►
From YouTube: IETF93-NFSV4-20150722-0900
Description
NFSV4 meeting session at IETF93
2015/07/22 0900
A
A
B
B
B
F
H
B
B
B
J
H
B
I'm
gonna
go
buy
internet
time.
Let's
start
5ft
up,
let's
talk
after
we're
already
yup.
Okay
next
slide
sign
the
blue
sheets.
If
you
don't
sign
the
blue
sheets,
nobody'll
like
us
anymore
and
that's
an
old
go
back,
go
back
that
it's
now
organized
ation,
not
email,
we'll
fix
that
Spencer
and
I
keep
reusing
the
same
slides
all
in
thumbs
up.
The
note
well
is
correct.
You
would
have
read
it
where
you
would
have
had
a
gun
through
an
interstitial
on
your
way
to
registration,
to
read
the
note.
B
G
B
B
Okay,
one
more
comment
within
the
Spencer
and
I
between
us
sometimes
have
a
lack
of
communication.
The
minutes
for
the
92nd
meeting
did
not
get
posted
with
the
materials,
though
I
had
them
and
they
exist
in
the
ether
and
they
are
in
various
mailboxes
throughout
the
universe
I'm
going
to
attach
them
to
the
93
minutes.
Instead
of
trying
to
figure
out
I,
think
I
can't
Spencer
try
to
get
and
post
it,
but
he
missed
the
cutoff.
So
you
can
okay
I'll.
Do
that.
Thank
you.
Okay,
I.
F
F
Where
this
role
came
from
and
who
gave
it
to
me,
I
don't
know
but
I'm
now
the
document
editor
7
update,
yeah
a
success.
Well,
we
got
RFC
75
69
past.
This
is
the
registry
for
the
label
that
FS,
among
other
things,
but
in
case
anyone
else
wants
to
start
using
the
different
mac
models.
They
could.
So
it's
not
specifically
tied
to
be
for.
F
L
B
F
B
B
M
M
You
know
today
it's
very
annoying
to
customers
that
you
know
the
copy
file
over
in
the
face
and
they
lose
the
extended
attributes
that
they
used
very
commonly
I
believe
that
there
is
a
strong
interest
in
community
are
to
edit
and
some
of
the
comments
that
the
coming
back
in
wine
not
named
attributes.
You
know
we
tried
it,
it
didn't
go
anywhere,
so
this
is
a
simpler,
easier
set
of
our
pieces
to
internet.
A
B
B
P
C
P
I
I
M
Much
more
lightweight
banking
mechanism
anyway,
so
the
actual
extended
avenues
that
you
know
I
started
from
what
about
two
years
ago
and
presented
it
in
those
different
meetings.
If
anybody
is
interested
but
fully
enough
territories,
will
you
know
start
to
make
some
truck
and
get
you
know
the
graft
going.
M
M
The
objection
in
the
past,
you
know
people
immediately
were
associating
it
with
extended
attributes
on
linux,
the
issues
with
the
different
prefixes
of
security
and
trusted
we're
not
going
to
get
into
that.
This
is
just
basically
user
application,
external
reviews
that
are
completely
opaque
to
the
client
and
the
server
the
server
and.
M
F
P
F
M
So
the
four
basic
operations
that
are
going
to
have
is
least
in
the
name
of
that
abuse,
its
list
of
keys,
given
a
key,
get
the
external
attributes
of
the
value
associated
with
that
setup.
Fajada
Butte,
which
is
given
a
value
and
key,
basically
said
it,
save
it
and
given
a
key,
just
remove
that
extent
attribute.
M
We're
going
to
have
a
new
recommended
attributes
at
the
basic
getattribute
operation
that
basically
will
tell
a
client
if
external
attributes
are
supported
or
not,
and
then
we're
going
to
have
four
new,
our
pcs
to
implement
the
gates
at
least
and
remove
excess
other
goods.
We're
also
gonna
head
to
so
much.
P
K
F
A
M
That's
basically,
what
I'm
going
through
more
details
of
you
know.
The
draft
we're
gonna
have
a
new
bit
in
the
gate.
