DAFS?

Started by Nonameover 23 years ago16 messagesgeneral
Jump to latest
#1Noname
ghinchliffe@vianetworks.co.uk

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

#2Bruce Momjian
bruce@momjian.us
In reply to: Noname (#1)
Re: DAFS?

I know some folks are using NetApp and haven't had any problems. We
don't officially support it because we discourage NFS because the file
semantics are not the same as Unix file semantics.

---------------------------------------------------------------------------

ghinchliffe@vianetworks.co.uk wrote:

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073
#3Medi Montaseri
medi.montaseri@intransa.com
In reply to: Bruce Momjian (#2)
Re: DAFS?

Any feature (or current) support of raw devices. With iSCSI on the move,
that would be
a good thing.

Bruce Momjian wrote:

Show quoted text

I know some folks are using NetApp and haven't had any problems. We
don't officially support it because we discourage NFS because the file
semantics are not the same as Unix file semantics.

---------------------------------------------------------------------------

ghinchliffe@vianetworks.co.uk wrote:

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

#4Bruce Momjian
bruce@momjian.us
In reply to: Medi Montaseri (#3)
Re: DAFS?

No, we feel raw devices would be only a small win with a major pain to implement.

---------------------------------------------------------------------------

Medi Montaseri wrote:

Any feature (or current) support of raw devices. With iSCSI on the move,
that would be
a good thing.

Bruce Momjian wrote:

I know some folks are using NetApp and haven't had any problems. We
don't officially support it because we discourage NFS because the file
semantics are not the same as Unix file semantics.

---------------------------------------------------------------------------

ghinchliffe@vianetworks.co.uk wrote:

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

---------------------------(end of broadcast)---------------------------
TIP 5: Have you checked our extensive FAQ?

http://www.postgresql.org/users-lounge/docs/faq.html

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073
#5Medi Montaseri
medi.montaseri@intransa.com
In reply to: Bruce Momjian (#4)
Re: DAFS?

FYI, some of my kernel engineer friends are totally scared of File
Systems on top of block devices,
particularly the buffer cache....maybe they know something....I was
always under the impression
that DB engine writers don't like to deal with Filesystems as a file
could consists of
spread blocks (even zero padded ones) and the only way to be sure that
you do have
contingues blocks is to go after the raw device itself.

Anyways, thanks for the reply....

Bruce Momjian wrote:

Show quoted text

No, we feel raw devices would be only a small win with a major pain to implement.

---------------------------------------------------------------------------

Medi Montaseri wrote:

Any feature (or current) support of raw devices. With iSCSI on the move,
that would be
a good thing.

Bruce Momjian wrote:

I know some folks are using NetApp and haven't had any problems. We
don't officially support it because we discourage NFS because the file
semantics are not the same as Unix file semantics.

---------------------------------------------------------------------------

ghinchliffe@vianetworks.co.uk wrote:

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

---------------------------(end of broadcast)---------------------------
TIP 5: Have you checked our extensive FAQ?

http://www.postgresql.org/users-lounge/docs/faq.html

#6Graeme Hinchliffe
graeme@vianetworks.co.uk
In reply to: Bruce Momjian (#2)
Re: DAFS?

but dafs isn't nfs. we were trialing netapp for storage but are going
back to local disks as the db ran VERY slowly in comparison. dafs should
accelerate things from what i have read.

graeme

On Tue, 29 Oct 2002, Bruce Momjian wrote:

I know some folks are using NetApp and haven't had any problems. We
don't officially support it because we discourage NFS because the file
semantics are not the same as Unix file semantics.

---------------------------------------------------------------------------

ghinchliffe@vianetworks.co.uk wrote:

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

--
Bruce Momjian                        |  http://candle.pha.pa.us
pgman@candle.pha.pa.us               |  (610) 359-1001
+  If your life is a hard drive,     |  13 Roberts Road
+  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

#7Doug McNaught
doug@mcnaught.org
In reply to: Graeme Hinchliffe (#6)
Re: DAFS?

Graeme Hinchliffe <graeme@vianetworks.co.uk> writes:

but dafs isn't nfs. we were trialing netapp for storage but are going
back to local disks as the db ran VERY slowly in comparison. dafs should
accelerate things from what i have read.

If it presents a POSIX filesystem API then PG should work OK with it.

-Doug

#8Dave Thompson
thompd@guppi.net
In reply to: Graeme Hinchliffe (#6)
Re: DAFS?

Interesting... I've been using DB2 7.2 on Linux on both local disk and with a
NetApp and have seen a performance increase going to the NetApp for storage.
We have a gigabit network connection between the two, but I'd have thought that
local disk would still be faster, but it wasn't.

On Wed, 30 Oct 2002, Graeme Hinchliffe wrote:

but dafs isn't nfs. we were trialing netapp for storage but are going
back to local disks as the db ran VERY slowly in comparison. dafs should
accelerate things from what i have read.

graeme

On Tue, 29 Oct 2002, Bruce Momjian wrote:

I know some folks are using NetApp and haven't had any problems. We
don't officially support it because we discourage NFS because the file
semantics are not the same as Unix file semantics.

