Plenary sessions followed by around 700 people (my estimation – room pretty full and lot of queue for registration)
Look at Agenda
Jim Zemlin (LF)
He made a very lively and strong defense of Linux. He especially shows that 3 major IT evolutions of today (well 4 but forgot the last one ;-))lead to more Linux adoption. 1/ The evolution of Internet from a people connected network to a devices (running LInux mostly) network. 2. It supports the cost reduction everybody is looking at in IT. 3. The desktop evolution to a more large set of various devices (smartphones, MID, Netbooks, …) all running Linux.
He also made the case of LInux powering lots of critical systems from NYSE to Oil research in Brasil, from french parliament to CERN accelerator….
Dirk Hohndel (Intel) on MeeGo
Made a point around MeeGo evolution. Still not there completely, but lots of progresses being made. Insisted on the FLOSS nature of the project (no vendor lock-in, external repo available, collaboration, …)
Round Table (Linux Torvalds, Andrew Morton, Jim Zemlin) (LF)
Jim asked some questions to Linus and Andrew around their perception of the kernel dev. Andrew explained that he is the sort of guy doing this in order to help others, to what Linus answered he wasn’t so nice and was doing it for its own pleasure and needs ! An overall enjoyable session !
Lunch with various Red Hat Brasil (among them Arnaldo Carvalho de Melo) and HP Brasil (Clesio Takahashi) colleagues. Very interesting discussions around performances (perf, perfmon, system tap, ftrace – new for me). Also met with Ric Wheeler manager for FS dev at Red Hat.
Jeremy Allison (Google) on SaMBa for your product
SaMBa is a product *and* contains technologies (CIFS linux kernel client, lib used by KDE/Gnome, …)
SMB2 in next release.
Support CTDB (CLuster)
SSO (AD in v4 + LDAP and Kerberos)
He explained how vendors producing proprietary code can work effectively with the SaMBa team.
Success comes from discussing with the SaMBa team (business friendly in fact !).
Advocated the use of smbtorture to at least test what ever server chosen (even MicroSoft uses it !)
Of course an excellent session by a SaMBA co-creator. Had the opportunity to shake hands, as a former HP colleague😉
Josef Bacik (Red Hat) on Btrfs: the story so far.
Developped by Oracle (Chris Mason, Yan Zheng), Red Hat (Josef Bacik), Fujitsu, Intel and Dreamhost. HP tests
What is available:
Large scale support, SSD, Built in Raid (0/1/10), compression (zlib) & checksum, cow, snapshot, volume management & balancing (to handle better errors)
Dynamic metadata and instant mkfs, online resize and defrag, online device addition/removal.
Converter from ext3/4 => Btrfs – special df
Tries to minimize corruption and increase security.
What remains to be done:
However, offline fsck coming soon !!
Raid 5/6 and free space cache soon,
encryption, deduplication, online fsck, quotas sometimes
pbs: cow fragmentation, df lies (ignores metadata)
On Fedora use linux btrfs at boot prompt (but grub doesn’t support it yet)
Default in F16/17. Nothing special in F15
Another great session by the people making the code.
James Bottomley (Novell) on Is the future Open Source ?
Targeting managers or dev who needs to convince managers. James is presenting well, but this is more interesting for new comers to FLOSS.
Ric Wheeler (Red Hat) on Linux File and Storage Ssstems: Are we done yet ?
Impact of new storage technologies on Linux FS choices (SSD, external Raid array)
Improvements in data integrity (checksum at btrfs, DIF/DIX flags)
ext4 improved streaming perf and FS repair. Supports extends and > 16TB FS. (default for RHEL6)
A btrfs summary of previous conf.
SSD mkfs is nearly similar between ext3/4/xfs/btrfs – on SATA 1TB ext3/4 is way smaller
Same for writing 50M files of some kB. Anywhere SSD is just removing differences between FS !
fsck ext4 on 70TB consumes 10 GB of RAM (xfs is 30 GB !) so large storage requires large server today.
But 2TB disks are consumer products today.
1 billion files (100 KB) => 100TB (mkfs: 4h, fill it: 4days, fsck: 2.5h)
Extremely well documented and comprehensive presentation on the state of the Art around Linux FS. Recommended if you can see it.
Sunil Saxena (Intel) on Building MeeGo: where are we and what’s next ?
MeeGo = Moblin (Intel) + Maemo (Nokia)
There will be a Meego conf in Dublin in Nov 2010.
Focus on compliance between MeeGo devices.
Releases April and October.