I
deduce
that
will
tell
you
if
this
is
supported
or
not
next
and
some
definitions
of
what
is
extended
attributes,
basically,
in
order
to
key
a
name
and
a
value,
then
enough
funding
in
the
two
new
baits
for
permissions.
M
There
was
run
one
comment
on
the
draft
already.
This
is
much
detail
here,
but
actually
the
only
comment
that
came
back
on
the
draft
that
was
possible
manuals
from
crystals
and
Benny
about
what
is
the
error
that
you
can
get
if,
basically,
the
value
that
you're
about
to
return
is
bigger
than
a
negotiated
session?
M
M
A
M
F
M
M
Your
connects
removed
most
likely
given
an
end
you're
going
to
remove
that
extent
attribute
sure
permitted
again.
All
those
operations
are
very
simple,
the
linguist
one
at
a
time.
The
first
draft
was,
you
know,
overloading
a
bunch
of
different
things
too,
on
our
pc
and
a
significant
Dacian
of
the
interest
from
Christopher's
to
break
it
down
to
simple
operations.
M
The
next
and
least
this
is
the
more
complicated
one,
and
here
we
had
to
go
to
some
model
that
will
allow
you,
if
the
if
the
thing
is,
does
not
fit
into
the
session
sighs,
then
you'll
be
able
to
come
back
with
a
cookie
and
it's
basically
the
logic
of
radio
credit
you
can
continue.
This
is
basically
just
giving
you
the
list
of
the
names,
not
actual
values,
but
again
they
can
be
long
bigger
than
you
know.
You
can
fit
into
one
buffer,
so
you'll
have
the
option
of
leading
them
in
in
chunks
and.
F
F
O
M
O
A
C
P
Sexy
other,
if
I
don't,
has
a
circus
well
I
day
the
club
that
worked
up
suppressed
for
it
again
get
unfollowed
level
on
selectors
different
things
when
positives
might
have
a
limit
for
the
length
of
the
key.
Oh
I
see
a
might
have
a
limit
for
the
lines
of
a
given
value.
They
might
ever
load
for
all
the
values
for
a
file
edit
out
Cashin.
In
so
I
know,
his
linux
is
pretty
inconsistent
reporting
bureaus
I,
suspect.
You
know.
Devices
are
better
in
different,
okay,
so
having
a
specific
terror.
P
M
There
was,
if
you
all,
if
you
have
a
if
your
client,
the
kind
does
not
have
the
legation,
then
it
you
know
it
can
catch
the
attributes,
and
you
know
check
validated
basically
will
change
attribute
to
see
that
he
has
up
to
date,
but
if
he
wants
to
update
it
it
should
be
done.
Synchronous
in
case
you
know,
somebody
else.
Has
a
delegation
or
below
you
just
don't
want
to
keep
it
on
your
cash
and-
and
you
know
assume
that
it's
set
and
allowing
other
people
to
change
it.
So
you'll
have
to
do
it.
Synchronously.
M
F
So
procedurally
I
thought
we
agreed
last
meeting,
make
it
work,
attachment
yeah,
so
I
think
what
you
have
to
do
is
you
have
to
take.
You
have
to
resubmit
the
draft
in
the
draft
ietf
whatever
for
and
when
you
do,
that
it'll
send
email,
dispenser
and
BB
telling
them
that
you're
requesting
it
via
a
working
group
document
and
they'll
promote
it.
At
that
point,
I.
M
M
F
M
I
P
Alright,
so
this
is
the
crazy
world
of
bypassing
NFS
for
storage
io,
when
using
good
old
block
storage,
that
most
people
probably
don't
care
too
much
anyway.
So
we
have
the
PFS
block
layout,
RFC,
5,
6
63,
ratified
down
about
five
years
ago
in
2010
I
actually
started
implementing
it
and
it
turned
out
that
we
actually
needed
a
lot
of
carotid.
N
P
P
So
I
was
just
running
the
basic
falsin
test
like
having
huge
data
integrity,
holes
just
corrupting
data
crashing
the
colonel
easily
exploitable.
There
are,
as
far
as
I
know,
known
on
Linux
clients.
So
even
asking
and
figuring
around
I
haven't
found
one
anywhere
on
the
server
side.
The
first
implementation
was
from
MC
and
tnx.
