Today, the Data Center Ambassador conference that I was attending after the CEC ended around noon. I had to do the last presentation, about my usual pet subject Oracle RAC on Sun Cluster. It went fairly well, and anyway half of the attendees already left in the morning to catch their planes so even if my talk  really was not interesting not much harm was done :o)

As my plane leaves tomorrow at 7:20am (uhhh), I had the afternoon free. My original plan was to first visit Stanford University and its supposedly very nice bookshop and then try to get to Santa Cruz using public transportation to have a nice dinner with view on the Pacific. According to the people at the hotel reception the latter plan was unrealistic, as it would presume a one hour bus drive to San Jose and then 2 (!!) hours to Santa Cruz. No Pacific for me this year :o( (Despite everyone's saying that it is cold and full of sharks I really want to swim in it at least once in my life!).

So Stanford Campus it was. There is a free shuttle driving from Palo Alto train station right into the heart of it. It is a nice area full of trees, happy chatty students and grey-haired professors. And what's more, I discovered an impressive collection of statues by Auguste Rodin. The sculpture called the 'Burghers of Calais' shows barefoot inhabitants of the city of Calais on their way to a certain death. They volunteered to offer themselves as hostages to the English king to spare their city and to accelerate the ending of the war. Just look at the expression on their faces. I think the existence of heroism is an equally difficult philosophical question than the existence of evil.

The Stanford bookstore was OK, lots of philosophy books but none about dogs sports (I thought that was a big thing in the US???). Only a very disappointing offer of comic books. I  liked the big bookstore in Denver much better also because it has a much cozier interior design. The cafe latte in the cafeteria really sucked. But books will be books and needless to say I spent a fair amount of money anyway.

 

 

 

So Monday morning I went to the presentation on Sun Cluster Futures and Solaris 10 Integration, by Paul Faramelli and Carlos Morillo. (the latter is also PTS Cluster Americas so glad to meet him as well :o). This talk started with a general overview about new cluster stuff and then Carlos gave details about how they are making Sun Cluster work with Solaris Containers and SMF.

The original plan was then to go to another SMF talk, but as I had to get to Menlo Park the afternoon and got a ride from Hartmut Streppel I thought it only fair to visit his presentation called 'Sun Cluster is Operating your Telephone Call'. This was really an interesting feel-good talk and showed how Sun Cluster can virtually solve all your availability demands.

Then there was the final session with Scott McNealy and then we headed off to Menlo Park and to sunshine! It was good to be out of the city! And good news is that finally during the evening someone asked my ID when I ordered my glass of Merlot! But then the waiter told me they were obliged to ask everyone who looked younger than 35. grrr....

Back after a few days interruption with the rest of my report on CEC.Because the ZFS talk was cancelled due to illness of the speaker, we decided to go to the talk about Trusted Solaris, as one of the speakers was Bart Blanquart who is from Belgium and we must encourage all Belgians :o). To be honest I am quite unfamiliar with the security world so this talk was a revelation. Trusted Solaris 8 is an different OS from 'normal' Solaris 8 and is compliant with the high demands markets such as the military pose on OSses. For example, it uses something called 'labelling' of files. For example, you log into trusted CDE and open 2 windows. Each window has a 'label'. It may very well be that you cannot just copy and paste from one window to another. Trusted Solaris can be deployed in the military, but is also suitable for commercial environments: it can help you to make sure confidential information remains confidential. You can run trusted Solaris on all your systems or as a gateway between networks to make sure that at least the labelling policy is enforced when one document goes from one network to another. Another interresting thing was that as of Solaris 10, Trusted Solaris is no longer a separate OS but the 'Trusted' option is included.

Then I went to a talk about Infiniband,  basic and in clustering. This talk was done by Patrick Lam from PTS Kernel and Matt Ostberg from PTS Cluster (hey at last I see some of my US colleagues in real life!). They covered the basic architecture of Infiniband and how it will be integrated in cluster. Seems like this will already be available in the next release of Sun Cluster 3.1, which is a good thing if you want to run Oracle RAC as the latter can take full advantage of the low latency protocol that is Infiniband.

Another interesting talk was about Solaris Update and Patch Strategy. This focussed on the Sun Update Connection, a network for developers giving them access to the latest patches and updates for Solaris 10.

Gifted speakers were Torrey McMahon and Scott Tracy with their talk about 'Arrays, Ports, Controllers and LUNs'. Not so much new stuff but a very good overview and definition of these 4 things and of what it means to be an assymmetric or symmetric array AND its repercussions on path failover. One thing to remember here: ONLY ONE CONTROLLER OWNS A LUN. No, Torrey I don't think anyone in the audience will forget!  As an extra, Scott talked about iSCSI which is available for Solaris as of NOW and which can be used with a combination of IPMP and STMS.

