May 08 2006

Bugs in the Kernel

So Andrew mentionned he thinks kernels are getting buggier last week at LinuxTag, yet he has no realdata except his inbox to confirm that.

Dave thinks

I think a more likely explanation is just the fact that we're getting better at finding bugs, and getting better (and more) reports from end-users.

Kernels aren't getting buggier.. we just find more bugs.. How many of these bugs are life threatening. ?
Some other questions ... What is growing faster .. the number of contributors, the size of the code , or the number of bugs.
As we don't have exact figures yet... how can we be sure that this is a bad trend ?
It would indeed be bad if the number of bugs is growing at a faster rate than the code base. But given the growth of the code base it might as well be otherwise. We just don't know. It is perfectly possible that eventhough Andrew is getting more bugs reported on a day by day base , the overall stability of the Linux kernel is increasing.

Afterall .. there's Lies, damn Lies and Statistics !

May 07 2006

LinuxTag 2006 Report Continued

The after party... somehow we ended up sitting at the same terrace having dinner as some of the other Linuxtag people ..
The result being that our tables eventually merged into a fun talk with Philip Copeland (Bryce) the ocfs2 guy ..and some late night open air irc session with seklos
who apologised for having to send someone else to actually give a talk. Philip tried his best to give a good talk.. but as posted earlier.. it is difficult to do so when it's not your talk.
I hope to see him give one of his own talks one day .. promises to be really good.

On saturday morning Ruediger was so kind to fill in the open spot we had which saved the audience from a 4th rerun of my Xen Deployments talk , I managed to miss half of his talk however.. kind of annoying...

Then time to introduce Werner.. who really is giving a talk I have been proposing to different conferences lately.. except that I was focussing on GFS and Xen where he takes a more general approach. Should be really interresting... As Werner mentions a problem with virtualisation .. when you consolidate.. and one machine fails it's actually more machines that fail.. therefore the need for ha becomes important. which is also one of the reasons why I am looking at the topic.. that and many-2-one ha setups which I think can save us a lot of idle hardware.

Werner mentions 5 possible setup .. he wonders why people would want to failover from virtual to physical, the only reason I can come up with for now is someone who has a lot of old machines and buys one new bigger machine where he consolidates his apps. Some different virtual machines might have a physical failover on an older left over machine. Looking at it from a budget point of view it makes sense.

Werner is doing his storage mirroring on his physical host , I've been looking at mirroring storage between virtual machines or between physical and virtual machines (certainly in a many 2 one setup) , this definitely needs more thinking, and more work to be done on testing integration between Xen and different clusterfilesystems ocfs2, gfs etc.

Time for me to check out of the hotel and do some small shopping (food) before heading back to the Xen talk ... not by Ian this time... but by Steve Hand, a nice change to see this presentation given by someone else who's equally skilled and involved into the stuff. I sneaked out for 5 minutes to see how crowded the other talks are... Mark's talk was crowded... you couldn't get into the room.. As the Xen talk is also rather crowded one can only assume that this last day at Linuxtag is really crowded ..

After the Xen talk I took some time to talk to Aaron about our ideas on openQRM and SystemImager, talked some more with Werner Fisher about Virtual Clusters tried to get back into the talks but sat down with Steve Hand for some Xen chat instead (I really got to look into buying a new Laptop :))
But luckily I got into last talk by Klaus Rabbertz on Software Deployment in Grids. Problem sounds familiar... solution doesn't.. it's really grid specific. Juliux really needs to take off soon :) Some interresting tought on using different RPM databases on one machine however. I need to buy some time somehwere to have a look at how stuff they have been building can be used in a broader area.

LinuxTag 2006 is over now.. I`m satisfied with it.. met some interresting people and learned some new stuff .. after all that's why you go to conferences isn't it.
I've seen LinuxTag from different viewpoints already.. as a regular visitor, as a speaker and this year as a programcomitee member. Depending on location and time I assume this 6th linuxtag I visited certainly won't be the last.