P
They
had
a
predecessor
to
that
in
earlier
products
which
wasn't
standardized
and
a
linux
server
that
I
wrote
of
merged
early
this
year
and
a
nasty
partner
that
people
actually
got
upset
about
is
because
I
assume
no
one
would
have
even
dared
to
use
decline
because
it
was
so
broken.
But
some
people
were
using
it
as
an
early
prototype
against
Lucy
and
the
gnu/linux
climbing
the
MC
server
don't
interact.
P
Actually
the
old
one
didn't
really
do
either,
but
people
didn't
notice
because
they
were
trying
hard
enough,
and
the
problem
is
that
the
Linux
virtual
memory
subsystem
really
doesn't
support
file
system
blocks
as
an
allocation
unit
that
are
bigger
than
the
vm
page
size
and
the
old
coke
kind
of
worked
around
it
in
a
way,
that's
very
easily
dead
lockable
for
the
new
one
I
removed
it
and
there
were
two
people
that
got
a
little
upset
about
it.
But
it's
not
much.
P
I
can
help
them
right
now,
so
to
go
back
to
the
the
issues
in
detail
with
the
with
the
spec
outside
the
bits
easily
fixed
by
errata.
The
real
main
issue
is
fencing.
So,
whenever
we've
got
a
misbehaving,
client
crashed
or
not
responding
to
recalls,
that's
the
main
issue.
We
need
to
cut
it
off
and
the
traditional
way
to
do
this
for
block
storages
is
to
have
fencing
at
the
switch
level.
Sometimes
at
the
array,
but
at
least
a
fibre
channel
wrote
the
classic
way
to
do
it
was
switch.
P
The
problem
is
that
the
protocol
doesn't
really
talk.
The
RFC
for
the
pro
called
isn't
really
talk
about
it,
so
it
basically
tells
the
MDS
to
fans
decline
and
the
only
thing
the
MDS
knows
about
the
client
is
the
IP
address
it
connected
on
which
trying
to
assumes
a
that
your
block,
storage
protocol
is
ip-based,
be
assumes
you're,
actually
using
the
same
IP
address.
N
P
P
As
well
and
the
try
the
attempt
fix
for
that
is,
and
it's
another
RFC,
which
is
RC
66
88,
which
basically
uses
a
partition
format,
a
very
common
one
out,
the
employee
efi
partitions,
which
most
people
use
these
days
and
basically
has
a
special
partition
type.
That's
a
PMF
s,
volume
which
clients
are
not
supposed
to
touch.
It's
not
actually
implemented
anywhere,
at
least
on
the
client
side.
As
far
as
I
know
it's
very
advisory.
It
requires
your
client
to
actually
look
at
that.
A
P
So
what
I
try
to
do
to
fix
all
these
issues
is
I
came
out
with
a
new
draft
for
you
laid
out
type,
which
I
call
the
scuzzy
layouts,
and
it
literally
takes
everything.
That's
on
the
block,
layout
and
tweaks
a
few
bits
and
pieces,
and
the
main
feature
here
is
that
we
use
scuzzy
reservations
or
persistent
reservations
which
are
used
between
different
users
of
scuzzy
block
storage
to
coordinate
access
to
it.
So,
basically,
you
can
cut
off
initiators
from
excess
at
all.
P
You
can
only
give
them
reap
style,
excess
and
there's
all
kinds
of
way
to
kick
people
out
of
that
pool
during
and
the
other
thing
is
that
we're
going
to
use
scuttling
specified
identifiers
to
find
our
volumes.
So
the
it's
actually
not
the
nicest
spec
for
finding
it
Baku
stirs
what
seven
different
variants
to
actually
identify
a
volume,
but
at
least
they're
all
standardized-
and
this
is
how
I
came
up
with
it,
and
the
status
is
that
I
submitted
a
first
personal
oft
in
2014.
P
We
got
it
accepted
as
a
working
group
document
in
April.
The
changes
have
been
pretty
minor
to
the
first
version.
I've
got
prototypes
for
the
client
and
server
implementations
and
I'm
planning
to
actually
give
them
up
to
full
status
and
submit
them
really
soon,
which
also
means
I
really
wanted
get
the
draft
moving
forward.
So.
