Alan Hargreaves' Blog

The ramblings of an Australian SaND TSC* Principal Field Technologist

Touring with Bryan Cantrill in Australia

I have had the most amazing week. I was privileged to be a part of Bryan’s trip to Australia. This involved such things as

  • SOSUG on Monday night
  • Tutorial and Plenary at AUUG
  • Customer briefings in Brisbane (which I missed), Sydney and Melbourne
  • Customer visits
  • Press activities

I’m not going to be able to do each of these justice, but I’ll give it a go.

SOSUG

James has done a review of this meeting. I will add that his ZFS presentation was extremely well done and received. While on the subject of ZFS, while it has not been integrated (yet) into the nevada source tree, it’s getting close to the time that you really should be watching the Open Solaris Source Tree and the Solaris Express Community Edition, as this is the first place that you are likely to see it (apart, of course, from the press releases and blogging that is sure to take place).

Bryan’s talk was more about what has been added recently to DTrace (and not yet been documented) and what is in his workspace (on his ferrari). The meeting also helped decide on the sematics of one of the features that he has not yet integrated. I’m really looking forward to be able to both print multiple aggregations in the same printa() and be able to pick what value or key to sort it on.

This also bears comment. When Bryan travels he has a notebook that he carries with him (a paper one in addition to the ferrari). If someone comes up with an idea he likes, he’ll write it down and then do something about implementing it. I witnessed this a number of times.

Tutorial and Plenary at AUUG

Unfortunately we did not get a lot of folks at the tutorial as we confirmed late. The good news about that is that the folks who did attend got a lot more interaction with Bryan throughout an excellant tutorial.

I suspect that Bryan’s plenary on the Wednesday morning will be an experience that people will talk about for some time to come. We knew that he had a lot to present and that 45 minutes was not a lot of time. I swear that he squeezed a two hour presentation into this 45 minutes; and throughout he had the audience spellbound. One of the metaphors that he uses when demonstrating DTrace (particularly watching all the system calls on a system run past) is that of winding down the window and sticking your head out at 120km/h. He certainly gave that impression in this presentation. One of the questions asked was about what kind of coffee he drank to get up to that speed.

Customer Briefings

There were customer briefings held (by invitation I think) in Brisbane, Sydney and Melbourne. As I was preparing for SOSUG on Monday, I didn’t get to Brisbane, but I believe that it was as well attended and received as the other two.

In these talks Bryan spoke more generally about the revolutionary technologies in Solaris 10, deep diving into DTrace and to a lesser extent Containers. One quote that sticks with me is:


Any of these six technologies on it’s own would be the most revolutionary thing that Sun has delivered in the last ten years, … and there are six of them!

Customer Visits

Obviously I can’t really say much about specifics of who we visited and what we talked about, but each of these went down really well. Bryan really “clicked” with some of the folks we visited and will be keeping in touch. The highlight of the customer visits was the last one, where we got to do some “hands on”. We managed to identify a number of issues with the application in question and made some suggestions on how to address them. I’ve also got to get some information back to these guys next week (as I didn’t have it on me at the time). From my point of view, it was great to actually be a part of that work with Bryan and we each managed to add something.

Press Activities

Again, I can’t really go into specifics here. Working with these folks was relatively new to me, so I tended more to watch and listen to what Bryan had to say, chiming in when appropriate. This was actually quite enjoyable and I hope to do a bit more of it in the future.

Developer Conference

This was held at the Maritime Museum in Darling Harbour in Sydney. I’ve never been here and after the brief look I got, I have to bring the kids down as they would have a ball (as would I).

Bryan modified what he had been speaking out to put a developer flavour to it. When he not only showed a stack trace involving a java stack interacting with the libraries and the kernel, but then proceeded to show a flow of control through this same stack (java->libraries->kernel and back), you could see how people were getting excited about this.

Then weave in the fact that once BrandX is out (I believe it won’t be long now for the pre-release) you will be able to use DTrace on linux code inside a BrandX container, the reasons for not having Solaris 10 at the very least on a development box start rapidly disappearing. Of course any improvements that you can make from andy kind of analysis can immediately be put into the code to run on what ever platform you need to make it run on. Add into the mix here that if you are running Solaris on an Opteron box, we run a 64 bit kernel as well as seamlessly running both 32 adn 64 bit applications. In the case of Brand-X while you may be running a 32 bit linux user space, you are actually mapping to a 64-bit kernel underneath, gaining you the speedups in data transfer and registers that you get by doing this. I have not seen figures yet, but logically one must expect that this may even make some applications actually run faster under Brand-X than they would natively under Linux (note when I capitalize “Linux” I am referring to the kernel. When I don’t I’m referring to an operating environment that contains the Linux kernel in the distribution).

In the food break at the end of the talk, discussion with the delegates was animated and excited. We have brought a whole new level of observability that no-one has ever had before.

Conclusions

This week has been absolutely amazing. In the times that he has not been presenting, simply interacting with Bryan has been fantastic. His knowledge of the history of software is nothing short of incredible. It’s also been great to catch on to some of the philosophy behind how and why he works the way that he does.

In addition to this we (Bryan, James Eagleton and myself) enjoyed a number of really very nice meals and dinner discussion1. On the flight back from Melbourne he asked me what I was doing I’d just finished a suduko in the in-flight magazine. We explained what it was and his considered response was that he thought he could write some code to solve that faster than we could solve one by hand. Thus started a challenge. By the following morning both of us almost had code ready. It’s going to be interesting to see how they compare.

It would be remiss of me if I were not to thank James Eagleton and everyone involved in getting Bryan out to Australia. I believe that it was certainly worthwhile and we must try to do this again.


1. “and like in good cocktail conversation, the answer to one question leads to the next”. I had to add this as I heard it so many times this week when Bryan was talking about using DTrace to look at a problem.

Technorati Tags: ,
,

Advertisements

Written by Alan

October 18, 2005 at 9:42 pm

Posted in OpenSolaris

%d bloggers like this: