Cloud Performance Training

Next week (Nov 18-22) I’m teaching a 5 day class on Cloud Performance, based on my book Systems Performance: Enterprise and the Cloud. I’ve taught this twice internally, and this will be the first class available publicly. I hope to teach it again, but one never knows, this may be the first and last chance to attend! Big thanks to Deirdré Straughan for making this all happen (including editing the book!).

As the website summarizes, the class is about:

This course is about cloud performance from a systems perspective, focusing on the systems that make up the environment. This covers both the challenges of traditional systems performance and also the performance of virtualization and multi-tenant environments. It is intended for system support staff, system administrators, developers, and anyone who would like to learn how systems work and perform, and the additional characteristics of the cloud. This is for both users and operators of the cloud.

A large part of Cloud Performance is understanding systems performance, which is common for all environments.

This class involves solving simulated issues, which my students have found is the best way to learn. This is where experience can be rapidly gained for putting methodologies and analysis tools into practice.

Posted on November 13, 2013 at 2:36 pm by Brendan Gregg · Permalink · Comments Closed
In: Training

Systems Performance: available now

My new book Systems Performance: Enterprise and the Cloud is shipping now from your favorite online book supplier (informIT, amazon). Thanks to those who have ordered a copy! I’ve already received feedback from people finding it useful, which is really gratifying. For an intro to the book, see my previous blog post about it.

I’ve also been using the book to help people more directly: last week I delivered a beta version of a Cloud Performance class, using the book as courseware. This class has additional labs, where the students solve simulated performance issues. It worked well, and I hope to be teaching more of these; the first will be held Nov 18-22 at Joyent’s San Francisco offices.

I taught similar performance classes during the book’s development, for students from a variety of backgrounds, especially system administrators, devops, and developers. The questions students ask, and where they get stuck during lab work, have been critical feedback during this book’s development, leading me to improve and develop new content. This has included documenting methodologies like the USE Method and the TSA Method, which are go-to methodologies for real world performance issues.

If you’d like to hear me talk about the book – and enjoy a free lunch on Dyn and Joyent! – sign up here to join us this Thursday, October 31st, at:

Le Meridien San Francisco, 333 Battery Street, San Francisco, CA
Thursday, Oct. 31 | 11:30 AM – 2:30 PM

My earlier post about the book links to Chapter 6 CPU as a sample chapter (PDF), which I’ll talk about during this event. See you there!

UPDATE 29-Oct-2013: This sold out immediately on amazon, who are waiting for another batch to be printed. informIT currently claims 24 hour shipping, so I assume they still have copies.

Posted on October 28, 2013 at 3:57 pm by Brendan Gregg · Permalink · One Comment
In: Performance · Tagged with: ,

Open Source Systems Performance

There is an important and untold story about open source and systems performance, which I’m uniquely positioned to tell. I presented it at OSCON earlier this year, which was the perfect venue. My talk, Open Source Systems Performance, was a play in three acts:

Linux appears in Act 2, but this story can only really be told by the Solaris operating system. Solaris was closed, then opened, and then closed again, providing acute examples of the value that an open source operating system provides for systems performance.

A video is on youtube:

The slides are available on slideshare and as a PDF:

Act 1 describes system performance without kernel source, a world that was documented, taught, and is still practiced today even when it doesn’t make sense.

In Act 2, DTrace, a pioneering Solaris technology, heralds a new era for systems performance, one where all performance issues can be debugged. DTrace stands for dynamic tracing: where “dynamic” refers to its ability to trace compiled code. However, dynamic tracing for the kernel – where many major wins are found – is only practical when the code can be read for reference.

Act 3 brings a twist of fate: Oracle closes the Solaris source, making it very hard for customers to use kernel dynamic tracing on their own. And, they begin porting DTrace to Linux, where the kernel is – unlike Solaris – open source. Will Oracle’s premier OS for DTrace become Oracle Linux, not Solaris?

Note that this story tells my personal experience and opinion; I am not privy to the official Solaris stance, nor work with Solaris at the moment. I work on illumos and Linux, and the illumos distributions, like SmartOS and OmniOS, enjoy a fully featured DTrace where kernel dynamic tracing is still practical.

There was good news with Solaris 11 and DTrace: more DTrace providers are available, including the long awaited tcp, udp, and ip providers. There was some other bad news too, with changes to the syscall provider. I personally think our industry continues to benefit from the existence of Solaris, and wish it the best. Hopefully, for Solaris, there will one day be an Act 4: the re-opening of Solaris!

Posted on October 20, 2013 at 12:17 pm by Brendan Gregg · Permalink · Comments Closed
In: Open Source · Tagged with: , , , , ,