P
L
They
were
black,
just
hang
up
to
ya.
I'll.
Take
oh
I'll,
take
a
look
at
this
and
yeah.
The
schedule
layout
is
doing.
The
schedule
layout
is
doing
the
right
things
when
we
started
working
with
what
became
the
block
layout
of
predated
in
one
case
appreciated
one
of
the
mechanisms
that
that's
being
used
and
I
think
the
other
case
there
was
it
predated
of
usable
implementations,
of
precision,
reservations,.
P
So
that
is
the
part
of
the
document,
and
the
other
one
is
the
the
way
I've
structured
it
right
now.
I
really
need
an
update
to
5
6
63
to
incorporate
the
errata
or
move
to
the
next
slide.
What
my
other
idea
would
be
is
one
thing
is
we
could
try
to
divorce
the
scuzzy
layout
from
the
blog
layout
to
the
way
I
wrote
it
right
now
is
literally
take
the
blog
layout.
Xd
are
in
description,
and
here
is
a
couple
items
which
means
it's
a
really
small
speck
right
now.
P
It
probably
makes
the
spec
a
lot
easier
to
read,
because
that
was
actually
one
of
the
main
confusion
with
the
reviewers
and
it
could
actually
allow
changes
to
the
existing
XD
are
specified
in
block
and
there's
only
two
things
I
could
come
up
with
for
that,
and
one
is
that
I'd
really
like
to
reduce
the
size
of
the
layout
to
make
payload,
which
has
giant
waste
in
it
right
now.
The
other
thing
I
could
come
up
with
in
theory,
but
not
in
practice
is
figure
out.
P
If
there's
anything
interesting
layout
specific,
we
want
to
send
in
layout
stats,
which
made
it
to
42,
which
wasn't
around
at
the
time
of
the
blog
layout
yet,
and
on
the
next
slide.
I've
got
a
little
more
detail
on
the
layout,
commit
payload
and
Rhyno.
What
the
block
layout
says
sentence
is
an
array
of
the
PMF
s
block,
extend
structure
which
is
the
same
structure.
P
We
use
in
layout
gap
and
then
have
a
device
ID
and
offset
in
the
final
lengths
and
offset
on
the
storage
and
the
state,
and
the
spec
actually
specifies
the
state
must
always
be
invalid
anyway.
So
that's
already
taken
out
and
the
way
the
server
has
to
work
I'm,
pretty
sure
to
the
server
always
needs
to
be
able
to
look
up
which,
although
which
divides,
IP
and
which
storage
offset
it
handed
out
for
a
given
layout,
because
that's
important
for
commit
that's
important
for
a
recall.
P
So
the
server
really
needs
a
tracking
structure
that
perhaps
from
this
to
be
to
other
members
that
matter
and
if
we
can
get
agreement
on
that,
we
can
literally
shrink
each
record
from
44
to
16
bytes
and,
given
that
on
workloads
that
do
small
random
writes
the
commit
payloads
are
huge.
That
would
actually
be
a
significant
improvement
in
terms
of
network
traffic
and
that's
pretty
much
it.
L
If,
if
I
understood
what
you
said,
it
was
basically
you
you've
got
in
rotation
of
56
63,
that
sort
of
limps
you've
got
a
better
layout
implementation.
That
gets
it
right
with
respect
discuss
soon.
As
part
of
that,
you
want
to
clean
up
a
protocol.
I
don't
see
any
week,
isn't
to
stick
with
the
old
protocol
for
the
for
the
old
layout,
given
the
new
layout
is
going
to
is
going
to
have
to
do
things
differently.
Anyhow,
courtesy
at
least
a
persistent
for
precision,
reservations.
B
L
L
B
L
Let's
see
if
that
means
six
words
in
Christoph's
mouth,
it
means
you
ep
6
63
get
to
get
your
mind
warped
and
he'll
on
what
word
for
you,
a
self-contained
darkness.
Cuz
you
layout
is
the
right
thing
to
do
there
and
then
independently.
We
can
figure
out
if
you
want
to
do
anything
to
366.
If
you
just
leave
it.