Last talk  for today was Mastering the tradeoffs in I/O performance. Clive King started this time with a good overview how to analyse I/O performance with a VERY funny demonstration  of active queue and wait queue as you see it in iostat (cf picture).  Then Bob Bartlett took over with some examples of issues with large systems. Extremely good talk.

 

 

 

This talk was performed by Rob Romack and Detlef Drewanz. In Solaris 10 SMF is introduced, replacing the old process of using RC scripts to startup applications. Why did they do this? I think we can correctly assume that a 'service' is more than just a process. In the current RC model we do not know the different relationships between processes, we only have very limited capabilities to restart died processes (through use of respawn in inittab) and we cannot define startup dependencies between different processes. With SMF this changes as it lets you control processes as services, and lets you enable dependencies, parallel service startup, restarts failed services etc... Don't worry if you still like to use the old RC methodology it is still there and you still can. This was an interesting talk and surely SMF is a major evolution and a better paradigm for dealing with system booting.  I am attending another talk on it on Monday morning and will tell you more after that.

This presentation was done by Owen Roberts from Solaris Sustaining. He explained the different stages in the Solaris boot process on X86: BIOS, mboot, pboot, bootblk. He also explained some of the enhancements that will be done to Solaris booting. For example, it will be made faster, there will be better support for loading other kinds of drivers, GRUB will be used as a boot loader and a boot archive will be used rather than loading the kernel directly from disk. The latter will allow you to use other types of filesystems, such as ZFS, for your root file system.

 

Dtrace rocks. And the guys who did this presentation , Clive King and Jonathan Haslam, rocked too and gave a very funny and interesting presentation.

For those of you who do not know Dtrace yet: it is a new tool (but that's too petty a word) that will solve everybody's Solaris problems and still everyones hunger for Solaris knowledge. It is a tool that can get you answers to questions like 'Which process is hammering at the ce: stack' 'What is the userstack of a process that is generating a specific syscall', 'Hey I want to know how this thing works but I don't have the source code'. Moreover, it does so without performance overhead as its probes are only active when you need them.

Dtrace consists of 3 layers: The providers, which are documented interfaces, such as the I/O provider, the CPU provider, syscall provide, consumers the ones who ask questions and format output (typically this is the Dtrace utility itself) and in between these two the Dtrace framework. We can easily write scripts in the common D language to get the information from the providers  (using the dtrace consumer).

Well there is lots and lots more to it that this but believe me this is really revolution. Check out http://docs.sun.com for Dtrace tutorials.  I will follow half a day course on it on Monday so I may tell you more then.

OK, my original plan was to go and see the SF 15K performance talk after the Solaris Cat talk. But last week I installed Solaris 10 on my laptop and I had some questions regarding this installation so I headed to the Installfest area where many laptop owners were eagerly installing Solaris 10.  I found the answers to my questions regarding the keyboard layout and the right X configuration of this laptop thanks to Joe Cicado and Volker Hergesell who patiently explained me some of the X86 specific installation features. Thanks guys!

 

 

The first breakout session I followed was about new features in Solaris Cat. As you may know, Solaris Cat is an excellent tool for crash dump analysis. You may be familiar with adb and mdb. These tools are useful if you want to debug a subsystem that you know well. If, however, you want to perform a general analysis of a crash dump, Solaris Cat is the easiest way.

John Harres, one of the authors of Solaris Cat, talked about the new features in 4.2: When opening a crash dump, Solaris Cat now performs more sanity checks (/etc/system, ndd settings, resource exhaustion, CPU changes, hung clocks, disabled pages ...). Also it performs more cross-checks wrt the callout tables, dispatch queues ... It has a feature to calculate which arguments are passed to the current funtion and uses the CTF (Compact Type Fromat). The new Scat contains SVM commands which allows you to look at the various types of information, including metasets (thanks!! this will be useful for me!!). Other new features are: support for CPU groups, the new 'sig' command to check signals sent to specific processes, constants in assembly is decoded automatically to symbols andsoon.

Then James McPherson talked about how they ported Scat to Solaris on X86 and X64 and gave some demos. This Scat version is in alpha now and will be available for Solaris 10 and beyond. Good!

If you want to try out this tool yourself you can download it from http://www.sun.com/downloads

 

 

 

CEC's first morning was filled with talks from Sun's VPs. They made Sun's vision crystal clear: Utility Computing is the future. If you are wondering what that is supposed to mean think about the following: electricity is provided by an electricity grid. You pay a monthly sum to the electricity company but then you can freely use the service they provide. Moreover, the appliances you want to attach to the grid are adapted to the grid and not the way around. So for the consumer this is easy to use. The technology on the side of the power plant is complex but there is no reason you and I would need to get involved with that unless we want a career in electrical engineering.

