ext4 paper
Andres Dilger
============
ext4 coming
ext3 becoming outgrown, but layout compatible, but developed in separate tree to allow stability

uninitialized areas - faster checking, easier growing
unwritten extents - sys_fallocate() - fast deletes, metadata overhead reduction
nanosecond timestamps
inode versions
16TB file size
2^64 bytes filesystem size (up to)

------

delayed allocation (until page flush)
allows multiple block allocation
no alloc on short lived files
avoid out-of-order writes (gcc)
helps with alloc fragmentation
buffer flag tracking of delayed blocks
free blocks need to be tracked
not done yet: journal quota accounting

hi i/o rates use less cpu
effecient free space allocations
large and small file policies possible
continguous allocation for large files
reserved space for large file growth

large allocation:
 length normalized 2^n or stripe size
 align allocations to raid stripe
 avoids read-modify-write on raid 5/6
 per-inode preallocation space

state today:
all major features in 2.6.27, 
all but mballoc (multiblock alloc) in 2.6.26
e2fsprogs 1.41 handles most features

===============================================================
===============================================================
virtual server solutions

single guest summary is that linux-Vserver is quite good
Xen isn't quite as good but often fairly close, except on one benchmark

multiple guests scalability:
linux-Vserver doesn't scale like Xen or KVM

uber-conclusion: test your use-case footprint against 3-4 different solutions to carefully understand which ones might be most suitable.

linux-Vserver is most likely the one we want because it's not emulation tech. like the others.

lots of movement and support in the last year from IBM, Intel and
Google for containers - development in fast progress.


=================================================================
================================================================
corosync cluster engine
Steven Drake, RH

IBM and Redhat behind project
open source HA cluster engine

history-
 ATCA hotswap infrastructure - 2002
 SA Forum Implementation 2003
 Corosync - 2008 (SA something removed)

Totem - communication protocl :: check it out {recv ordering, sync, reliable}
multicast required - or broadcast for less performance
membership algorythm built in - sweet

memory checkpointing

service engine is heart of system

plugin architecture - {configuration system, service engines, simple
development}

openais.org
corosync.org
openais@lists.osdl.org

=================================================================
=================================================================

LTTng by Mathieu Desnoyers

currently working on patchset submission:
	users: ftrace, LTTng, kvmtrace, blktrace


=================================================================
=================================================================

Advanced XIP file system: axfs

Jared Hulbert, Numonyx

very impressive; used on ultra large systems like S390, as well
things like cell phones benefit greatly.  faster performance combined
with RAM savings (use that RAM for other things)

great to use from live CDs because fast booting and smaller than squashfs