Some hilarious stuff coming out of the morning session. They started with a video from OK GO called “This Too Shall Pass”. That caught the group’s attention really quickly. Then Souders jumped on the stage with Robbins to poke some humour at himself. He created a JibJab card with his picture super-imposed on a belly dancer.
The Internet is running out of IPv4 addresses. Only a small percentage of users have IPv6 and it’s preferred. A lot of people have IPv6, but it’s broken. Looks like the problem is that the web sites need to enable dual-stack (IPv4 and IPv6). Needed someone to take the first leap. Since no one was doing it, a group of sites decided to do it together. There was a World Day for IPv6.
On 8 June, 2011, top websites and Internet service providers around the world, including Google, Facebook, Yahoo!, Akamai and Limelight Networks joined together with more than 1000 other participating websites in World IPv6 Day for a successful global-scale trial of the new Internet Protocol, IPv6. By providing a coordinated 24-hour “test flight”, the event helped demonstrate that major websites around the world are well-positioned for the move to a global IPv6-enabled Internet, enabling its continued exponential growth.
Yahoo Home Page was one of the first sites to do it. In fact, all of the major players (Google, Facebook, Verizon, etc…). Yahoo even had a help site to help customers work through any issues.
Lesson #1: Monitoring is not the same in a dual-stack environment
Lesson #2: Don’t start something big and risky at a traffic inflection point
Lesson #3: Always have more than one way to look at data
Lesson #4: Always have more than one way to look at data (again)
Lesson #5: Practice Makes Perfect. For a major change, schedule one or more test runs.
Pace of innovation accelerated at Facebook even as the growth was happening (jump from 40 million to 500 million users). Facebook was an early adopter of MemcacheD. Then broke the architecture down to service clusters. Clusters for Web/Memcache, ads, multifeed, other small services (hadoop for data analysis, as well as messages platform). They have over 10,000 mySQL databases today. Size of a cluster is based on the number of switches in a 10G core.
Big Change: Efficiency
- Started by looking at Software: Had 3 parallel efforts (PhP, Quercus and HPHP) to see risk and reward. Dimmed the lights on PHP project. Dimmed the lights on Quercus as well. Maximized performance and chose HPHP as the solution for today. Software is 6X more efficient with equal traffic, 30% less CPU w/ 2X traffic and major cost benefits (10’s of millions of dollars).
- Continued with Servers
- Data Center
Power Grid –> Data Center –> Server –> Motherboard (Grid to Gate way of thinking)…Facebook was literally building their own servers and data centers. Went with a 1.5U server in order to get a bigger/better fan. Built even their own battery cabinet as a distributed UPS to be closer to the server. How wold they be racked? They chose a triplet rack as well. No screws, just pins to put in a rack. Overall they went from a 11%-17% power loss down to 2%.
Key Point: Small Teams Win Fast…Teams of 1 are too small.
Velocity culture depends on linking to the “business focus”. Case Studies from Bing/Google, Shopzilla and MSN/DoubleClick are the folks have shown some real business value. Haven’t shown a ton of business cases over time in which Operations = Business.
Amazon uses Utility Computing (aka…the cloud) as their way to make opps affect business. Typical week at Amazon, they waste about 39% of server capacity. During the month of November they actually waste about 76% of servers just to take advantage of about 24% of computing. Capacity Planning = Spending Money…Capacity Optimization = Saving Money.
November 10, 2010 was the day they turned off last physical web server. Everything was moved to EC2. Fleet of servers scales dynamically. Rock stars figure out how to make fleets small, not big. System at Amazon called Apollo. Allows developers to deploy their code to a fleet at any time. Big fan of how they call their servers as fleets. Each fleet has a purpose.
- Every 11.6 a new deployment is happening on Apallo
- 1,079 deployments in an hour
- 10,000 simultaneous hosts receive a deployment
- max of 30,000 hosts simultaneously receiving deployments
Main point is buy SSD disks. If not you are wasting your day. Think about how much money you save.
Cisco plans on putting OpenStack platform on Cisco UCS.
OpenStack is a global collaboration of developers and cloud computing technologists producing the ubiquitous open source cloud computing platform for public and private clouds. The project aims to deliver solutions for all types of clouds by being simple to implement, massively scalable, and feature rich. The technology consists of a series of interrelated projects delivering various components for a cloud infrastructure solution.
We had one of our first female presenters, Rachel Chalmers from the 451 Group.
Her main point is that the future of ops is devOps.
Take a look at this blog post by John. Basically covers his entire presentation.
- Use a tool called JSPerf for proving out issues.
The first demo is PageSpeed from Google. Full integration with Chrome, just like Firefox. Can identify Performance Patterns and Anti-Patters. Love that after 4 years of Velocity they are finally using the right terms of SPE to describe software problems. Pagespeed Online can analyze the performance on behalf of you the user. It even has a mobile interface, which is fantastic.
The more we keep seeing these tools and the more they get thrown in our face, it just makes me think how can we find a way to consistently use these tools? We bring them into our lab from time to time. What about making it a part of our practice.
The second demo is from dynaTrace. Earlier this year they announced Firefox 3.6/4 with AJAX edition 3. This year they are announcing a tool that allows cross-browser testing side by side. It also has web 2.0 awareness. This is a new report that shows Web 2.0 action. They added automation support for Selenium, ShowSlow, WebPageTest, etc…They even allow comparing with Alexa top 1000 sites via the Speed of the Web. It’s a new Beta app.
There are 15 YouTube video tutorials on how to use the tools.
Note to Self: The new version has support for both browsers that can be integrated with our server tool. This is a big deal for us.
The third presentation was from the Chrome development team. Chrome now has a task manager feature. If you right mouse click, you can see even more. Can even see the number of “goats teleported”. Can also see performance.timing (), performance.memory() with memory flag enabled, window.onerror(), console.profile(), console.profiles(), console.marktimeline() for adding notes. A lot of these tools can be used for Continuous Integration.
Great thing is you have the ability to use the command line feature to grab any of this feature. All of this information is available in an object (console.profiles()).
Now have a heap profiler. It can dig into memory usage of application. Can find DOM related memory leaks and highest memory consumers in the application. Even has remote debugging capability. THIS IS AWESOME!!!
The last presentation was from Sergey from Show Slow. All tests from WebPageTest going forward will go to ShowSlow.
Referenced an unfamiliar tool: DomMonster bookmarklet
Arvind Jain from Google
- 5 seconds to load a page (Yuck!)
- How do we make the web instant?
- In their example, it takes about 300ms to click on a page. This is key because it shows that the user wants to interact super quickly.
- Google Search: Instant pages announced 2 days ago.