paging: a DBMS works very hard to keep the "relevant data" hot (in memory and in cache) -- IO is slow, slow slow.

This is because database data is often [and has historically been] significantly larger than main memory. That and main-memory is volatile :-) [ACID DBMSs do lots of works with write-ahead logging -- to a non-volatile store -- and other techniques to ensure data is never corrupted, even in case of a unexpected shutdown.]

Some databases, like SQLite use the same format for the disk and memory stores even though they explicitly support an in-memory store. Support for other [in-memory] back-ends and memory usage tuning vary by provider.

Happy coding.

回答2:

You may be interested in VoltDB

回答3:

The key is not only is it in memory, but it also has simpler operations than a SQL DB. Redis has simple operations such as GET, SET (and so on) using hash tables, and other optimized data structures.

SQL Databases generally take longer to compute, however they are a ton more flexible and in most cases more powerful (in terms of what type of queries). You most certainly cannot run JOIN queries in Redis, for example

回答4:

You may be interested in TimesTen (which is now Oracle).

In 11g its SQL has improved significantly, though still is not as powerful as that of Oracle.

回答5:

You can do that natively with some SQL database management systems. But there are risks.

You stand to lose data if the server fails, for example. I don't think you can get ACID compliant transactions; any log file would have to be written to disk to survive a server failure. (I imagine it's possible for an in-memory SQL dbms to still write log files to disk, but I've never run across that myself. Not that I've looked much.)

回答6:

On DB's in RAM: Traditional databases will eventually wind up in RAM:

Traditional database data — records of human transactional activity [...] — will not grow as fast as Moore’s Law makes computer chips cheaper.

And that point has a straightforward corollary, namely:

It will become ever more affordable to put traditional database data entirely into RAM.