May 05 2006

LinuxTag 2006 20060505 Kernel Questioning

This is the time of the day I`m happy I never have really committed code to the Linux Kernel... I've only removed code from openMosix and the fact that my nickname is still somewhere in the tree really doesn't count.. (hmm.. yep still there.. just verified.. weird..)

Nils called everyone who raised his hand and they are all on stage now :)

I didn't catch all the names but AGK, Heinz, Jan Blunck, Jorg Engel, Christophe "I got a new Haircut so I can attack unsuspecting speakers with questions again" Andrew Morton, Marcel Holzmann

The award for the most funny introduction goes to
"I`m Andrew Morton .. I maintain the 3Com network driver "

May 05 2006

LinuxTag 2006 20060505 Kernel Questioning

This is the time of the day I`m happy I never have really committed code to the Linux Kernel... I've only removed code from openMosix and the fact that my nickname is still somewhere in the tree really doesn't count.. (hmm.. yep still there.. just verified.. weird..)

Nils called everyone who raised his hand and they are all on stage now :)

I didn't catch all the names but AGK, Heinz, Jan Blunck, Jorg Engel, Christophe "I got a new Haircut so I can attack unsuspecting speakers with questions again" Andrew Morton, Marcel Holzmann

The award for the most funny introduction goes to
"I`m Andrew Morton .. I maintain the 3Com network driver "

May 05 2006

LinuxTag 2006 20060505

So after breakfast it's time to figure out where the conference actually is..
Whee.. that's far .. and you have to cross a street .. dangerous .. at least compared to stepping out of the Dorint's doors in Karlsruhe :)
It's right across the street , but the conference entrance is around the corner..

Well hidden on the schedule is a talk about OpenVZ it's in one of the Community workshop rooms. It means I'll have to miss AGK's talk about LVM fun... tough choice ...

They go over a short overview of the different Virtualisation types etc interresting to see that they claim as the advantages compared to their collegues ..

One of the claims against Qemu and VMWare is the one by one management etc.Hello .Large Scale management anyone ?
As cons for Para Virtualisation the "Needs modified guest OS" is acceptable but "Static Resource Allocation , bad scalability and bad manageability.. they don't consider it different from vmware. and also one by one management..
According to them "ballooning" isn't a kind of real solution but just a hack.. works for me ;)
And the fact that I can add disks on the fly when I need to is also not really static for my needs.

Also I personally really prefer to have identical tools for my physical and virtual the fact that a collegue just doesn't need to know about wether this host is physical or virtual and use the same tools he is used to makes it a lot easier.

3rd kind is OS level virtualisation .. BSD Jails, VServer .. Solaris Zones , OpenVZ. no multiple kernels.. single kernel ...
Hmm.. this won't fix eg netfilter conflict issues .. Xen does .. I want to be able to runn different kernels in different environments. Say old and antique 2.4 kernels on a machine that is phyiscally only supported on a 2.6 kernel. because we have to keep an old legacy application around.. (yeah those already exist on Linux .. strange huh .. )

Nice viewpoint on the evolution from Multitask to MultiUser to multiple Execution envirnments ...
Open VZ has also checkpointing and Migration support altough according to their blog it is really recent.

The real fun part is how competing projects often don't know the correct facts aoubt their peers, "If the sshd on your xen box dies you have to reboot it because you can't access it ?" Huh .. strange .. I must have misheard but it certainly ain't true .... I definitely don't claim to know everytyhing about other projects.. but I won't make claims I do .. and I know it's difficult to catch up and keep track of what's happening as we are only humans..

Got in late into the Git/ Cogito talk , Gitk really really looks nice
It's really really time I set up that openMosix Git tree sometime...

I ran into Matt after that talk.. he just arrived ..with a small hmm.. not that small surprise .. Matt already mailed about doing a Aaron visits Europe and Matt the USA.. but I really didn't expect to meet him here... really great to finally meet him.. We had lunch downstairs (much better Food arrangements than Karlsruhe btw !) and then moved on to the Kerneltrack Keynote by Andrew Morton

The talk went over the which trees exist , what's the model for review, and how to get your code included...

