厚积薄发: accumulating steadily

- Impressions "Impressive" I have to say. The OSCON 2007 conference has 2500+ attendees. The small town of Portland is filled by the attendees of the conference to the very tiniest corner. Whenever you see a people who is IT-like, there's a 90% possibility that he/she's here for the conference. Different from GUADEC (compared with my idea of GUDEC of course, since I've never been to a GUADEC), OSCON is not so technically focused. A lot of the people here are not developers. And the sessions are not all about technical stuffs. Approximately 40% of the sessions are about people, legal or future of opensource. And of course, the technical sessions are not desktop focused either. The sponsors of this conferences are famous companies, either involved in Opensource or not, like Intel, Microsoft, IBM, HP, Google etc. Almost any famous company that you can named in the IT industry can be seen here.Of course the exhibitors are even more. The exhibition hall is HUGE, and Intel has the biggest booth, followed by Sun. However, there seems to be few things in the Intel booth compared to its size, while Sun's has made used of it's booth to the very most. The thing that I have a problem with it of course, the food: little of them are cooked, especial veges. - Sessions -- More than licenses: the legal policy of the free world in the age of web 2.0. Speaker: Eben Moglen, Director, Software Freedom Law Center I took this session partly because of the heated discussion of GPL 3 that are being carried on recently. But it turned out to be mainly an inspiring speech. Eben talked about what we need to think about licenses in the Open-source world, stating that licenses are not used to make sure that software is free or safe but to keep the community strong. It's role is just like a constitution of the community to regulate the activities. Opensource features the 21st century, while previous versions of GPL which are developed in the 1980's is not enough for the new century. There's need that community interest must be served, this is why GPL 3 is proposed. It's purpose is to structure the conversation between different stakeholders to set the principle. Licenses is not a tool to fight against the competitors with. Eben also talked about the vulnerabilities of the license. However, nothing is meant to be perfect. Also, the patent is still threaten in the opensource world whose system still needs discussion. -- Project Indiana: the road to Solaris .next Ian Murdock, Chief OS Platform Strategist, Sun Microsystems

Ian Murdock

For products and services, the rooms are relatively smaller than those for technical sessions, however, the room was full of people when Ian started to give his talk. The presentation started with a brief of Ian himself, having been involved in Open-source for 14 years, and ever worked as CTO of linux foundation, the founder of Debian. People kept asking Ian why he choose to work for Sun and his answer is that, 1. when he first get to start working on open operating systems, UNIX, especially Solaris was what he was familiar with not "Linux". However, now, Linux is known by more. 2. most of the deployment decision now are bottom up instead of top-down, which lead to heavy weight processes, even in the enterprises 3. the dividing line between OS and apps is blurred, Operating system now seems to be a collection of software. in 2005 Solaris was opened. Ian was familiar with Solaris and is fascinated by some of the innovations of Solaris like DTrace. However, the opensolaris model is not so good, it provides code base, and a long list of things to do in order to make it build. Ian has the idea that people are familiar to Linux and the concept of Linux nowadays is more about the desktop/appearance but not the kernel. Sun's goal is to make Solaris more Linux like in the appearance while keeping Solaris' strengths like being innovative in the kernel part and backward compatibility and stability. The strengths of Linux are that it's free, it has community and it has a distro model. The most remarkable innovation in Linux is the "distro" model. Linux releases binary distributions from time to time for users. Project Indiana is trying to build a distribution model for Solaris. Different from SXDE, which we release on a 3 months base without giving support for customers, Indiana will be release on a 6 month base and feedbacks from customers will be considered and supported will be given to make Solaris a better platform. None registration is required for involving in Indiana which is different from SXDE as well. Sun is expected to be paid when value can be seen. The goal of Indiana is to make Solaris technologies more broadly and quickly available for the users and to expose more of the kernel . Test release will be available in fall of 2007 and the first release will be in spring of 2008. --Mozilla Firefox and the Internet as an Open-Platform Speaker: Mitchell Baker, the Mozilla Cooperation The presentation focused on how the success of Firefox highlights opportunities for all to participate in creating a healthy internet. She defined a healthy web as a web where the users have a choice, and can add things that you want, a.k.a. customizable. The opensource community enables the establishment of such health web. What matters is a to establish a model of mess participation. The most important thing that the participant can obtain from this model is a sense of achievement (she gave an example of a silent, introverted engineer who become talkative and confident through such participation.). Here's the what need to compose a health web according to her: Healthy web = Choices + security (in basic sense) + interoperability + decentralization + transparency + accountability + trust + multiple source of innovation. --escaping image storage hell: Xen Image manager (XIM) Speaker: Jonathan Oxer, internet vision technologies In the world of virtualization, the problems of machine configuration has been handled properly, yet one of the problems with Xen administration is how to manage the storage of a great number of virtual machine images. Jonathan proposed a Xen image manager(still under development) for handling this problem. The Xen image manager can store VM image on Xen nodes using LVM, survey resources, propagate between using LVM snapshots and encapsulate Xen function for convenience etc. Currently, the image manager can build new VM image from a master VM, demolishing existing VMs, propagate a VM snapshot to other nodes, rename duplicate image locally etc. Features to be implemented include relocation of images, auto migration and resource utilization etc. Slides available at joh.oxer.com.au/talks --Why Observability matters- how DTrace helped Twitter speakers: Leventhal, Gregg, Sun Microsystems Leventhal and Gregg gave a real life example of applying Dtrace. Twitter is a free social networking and micro-blogging service that allows users to send "updates" via SMS, instant messaging, email, the Twitter website, or an application such as Twitterrific. It is horizontally scaled, and has a centralized MySQL backend. It is successful, but it doesn't scale with its success and high latencies could make it unusable. Possible suspects of causes can be in operating system, ruby, mySQL, apache or the application itself. Demos were shown to explain how Dtrace helped to identify the cause of the problem and help to improve the performance significantly. The practice lead to joyent's interest in DTrace, who now start to provide DTrace-enabled Ruby. -- People Hacks Speaker: Adam Keys This is where I saw the hugest number of audience in all sessions. None of the seats were empty and some were sitting with their laptop on the carpet. Modern developers are faced with more social problems then technical. Adam gave some tips on how to do people hacks while being an engineer, including 1. influence people's emotions by making them like you, always smiling not frowning, more praisal than criticism etc. 2. Admitted humbly if you have make a mistake 3. being patient 4. make change little by little --Next generation Version control systems Speaker: Sam Vilain, Catalyst IT (NZ) Ltd. Sam summarized the evolvementof version control systems including revision model (simply copying, locking while editing, patch-based, unify diffs etc) and the tools (RNA, RCS, subversion, arch, bazaar, darcs etc). He gave ideas of next generation version control systems and compared three next-generation-VCSs: git, bazaar and mercury(hg). Pros and cons of each of them are summarized, speed, whether lightweight branches are enabled, portability, spaces required were also compared. --A11Y for Web 2.0 Speaker: Eric David, Right Media I became one of the audience of the session because the first part of the title: A11Y. Although I don't care about the detail implementations of A11Y with web 2.0, this session was a structured introduction to A11Y concerns for me. Usability factors of accessibility 1. avoid using colors to convey information 2. cognitive friendly interfaces 3. systematically correct make up 4. tables for tabular data only 5. front loading (important for screen reader) 6. avoid time limit on actions effects of ability impairment 1. loss of limb usage 2. photosensitive epilepsy 3. color blindness 4. poor vision 5. blind and deaf 6. cognitive barrier assistive tech include 1. enlarge text 2. head wand 3. special keyboards 4. screen reader 5. virtouch mouse for the blind He also talked about WCAG-WAI and section 508 guidlines. Wondering whether flash is also accessible (which it is supposed to) on Solaris currently? Slide available at ericdavid.cc/oscon2007 --Why user spaces sucks Speaker: Dave Johns, Red Hat Dave Johns talked about common issues with application running in user space that negatively impact system performance and offers tips and techniques for writing better applications He gave examples with astonishing statistics. Many of the applications are those we are familiar with. A lot of the applications use polls, some even poll for no obvious purpose. Some examples XOrg: scans through /proc/bus/pci/ (Ubuntu as an example OS)for PCI devices in order according to randome bus numbers Nautilus: lots of time is spent stating template and applications files even nothing is changed X font server, rebuild font cache from time to time even there's no changes mixer_applet wakes up every 100 ms to check changes of the volume, while gstreamer has a volume change notification signal. pcscd: scans all USB devices once per second to see if new devices are added, even without changes Xend has to finish 16 transactions every time you ask for information about a daemon, which requires huge disk I/O overhead. ... Tools can be strace, strace-account, ltrace, systemtap script, valgrind (these are suggested for Linux. Of course, for Solaris, we have a more powerful tool than these all: DTrace.) --What's in it for me? How your company benefit from opensourcing code. Brian Fitzpatrick, Google Inc. Two Google engineers talked about the ways for a company to opensource an internal project. The ways include 1. fake opensource 2. throw the code on the wall: publish a source tarball and do nothing 3. develop internally 4. open Monachy 5. Concensus based development (the best practice according to the developer but requires patience) They argued that Google is opensourcing their code in the 5th way which all the commiters can involve in decision making. However, neither Glynn nor I think Google is doing the thing that they talked about. -Other might-be-interesting topic which I didn't attend include 1. network monitoring with Nagios: Talor Dondic talked about the popular opensource host, service, and network monitoring program that streamlies network-monitor tasks while reducing the cost of operation. URL: www.nagios.org 2. Use puppet to handle your network: Luke Kanies presented a system admin automation tool written in Ruby, capable of expressing and managing the configuration of all the computers on your network. 3. Google code: a retrospective by Greg Stein talked about the google project hosting services: Google code website, including features, uniqueness and how to use. 4. 3 sessions are about Ajax.