Although I will go into the details of how the MySQL/Innodb/ZFS read-only
performance was improved almost 5x using SSDs for L2ARC, this blog is primarily
focused on understanding why I got that improvement for the configuration that I was using. I'll also discuss why the tuning tip I mentioned in my previous blog entry, of matching the ZFS block size to the database record size, is not
such a good idea for this particular test.[Read More]
All generalizations are incorrect Charles Suresh's blog
- All
- Performance
- Personal
- Sun
Tuesday Aug 18, 2009
Wednesday Jul 22, 2009
In some cases, system performance can be improved simply by reducing wasteful work.
So in this blog, I talk about tools and techniques you can use to find if your system
is doing more I/O than necessary and what you can do about it.[Read More]
Thursday Apr 09, 2009
A map is a lifesaver when embarking down unfamiliar territory. When I recently
started working on ZFS performance (in the context of MySQL usage), I needed a
code flow chart to know what parts I needed to focus on within ZFS. Take
a look at the map of ZFS I generated to help me along the way.[Read More]
Tuesday May 06, 2008
uperf is a network performance evaluation tool which is being open sourced
by Sun today. Take a look at how I used it to compare network "ping-pong" traffic
rate scaling on Solaris vs Linux.[Read More]
Wednesday Mar 12, 2008
Multi-core/Multi-thread CPUs can exacerbate scaling problems due to locking. A "mux"
that pipelines work from multiple producers to a single consumer that avoids high
lock hold times is discussed in this blog. Some of the flow control issues that needed
to be addressed to deal with fast producers and slow consumers are also discussed
along with the politics of source code changes to critical system software.[Read More]