New MariaDB/MySQL storage engines such as MyRocks and TokuDB have renewed interest in using engines other than InnoDB. This is great, but also presents new challenges. In this article, I will describe work that I am currently finishing, and which addresses one such challenge. For example, the left bar in the figure shows what happens… Continue reading Improving replication with multiple storage engines
Tag: performance
Understanding skew factors in Simplex/Improved Perlin Noise
[Here is a PDF version for readers whose browser does not understand MathML.] The Simplex Noise (or Improved Perlin Noise) algorithm uses a somewhat mysterious “skew factor” of 3–1 2 . I did not find any really satisfactory explanation for this factor in the descriptions of Simplex Noise that I read. But I managed to… Continue reading Understanding skew factors in Simplex/Improved Perlin Noise
Arduino to the max: 11x11x11 LED-cube
March 29 2014 is Arduino day, also in Labitat. This is a good opportunity to describe my LED-cube: This LED-cube pulls a number of tricks to get the most out of just a single normal Arduino Uno. A meager 16 MHz and 2048 bytes of RAM goes a long way with sufficient ingenuity and creativity.… Continue reading Arduino to the max: 11x11x11 LED-cube
No application changes needed: 10 times faster slave with MariaDB 10 parallel replication
Parallel replication is in MariaDB 10.0. I did some benchmarks on the code in 10.0.9. The results are quite good! Here is a graph that shows a 10-times improvement when enabling parallel replication: The graph shows the transaction per second as a function of number of slave worker threads, when the slave is executing events… Continue reading No application changes needed: 10 times faster slave with MariaDB 10 parallel replication
Using MASTER_GTID_WAIT() to avoid stale reads from slaves in replication
I have just implemented MASTER_GTID_WAIT() in MariaDB 10.0. This can be used to give a very elegant solution to the problem of stale reads in replication read-scaleout, without incuring the overheads normally associated with synchronous replication techniques. This idea came up recently in a discussion with Stephane Varoqui, and is similar to the concept of… Continue reading Using MASTER_GTID_WAIT() to avoid stale reads from slaves in replication
More on 40% better single-threaded performance in MariaDB
In my previous post I wrote about how I achived a >40% speedup on sysbench read-only using profile-guided optimisation (PGO). While this is a preliminary result, I though it was so interesting that it deserved early mention. The fact that any benchmark can be improved that much shows clearly that PGO is something worth looking… Continue reading More on 40% better single-threaded performance in MariaDB
40% better single-threaded performance in MariaDB
Continuing my investigation of single-threaded performance in the MariaDB server, I managed to increase throughput of single-threaded read-only sysbench by more than 40% so far: I use read-only sysbench 0.4.12 run like this: sysbench –num-threads=1 –test=oltp –oltp-test-mode=simple –oltp-read-only –oltp-skip-trx run And mysqld is run with minimal options: sql/mysqld –no-defaults –basedir=X –datadir=Y –innodb-buffer-pool-size=128M With modern high-performance… Continue reading 40% better single-threaded performance in MariaDB
MySQL/MariaDB single-threaded performance regressions, and a lesson in thread synchronisation primitives
I took a quick look at MariaDB 10.0 single-threaded performance (simple read-only sysbench). One thing immediately leaps to the eye, and I thought it worthy of mention. It contains an important lesson about the use of synchronisation primitives and in particular “atomic operations” in MariaDB (and MySQL). I am using the Linux perf tool on… Continue reading MySQL/MariaDB single-threaded performance regressions, and a lesson in thread synchronisation primitives
Even faster group commit!
I found time to continue my previous work on group commit for the binary log in MariaDB. In current code, a (group) commit to InnoDB does not less than three fsync() calls: Once during InnoDB prepare, to make sure we can recover the transaction in InnoDB if we crash after writing it to the binlog.… Continue reading Even faster group commit!
Benchmarking thread scheduling in group commit, part 2
I got access to our 12-core Intel server, so I was able to do some better benchmarks to test the different group commit thread scheduling methods: This graph shows queries-per-second as a function of number of parallel connections, for three test runs: Baseline MariaDB, without group commit. MariaDB with group commit, using the simple thread… Continue reading Benchmarking thread scheduling in group commit, part 2