L
R
R
L
B
P
So
the
only
other
objection
I
heard
a
while
ago,
actually
should
have
done
a
slide
on
that
is
there
were
a
few
people
not
directly
involved
concerned
about
non
skazhi
block
storage
back
then
I
didn't
see
if
I'm
tender
now
actually
involved
with
another
contender,
which
is
nvmd,
especially
in
the
fabric
version,
which
actually
has
persistent
reservations
and
it
does
a
comfortable
EVP
deep
age.
These
days
sure
you
write
another
gram,
no.
E
P
L
P
L
L
B
F
F
So
we
talked
about
since
that
a
few
IDs,
which
is
how
we
allow
for
fencing
I,
made
the
mistake
of
labeling
the
title
security
instead
of
client,
fencing
and,
of
course,
that
pools
p
bellow
the
woodwork
like
crazy
right,
so
I've
got
a
lot
of
discussion.
Well,
it's
actually
very
good
in
that.
It
taught
me
my
lesson
because
when
I
push
this
up
to
a
next
level,
if
it
had
the
word
security
there,
I
would
really
drives
the
cockroach
gel
is
right,
so
a
very
good
process.
F
So,
the
again
the
synthetic
you
IDs
are
strictly
there
as
a
means
of
doing
fencing
and
if
you
want
Kerberos,
don't
use
the
loose
coupling
use
the
type
coupling
don't
use
this
synthetic
get.
Ids
have
the
same.
You
ID,
or
whatever,
attached
to
both
the
MDS
and
ids
for
the
storage.
If
you
want
to
use
loose
coupling,
my
suggestion
is:
that's
more
of
an
implementation
choice
than
a
speck
definition.
F
So,
for
example,
you
could
take
the
MDS
and
you
can
make
it
a
KDC
such
that
it's
issuing
tickets
to
the
client
that
work
for
both
the
file
on
the
MDS,
the
metadata
file,
and
then
the
data
part
portions
on
the
vs
I've
had
several
other
suggestions
of
ways.
This
could
be
done,
but
they
all
turn
out
to
be
very
implementation-specific
rather
than
we
do.
We
want
to
specify
this
in
a
second
again.
This
comes
back
to
this
is
not
this.
Instead
of
you,
I
DS
Arnott
a
security
model.
They
are
as
I'm
saying.
F
So
you
know
wearing
my
other
hat
with
my
company.
We
go
through
doing
an
implementation
of
the
flex
file
server
and
we
wanted
to
separate,
read
verses
right
so
that
we
don't
have
to
keep
a
timestamp
consistent
on
them
both
and
we
want
to
make
the
updates
accumulative.
So
we've
done
that
so
then.
We
then
separate
on
a
moment
on
the
other
side,
but
we
separate
basically
between
operations
and
bikes.
What
type
of
operations
were
said?
F
F
So
we
have
this
structure
of
Iowa
latency,
for
we
have
the
operations
requested
operations
completed,
bite
requested
by
completed
in
the
bikes,
not
delivery
being
talked
about,
and
then
we
have
the
busy
time
which
I
forgot
to
write
down
and
I'm
I'm
jetlite
as
well,
and
the
aggregate
completion
times
in
busy
times
how
much
time
I
spent
no
I'm
very
good.
It's
a
reading
again
a
truant
actually
wrote
this
part
so
and
then
the
the
agri
completion
time,
how
long
they
take
to
complete
the
request
down.
F
If
it's
right
down
to
stable
storage,
it
so
read
to
come
back,
please
so
the
status
we
we
actually
tested.
This
is
the
last
big
athan.
We
brought
our
server
and
we
were
able
to
get
you
up
and
running
the
first
day
and
we've
already
delivered
the
client
into
the
links
Colonel
we've
sent
wash
on
patches
upstream
for
the
layout
type
and
I've
been
accepted,
so
you
can
decode
test
and
it's
time
for
another
working
for
blast
all
right.
My
goal
is
to
get
these
the
layout
type
of
sign
to
us
before
Christmas
document.
F
F
F
L
B
A
F
F
Alright,
so
requirements
for
layout
types
could
be
a
Scalise,
so
it
lives.
If
you
remember
actually
that
this
is
the
closest
I've
come
to
be
using
a
slide.
