Showing entries 21 to 24
« 10 Newer Entries
Displaying posts with tag: Ideas (reset)
An idea: create ha_trace tracing storage engine

Our exprience in solving query optimizer problems shows that a good chunk of optimization problems are incorrect choice of join order. The most frequent causes of the problems are

  1. Table engine returns index statistics or records-in-range numbers that are very far from reality;
  2. The WHERE clause has high-selectivity conditions for one table and low-selectivity conditions for the other. The optimizer is not aware of this and chooses a poor join order;
  3. Bugs or shortcomings in the optimizer or the table engine;

At the moment investigation of those kinds of problems is hard and time-consuming:

  • There is no easy way to find out what the storage engine has returned to the optimizer from records-in-range calls. One has to manually repeat the steps taken by equality propagation, constant table detection and other query rewrites, construct the table's condition and run the …
[Read more]
Nested-loops join speedup idea promoted to WL task

The idea how to speed up nested-loops join a little I've mentioned earlier has now been promoted into a worklog entry. It is now known as WL#3724 "Short-Cutting Join Execution: Speeding up star queries" and its text is publicly available at MySQLForge.

At the moment there is only a short description, but hopefully Martin Hansson (the assigned developer) will add more content there.

Backing up Data, Thoughts on an Alternate Ending

Let us recap for the audience on how to back up MySQL, or at least my favorite methods :)

1) Use Mysqldump
2) Shutdown the database or use an LVM snapshot method to back up the physical bits.
3) Use replication in combination with method 2.

And the winner for me goes to number three. So why is that?

Its cheap, and in a lot of cases I see, its very possible. This is the point where someone says, but I have 20 terabytes! I am sure you do, and in this case you can ignore the rest of what I am about to say.

One linux box with a bunch of disks with either one or multiple instances all acting as slaves is a cheap solution. From the point where you have a replication copy, you can make copies of the database off to tape or just create vectors (aka Monday, Tuesday, Wednesday... Friday one week ago) from the physical database. With this method you get live snapshots, but you end up …

[Read more]
Microsoft pays $115 million of Copyright Infringement

A Texas jury has awarded $133 million in damages to David Colvin, after finding Microsoft and Autodesk guilty of infringing upon Colvin’s two software patents for software antipiracy protection. Colvin’s company, z4 Technologies Inc., filed patents for ‘passwords and codes assigned to individual software copies to prevent unauthorized copies.’ Microsoft was ordered to pay $115 [...]

Showing entries 21 to 24
« 10 Newer Entries