---------------------------------------------------------------------------

ghinchliffe@vianetworks.co.uk wrote:

Hiya
We use NetApp filers for network attached storage, these support
the DAFS protocol. Is there any support or plan for support of this in
PostgreSQL? Would be VERY handy if it was supported.

DAFS info can be obtained at:

http://www.dafscollaborative.org

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

--
Bruce Momjian                        |  http://candle.pha.pa.us
pgman@candle.pha.pa.us               |  (610) 359-1001
+  If your life is a hard drive,     |  13 Roberts Road
+  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

-----
Graeme Hinchliffe BSc (Hons) _
Unix Systems Engineer (UK Datacentre) ASCII ribbon campaign ( )
- against HTML email X
VIA NET.WORKS & vCards / \
http://arc.pasp.de/
Tel +44-1925-484065
Fax +44-1925-484055

ICQ : 3842605 (name : Link)

www.vianetworks.co.uk

---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

http://archives.postgresql.org

Dave Thompson
dave@guppi.net

#9Vincent Janelle
random@goblinstudios.com
In reply to: Doug McNaught (#7)
Re: DAFS?

On 30 Oct 2002 08:47:18 -0500
Doug McNaught <doug@mcnaught.org> wrote:

Graeme Hinchliffe <graeme@vianetworks.co.uk> writes:

but dafs isn't nfs. we were trialing netapp for storage but are
going back to local disks as the db ran VERY slowly in comparison.
dafs should accelerate things from what i have read.

If it presents a POSIX filesystem API then PG should work OK with it.

-Doug

Doesn't appear to. A quick scan of the SDK docs appears as though as if
its a direct implementation to access the storage of a device supporting
it by applications, such as database servers over a network.. Kinda like
raw devices.

#10Doug McNaught
doug@mcnaught.org
In reply to: Graeme Hinchliffe (#6)
Re: DAFS?

Vincent Janelle <random@goblinstudios.com> writes:

On 30 Oct 2002 08:47:18 -0500
Doug McNaught <doug@mcnaught.org> wrote:

Graeme Hinchliffe <graeme@vianetworks.co.uk> writes:

but dafs isn't nfs. we were trialing netapp for storage but are
going back to local disks as the db ran VERY slowly in comparison.
dafs should accelerate things from what i have read.

If it presents a POSIX filesystem API then PG should work OK with it.

-Doug

Doesn't appear to. A quick scan of the SDK docs appears as though as if
its a direct implementation to access the storage of a device supporting
it by applications, such as database servers over a network.. Kinda like
raw devices.

Well then I can safely say it won't work with PG as distributed. ;)

-Doug

#11scott.marlowe
scott.marlowe@ihs.com
In reply to: Vincent Janelle (#9)
Re: DAFS?

On Wed, 30 Oct 2002, Vincent Janelle wrote:

On 30 Oct 2002 08:47:18 -0500
Doug McNaught <doug@mcnaught.org> wrote:

Graeme Hinchliffe <graeme@vianetworks.co.uk> writes:

but dafs isn't nfs. we were trialing netapp for storage but are
going back to local disks as the db ran VERY slowly in comparison.
dafs should accelerate things from what i have read.

If it presents a POSIX filesystem API then PG should work OK with it.

-Doug

Doesn't appear to. A quick scan of the SDK docs appears as though as if
its a direct implementation to access the storage of a device supporting
it by applications, such as database servers over a network.. Kinda like
raw devices.

Look and see if there's some code out there for your OS (Linux???) to
mount a remote network device like this in loop back mode.

Then you might be able to let the OS turn it into a file system for the
database, which would get you caching on the database server box at the
file system level, but block access across the network for speed.

Then test it as thouroughly as an Apollo mission. :-)

#12Vincent Janelle
random@goblinstudios.com
In reply to: scott.marlowe (#11)
Re: DAFS?

*shrugs*, I use shared attached storage. Routing this stuff through
nscd (which is what most of those user-space apps do) would be silly.

You'd still need a filesystem on top of it for postgres, which negates
the whole issue. You'd might was well just use a SAN then. Netapp
announced that they're adding an option for this, the original poster
might want to look into that.

On Wed, 30 Oct 2002 15:15:01 -0700 (MST)
"scott.marlowe" <scott.marlowe@ihs.com> wrote:

Show quoted text

On Wed, 30 Oct 2002, Vincent Janelle wrote:

On 30 Oct 2002 08:47:18 -0500
Doug McNaught <doug@mcnaught.org> wrote:

Graeme Hinchliffe <graeme@vianetworks.co.uk> writes:

but dafs isn't nfs. we were trialing netapp for storage but are
going back to local disks as the db ran VERY slowly in
comparison. dafs should accelerate things from what i have read.

If it presents a POSIX filesystem API then PG should work OK with
it.

-Doug

Doesn't appear to. A quick scan of the SDK docs appears as though
as if its a direct implementation to access the storage of a device
supporting it by applications, such as database servers over a
network.. Kinda like raw devices.