Deck
I
wanted
to
keep
the
same
titles
from
last
time
and
that
the
same
content
last
time
I
said
it
lives
!
this
time
it
lives.
So
there
was
confusion
on
the
status
we
had.
I
affect
an
email
was
shared
that
went
to
a
working
group.
F
Basically,
that
said,
since
we've
changed
from
an
info
to
a
standard,
here's
a
set
of
issues,
and
that
only
came
to
me
evidently
and
so
I
I.
I
thought
well
we're
not
proceeding
with
this.
I
have
to
go
address
these
comments
and
then
I
started
getting
email
from
iesg
about.
Oh
we're,
reviewing
your
document,
and
this
is
an
Islamic
very
confused,
so
it
actually
went
through
the
ice
I
ESG.
There
were
some
comments.
I
did
not
address.
F
Oh
I
will
address
them,
but
you
know
we
had
this
I
think
David
finally
sent
out
the
email
last
week.
Saying
here
are
the
set
of
objections.
I
have
that,
since
we
went
from
info,
that's
very
informational
to
us
a
standard.
He
wanted
to
be
a
little
bit
more
stricter
about
so
right
now,
so
those
finally
came
out
and
I
think
Martin's
going
to
talk
so
we'll
look.
N
F
F
F
One
word
on
the
document
changed
the
whole
whatever
so
I
want
to
address
the
issues
raising
the
address
days,
comics
first,
because
that'll
dictate
how
I
do
the
next
part,
which
is
to
address
the
issues
raising
the
iesg
I,
have
to
engage
in
early
I
have
to
go
address
those
before
we
send
it
back
out
for
another
working
group
last
call
and
then
send
it
back
to
the
iesg.
It's
the
nice
thing
about
having
detailed
comments
is
you
can
go
address
them
individually,
very
succinctly.
F
It
was
just
again
the
sheer
magnitude
of
the
emailed
versus
where
I
thought
I
was
then
I
distract
down
that
was
like
you
know,
I,
don't
think
there
is
there
another
page,
nope.
Ok!
So
again,
it's
mainly
providing
detail.
Now
that
we're
very
you
know,
gonna
be
much
more
standardized.
I
thing
else,
I
think
Europe.
O
Alright,
this
is
andy
Adamson
I,
don't
have
any
slides
on
this,
but
the
multi
domain
draft
that
I've
been
working
on
for
several
years
now.
Sort
of
sort
of
similar
to
the
layout
types
in
the
sense
that
is
is
clarifying
some
of
the
the
issues
that
are
there.
You
know
presented
inside
the
nfsv4,
0
and
4.
Doubt
one
dress
on
on
how
you
run
a
multi
domain
name
space,
especially
with
fed
FS.
So
right
now
it's
a
standards
level
draft
I'm
has
gone
through
a
lot
of
review.
Where
does
it
stand?
O
B
O
So
you
could
read
it
in
one
place
and
understand
how
things
actually
worked
and
that's
very
similar
to
what
I'm
doing
with
the
multi-domain
draft,
in
the
sense
that
you
know
some
things
are
sort
of
obvious,
but
when
you
put
them
all
together
you
it
makes
it
very
clear
as
to
how
do
you
run
one
of
these
things?
So
so
I
guess
my
questions
are
number
one.
F
O
There's
a
couple
of
months
in
there
someone
some
of
them
come
from
capital,
must
some
of
them
come
from
the
draft
ins
5661
and
some
of
them
don't
such
for
it
is,
for
instance,
you
must
not
use
assess,
but
they
could
be
worded
as
an
informational
draft
and
it
would
be
it
would
serve
its
purpose
in
that
way
as
well.
There's
there's
no
protocol
in
you
know
just
like
the
layouts
type,
there's,
no
there's
no
protocols
pieces
inside.
It
is
just
how
to
run
everything.
David
I'm,
real
good.
L
Martin
as
I
say
this,
but
my
recollection
is
that
with
suitable
disclaimers
and
we
can
find
you
the
magic
work,
Bert's
2119
must
shoulds
and
maze
can
be
used
and
informational,
doc,
good
and
I
think
this
is
I.
Think
information
is
probably
correct
because,
on
the
one
hand,
you're
telling
protocol
designers
how
to
get
it
right.
Yes,.