Andrew wonders if kernels are getting les stable day by day , he thinks so but he doesn't have real data about it.. it's just an impression.
Maybe we should take all the kernel hackers after an event such as OLS and lock them in a room for 24 hours forcing them to do nothing else but fixing bugs

On to the questions then .. lots of them around ..

Q: USB is "slow" these days
A:...vfat is the actual problem ...
distributions implmented mounting usb sticks with a "stynk" option..
eheh :)

Q "Version numbers are marketing not technical stuff.. so where is 2.7 ? "
A "We have no plans on currently going to 2.7 or anything else..."

Q: why are the X drivers not in in kernelspace
A: because Linux isn't the only OS using X.

Q: What's the status of virtualisation in the kernel A: Andrew expect Xen to be in the linux tree within the next 12 months..
Xen people seem to be reluctant to talk to the general kernel people..
they talk more to the distro people ... Lets fix up Steve with Andrew tonight :)
Andrew doesn't want to choose between different virtualisation techniques.

Maybe rather than choosing there should be an API one way or another.. a common api that works for different types of virtualisation techniques.

There seems to be a need for a Meta Change log for the kernel rather than a great detailed changelog.. a quick 1-2 page overview where you can see in which compomponents stuff has drastically changed.

Isn't Kerneltrap and Zach's stuff enough summary ..

Matt's talk ...
ff course it didn't attract the huge public Andrew attracted but still fairly well attended , everyone who doesn't know what iSCSI is about , knows now.
Hmm.. so I wonder where my test setup has gone again .. I think I was exporting the USB disk at home via iSCSI. Maybe I should actually use iSCSI to export my MythTV and Music storage. Ah.. the lack of time..

Then on to the OCFS2 talk...It's extremely difficult to have to fly into a conference, give a talk, which you are not used to giving, didn't write yourselve but have to give since the marketing people planned.. annoying which explains why they guy presenting has .. some things on the slides he just doesnt know about.. so after about 25-30 minutes in the talk .. after the room has partly flooded the guy decides that rather than reading full parts of his notes he'd better start with a Q&A session. which was a really really good decision ...
but still .. We want Wim.. We want Wim .. We want Wim .. ..

May 05 2006

A load of C.R.A.P.

Someone finally found the correct acronym for what some people tend to call DRM it's
C.R.A.P : Content Restriction Annulment and Protection.

May 05 2006

LinuxTag 2006 Wiesbaden

We made it, Wiesbaden.. and even in a reasonable time slightly more than 3h30 minutes. Unlike our trip to Karlsruhe last year which took us trough every traffic jam in germany and changed our normal driving time from something like 4h20 to 9h. Time for a shower and breakfast.

May 04 2006

openMosixWiki - Guidelines

Wrote up some guidelines for #openmosix on Freenode,

http://howto.x-tend.be/openMosixWiki/index.php/Guidelines

Kind of started to annoy me that we get the same "Can openMosix do HA?" and ":Can openMosix scale my apache ?" kind of questions daily ..

The short answer.. no openMosic won't work for you.

(Hm.. strange.. I posted this yesterday but something ate my posting... )

openMosixWiki - Guidelines

May 03 2006

Me too

Liberie Cunha-Neto says "I need more time"

Me too! and for exactly the same thing ....

If only Whurley would send me that 60hourday2hoursleep patch.

May 01 2006

Linux Troubleshooting 101

Everything is a DNS Problem

Everything is a Fscking DNS Problem
No really, Everything is a Fscking DNS Problem
If it's not a fucking DNS Problem ..
It's an arp problem
If it's not an arp problem...
It's a Full Filesystem Problem
If youre filesystem isn't full
It's a Spanning Tree problem
If it's not a spanning Tree problem...
It's a USB problem
If it's not a USB Problem
It might be an ntp problem 
If it's not an ntp problem
It's a sharing IRQ Problem
If it's not a sharing IRQ Problem
But most often .. its a Freaking Dns Problem !