Interesting but lots of repeats with Dirk Hohndel’s presentation of the morning in Keynote.
Dinner with various Red Hat employees and FLOSS fans. Lots of discussions on … food😉
Around 250 people for plenary.
Round Table (Jon Corbet – James Bottomley (Novell) – Christoph Hellwig (himself) – Ted T’so (Google) -Thomas Gleixner (Linutronic)
Various questions from attendees around scalability, worst problem solved, background to be a kernel developer, …
Jane Silber (CEO Ubuntu) on Consumerization of IT ans its implication for FLOSS
Using the Producer -> Distributor -> Consumer chain
Notion of Prosumer – Similarities with the FLOSS movement
Quality is overtaking cost in FLOSS adoption (Accenture)
Ubuntu has a design.canonical.com project to improve user experience + PaperCut + openusability.org
Presented 6 monthes release as innovative and now std, but now slow commpared to Apps Store
LaunchPad Build System https://help.launchpad.net/Packaging/SourceBuilds
Now have an App Store (Ubuntu Software Center) to increase delivery pipeline.
Jane presented some interesting concept, even if she doesn’t have the same impact as Marc S. when speaking of Ubuntu😉
Ted T’so (Google) on FileSystems in LInux
2.6.35 == 70 different FS.
However FS are hard to deal with.
Concentrating on the different workloads (embedded, desktop, servers, DB, HPC, HAC, CFS, and mythical general purpose FS) and storage types (HDD, Raid, Flash – MTD, SSD, PCI-e attached).
Distributed Cluster FS: ceph, or ext4 in no-journal mode
Workload making FS evolutions: Cloud and embedded (compression, large flash ubifs).
Limits in HDD techno evolutions (last from 2005 perpendicular recording).
Shingled drives: overlaping of tracks (not reliable yet, and restrictions in tracks order) = cheaper tapes !
PCI-e attached flash: 100kIOPS => 1MIOPS
Phase Change Memory: 2x slower and 2x power consumption as DRAM (in Labs): Universal memory ? Then no more FS !!
SMP scalability (32 cores in a die)
Still 10 years of interesting evolutions !
As usual Ted made a very good presentation gathering history anecdotes and vision of the future.
Lunch was in a nearby restaurant with a specialty of buffet. Eated with Evaldo Gardenali, who was very helpful during all my stay there.
Bruno Cornec (HP) on Continuous packaging with project-builder.org
Between 80/90 people attended. Well received + questions + Fedora contact + contribution proposals.
Presentation available at http://trac.project-builder.org/browser/devel/pb-doc/pb-presentation.odp
Helio de Chissini de Castro (Collabora) on Linux and Mobile
(In portuguese ! so was a bit hard to follow :-))
Positioning Meego vs Android, Blackberry, Symbian, … and showing the openess of the platform, portability across platforms of apps, Usage of Qt (highly portable and LGPL), C++ and other bindings (python)
Easier for C developers with QtCreator and Qt Quick App
Made an impressive demo of QtCreator, including the demo of a game launched without any compilation of code, just using js + qml.
Enjoyable presentation, even for non portuguese speaking people !
Vinod Kutty (CME group) on Migrating to Linux and Open Source
Mission critical with Linux. Insisted on serial line, ssh, service processor support. Operations from CLI, at OS level even for HW related ops (FW upgrade, queries). Would like a std (not smash) for service processors. Do not use IPMI as unreliable.
Glauber Costa (Red Hat) on The KVM compendium
(In portuguese ! so was again a bit hard to follow :-))
HIstory, features of KVM.
Only SpecVirt published is KVM (sc2010) – 48 wihtout tuning, 54 + hugepage, 60 + SR-IOV, 72 + node binding
Thomas Gleixner (Linutronic) on It’s really time for real-time
History of real time linux, with Linus quotes.
2 approaches: Linux as idle thread of nano/micro kernel or Linux with implicit RT properties.
2004 preempt RT started by Ingo Molnar (in kernel RT approach + std POSIX)
He explained problems around RT in LInux (interrupt handlers in thread, sleeping spinlocks, break per cpu assumptions, memory alloc)
Working on RT patches helps improving a lot lock bugs.
Results: latency worst case is < 50µs at a price on throughput and perf (disk I/O, networking) (5-25%)
Ikea machines are driven by Linux RT
Extremely interesting presentation, which render the topic cristal clear even for people with no particular RT background.
The day finished with an extemely nice diner at with Thomas Gleixner, Christoph Hellwig, … among others and some kernel related discussions which could allow me to be updated of the latest news.