Look and see if there's some code out there for your OS (Linux???) to
mount a remote network device like this in loop back mode.

Then you might be able to let the OS turn it into a file system for
the database, which would get you caching on the database server box
at the file system level, but block access across the network for
speed.

Then test it as thouroughly as an Apollo mission. :-)

#13scott.marlowe
scott.marlowe@ihs.com
In reply to: Vincent Janelle (#12)
Re: DAFS?

Vincent, did you actually READ my post before writing yours?

I said quite clearly that the device would be MOUNTED. I.e. it would have
a file system on it.

My point being that if you remote mount it as a block device, then the
file system on the local machine caches (i.e. fast access) while the
remote machine provides (semi-)fast block access across the network.

diversity is a good thing.

On Wed, 30 Oct 2002, Vincent Janelle wrote:

Show quoted text

*shrugs*, I use shared attached storage. Routing this stuff through
nscd (which is what most of those user-space apps do) would be silly.

You'd still need a filesystem on top of it for postgres, which negates
the whole issue. You'd might was well just use a SAN then. Netapp
announced that they're adding an option for this, the original poster
might want to look into that.

On Wed, 30 Oct 2002 15:15:01 -0700 (MST)
"scott.marlowe" <scott.marlowe@ihs.com> wrote:

On Wed, 30 Oct 2002, Vincent Janelle wrote:

On 30 Oct 2002 08:47:18 -0500
Doug McNaught <doug@mcnaught.org> wrote:

Graeme Hinchliffe <graeme@vianetworks.co.uk> writes:

but dafs isn't nfs. we were trialing netapp for storage but are
going back to local disks as the db ran VERY slowly in
comparison. dafs should accelerate things from what i have read.

If it presents a POSIX filesystem API then PG should work OK with
it.

-Doug

Doesn't appear to. A quick scan of the SDK docs appears as though
as if its a direct implementation to access the storage of a device
supporting it by applications, such as database servers over a
network.. Kinda like raw devices.

Look and see if there's some code out there for your OS (Linux???) to
mount a remote network device like this in loop back mode.

Then you might be able to let the OS turn it into a file system for
the database, which would get you caching on the database server box
at the file system level, but block access across the network for
speed.

Then test it as thouroughly as an Apollo mission. :-)

#14Chris Gamache
cgg007@yahoo.com
In reply to: scott.marlowe (#13)
Creating a unique identifier...

I've created this elaborate scheme to create small unique identifiers for
external transactions. It is built on a compact representation of the
current_timestamp(6) + 2 peices of data (varchar(50)) + a sequence number. I'd
like the option of restarting the sequence at some later date, but I don't want
to replicate any IDs.

To shorten the identifier I thought it might be a good idea to run an MD5
digest using the functionality from contrib/pgcrypto on it. ("thought" and
"might" are the key words there...)

I just need another (thousand?) set of eyes to see if a) I've missed anything,
b) There's a better way... :)

-- Sequence: universal_sq
CREATE SEQUENCE universal_sq INCREMENT 1 MINVALUE 1 MAXVALUE
9223372036854775807 CACHE 1;

-- Creates an text/interger representation of the current time
-- with microseconds.
CREATE FUNCTION "dtid"() RETURNS "text" AS '
DECLARE
tstz timestamptz;
dtid text;
BEGIN
tstz := current_timestamp;
dtid := (((tstz::ABSTIME::INT4)/100 * 100000000::INT8) +
date_part(\'microseconds\',tstz))::INT8::TEXT;
RETURN dtid;
END;' LANGUAGE 'plpgsql';

SELECT encode(digest(dtid() || 'value1' || 'value2' ||
nextval('universal_sq')::text, 'md5'),'hex');

__________________________________________________
Do you Yahoo!?
HotJobs - Search new jobs daily now
http://hotjobs.yahoo.com/

#15scott.marlowe
scott.marlowe@ihs.com
In reply to: Doug McNaught (#10)
Re: DAFS?

For those interested in learning more about what DAFS is and what's been
done with it so far, here are the links I've found

http://www.dafscollaborative.org
http://www.eecs.harvard.edu/~vino/fs-perf/dafs/
http://www.cs.duke.edu/ari/dafs/

It appears there's been work to build in DAFS support for both Linux and
BSD, but there's not a lot of data on how far that support has gotten.

Basically, it looks like a SAN type of protocol.

#16Vincent Janelle
random@goblinstudios.com
In reply to: scott.marlowe (#13)
Re: DAFS?

and I read the DAFS SDK, which didn't make claims to that.

If you mount it as a block device, then you can put a filesystem on it.
Problem solved.

On Thu, 31 Oct 2002 09:06:04 -0700 (MST)
"scott.marlowe" <scott.marlowe@ihs.com> wrote:

Show quoted text

Vincent, did you actually READ my post before writing yours?

I said quite clearly that the device would be MOUNTED. I.e. it would
have a file system on it.

My point being that if you remote mount it as a block device, then the

file system on the local machine caches (i.e. fast access) while the
remote machine provides (semi-)fast block access across the network.

diversity is a good thing.