Computing in the future will be the same. If you still have the electricity metaphor in your mind you see what we are trying to describe here: everyone would think it overkill to have their own power generator in their house but in 99% of the cases we still have our own highly customized data centre. Wouldn't it be much nicer if we could just pay a monthly fee and use services of a big shared data centre?

First one to talk was Robert MacRitchie, VP of the Global Sales Organization talking about Sun becoming a much stronger solutions oriented organization. Interesting was his mentioning of future specialised Sun centres where customers could come and talk to specialists in different areas.

Then there was Marissa Peterson, VP of Sun Service  who talked about Preventive Services, a new service launched in June, which would in advance detect possible flaws in customer's configurations and data centres and thus make everyone happier.

Then there was Jonathan Schwartz. who did a short talk about the ideas of utility computing and the grid an,d OpenSource and spent a lot of his time answering questions. I must admit I quite like Jonathan, that's why I took a picture of him :o)

Robert Youngjohns gave an overview of what had already be accomplished in the area of utility computing using the Sun Grid Ecosystem and Opteron Servers.

 

 

 

Friday around noon I took the plane to San Francisco via Washington . I was actually looking forward to the flight: 8 hours without phone calls, mails, just reading a book and enoying a glass of wine :o) I managed indeed to finish Dan Brown's The Da Vinci Code and Anais Nin's A Spy in the House of Love. The glass of  wine was somewhat problematic as it seems that as of February United charges 4 Euro or 5 Dollars for alcoholic beverages even on transatlantic flights. Oh where did the charm of flying go? Ordinary people like me really enjoyed the glass of wine or the gin and tonic that you got served as aperitive and that made you feel for a short while like someone important. Drinking free Diet Coke isn't exactly the same ...

Anyhow, I had to wait for 3 hours, decided to have dinner, but finding some vegetarian food on TGI Friday's menu was a challenge.

The next part of the flight, from Washington to San Francisco was a pain but I spent most of the time sleeping on half of a sleeping pill. I arrived at 8:30 in SF so it was really too late to attend the opening ceremony or reception of the CEC and I went to bed.

Oh since we are talking wine here, since I am in the states this time nobody has asked my for my ID yet when ordering some !!! Last year they still did that ... It is probably the 30 that is coming up later this year that's starting to show :o( 

 

 

 

 

Hello,

I promised to add some blog entries about the CEC (Customer Engineering Conference). This is an anual conference for Sun Engineers. This year it takes place in San Francisco and I am leaving tomorrow!!

I took a look at the different talks and this is my provisional planning for the 3 days:

Saturday:  02/26
1:15-2:15PM New Solaris CAT features for Rapid Crash Analysis
--> Solaris CAT is a great tool and I would love to hear what they did to make it even better
2:25-3:25PM StarCat Performance: Issues and Answers
--> Just curious what this is about. And my ideas was to follow some performance talks anyway
3:40-4:40PM Dtrace – Vive la Revelucion!
--> Dtrace rocks! I am following a half day course on it on Monday but would like to hear what this talk has to say
4:50-5:50PM The Solaris X86 Boot Process
--> Just installed Solaris 10 x86 on my laptop AND it Sun Cluster is available on x86 since last year. 2 good reasons to follow this talk.
6:00-7:00PM Improving QFS and SAM-QFS
--> I think Shared QFS is a great filesystem in Oracle RAC environments. Curious to see what they did to make it even better.

Sunday:  02/27
1:15-2:15 PM Solaris 10 Service Mgmt Facitlity/No Longer RC
--> Want to know every gritty detail of Solaris 10. Following this talk is part of that plan.
2:25-3:25PM InfiniBand Basic Cluster and Troubleshooting
--> Hey the title contains the word 'cluster' so why shouldn't I go :o)))
3:40-4:40 PM ZFS: The Last Word in File Systems
--> Everyone seems to be over enthousiastic about ZFS so I would like to know what the hype is all about!
4:50-5:50PM Arrays, Ports, Controllers, Luns and Paths
--> I just love this title! Am curious what this talk will entail and I feel obliged to follow some storage stuff too.
6:00-7:00PM Mastering Tradeoffs in I/O Performance
--> Performance again

Monday:  02/28
8:00-9:00AM Sun Cluster on Solaris 10 and Future
--> Sun Cluster !!! And I promised the speaker I would attend their talk at 8:00 in the morning
9:10-10:10AM SMF for Administrators
--> More Solaris 10 and SMF for me...