Paul Roberts' Musings

Thursday December 22, 2005

London OpenSolaris User Meeting Report

On Monday (19th December) I attended the latest meeting of the London OpenSolaris User Group. This was a very interesting event. It was difficult for me to judge the ratio between Sun and non-Sun attendees (the majority of faces were unfamiliar to me) but I would like to think that it was a popular event beyond the Sun staff.

The event was opened by a detailed talk about DTrace from Jon Haslam. This talk turned out to be a highlight of the evening, as rather than being just another demo of the technology, it was instead an examination of the design and implementation of DTrace. This included an analysis of how your D code is converted into the DTrace virtual machine opcodes and how these objects are then injected into the kernel so that they are executed when a newly activated probe is 'hit'. In all it made for a very helpful precursor to delving into the sourcecode.

We were also joined by a welcome guest from overseas, Adam Leventhal. He is a member of the Sun engineer elite, probably most famous for his his membership in the DTrace team. In the latter capacity he contributed to the meeting by assisting Jon in his description of DTrace. However, he also presented us with a new project, BrandZ, which I had heard of in the past, but haven't had an update on in a while.

This is an interesting project which aims to allow binaries from non-native OSs (e.g. Linux) to run unmodified in a Solaris zone (on a matching processor architecture) by 'booting' a Branded (non-Solaris) Zone and emulating the guest OS' kernel API there (for the most part this means the kernel's system calls). Unfortunately the talk was only a quick (5-minute) demo and proof of concept, but it was very interesting to see that real progress is being made on this project, which is now available for download via OpenSolaris.

This was then followed by a general overview of ZFS, and why its great, as presented by fellow GMP-er, Tim Graves. This presentation is great, and I have seen it presented while I was in the states by the ZFS team themselves, but unfortunately due to a lack of time Tim had to run through it a bit quicker than it deserved. However, ZFS remains a fantastic technology which I love to play with. Its unfortunate that as a technology it can sometimes be a bit of an underwhelming demo because it is designed to make things easy, and things always seem easy in a well prepared demo. The only real way to get a feel for how revolutionary ZFS is is to download it and give it a go, something which I highly recommend doing.

Next up was another quick presentation about a new Solaris related technology, this time the port of Solaris to Xen (a free open source virtual machine type program, allowing you to run multiple OSs at once). The good news is that this project is making progress and has a currently working demoable status.

Next we had a talk on an important topic which is currently starting to get the attention it deserves in Sun: performance. Phil Harman, who presented this talk, is one of the leaders of the movement to put performance on the top of the agenda within Sun (he coined the oft-quoted phrase 'If its faster in Linux, its a Solaris bug'). One of the interesting things that he discussed was the work his team was engaged in in changing the default attitude of Sun, and the best way to convince engineers of something is to prove it in practice, so the team developed a portable benchmarking tool libmicro to produce real provable statistics. This tool is designed to make it easy to create benchmarks that test your favourite portable APIs on different platforms and it has been successfully run on most obvious modern Unixes (Solaris, Linux, Windows Services for Unix, MacOSX etc). This application is now available online and Phil encouraged everyone to download it and create thier own benchmarks to see how Solaris compares in thier favourite APIs. And if Solaris doesn't compare favourably, we want to know about it!

Finally, Darren Moffet closed the show with a quick presentation of Sun's efforts to integrate encryption into our filesystems. He gave a summary of the field and Sun's design decisions in implementing this forthcoming project.

All in all it made for a very interesting evening.

( Dec 22 2005, 04:59:12 PM GMT ) Permalink