Thread: RE: 2 computers 1hd 2 postgres daemons. Is it possib le?

Yeah, I was really just curious if the proposed system was at a system level
or a network level.  Having two physical machines share (at the system i/o
level) a storage unit is something I have never heard of.  Does such a beast
exist?  I assume other RDBMS's use replication to fulfill high-availability
requirements...

jc

-----Original Message-----
From: Tony Simopoulos [mailto:karkalis@earthling.net]
Sent: Monday, October 16, 2000 5:43 PM
To: Clark, Joel
Cc: pgsql-interfaces@postgresql.org
Subject: Re: [INTERFACES] 2 computers 1hd 2 postgres daemons. Is it
possible?


> Through an NFS mount maybe?
without a "distributed lock manager" sharing cannot be done efficiently.
NFS creates problems both desired features of this set-up.

1) High Availability
if the computer that is serving the NFS mount goes down, any other machine
using the "share" goes down as well -- i.e. single-point
of failure

2) Load Balancing
NFS is extremely expensive.  all that would be balanced is the database
logic; in place sorting, connection overheads, etc.
expensive reads and writes will all hit the NFS mounting server -- i.e.
single-threading every query to the disk i/o of one machine.

tonys.

>
> -----Original Message-----
> From: Tony Simopoulos [mailto:karkalis@earthling.net]
> Sent: Monday, October 16, 2000 3:03 PM
> To: Oscar Serrano
> Cc: pgsql-interfaces@postgresql.org
> Subject: Re: [INTERFACES] 2 computers 1hd 2 postgres daemons. Is it
> possible?
>
>
> a redundant, replicated, or parallel system (choose whichever adjective
> befits your desires) is being developed.  check out
> http://www.erserver.com/.
>
> sharing a single disk is an O/S & hardware problem.  what is your intended
> platform?
>
> good luck,
>
> tonys.
>
> ----- Original Message -----
> From: Oscar Serrano <oserra@fondos.net>
> To: Postgres interfaces <pgsql-interfaces@postgresql.org>
> Sent: Monday, October 16, 2000 2:15 PM
> Subject: [INTERFACES] 2 computers 1hd 2 postgres daemons. Is it possible?
>
>
> > Hi men :-)
> > Just one question for those of you who have spent many time with
postgres.
> > I want to create a load balancing and High Availability system with two
> > servers.
> > This two server have a common hard disk: A RAID 5 server connected to
them
> > by the LAN.
> > I have a hardware machine (WSD+ from RADWARE) that manages the load
> > balancing.
> > I want to have, two servers, and two postgres installations, one in each
> > server, but with a  common PGDATA directory in the RAID.
> > So, if one of the servers crashes, people can still get the information
> from
> > the other server.
> > Esentially, what I want is, two postgres (one in each machine) with a
> shared
> > PGDATA directory, and that both postgres can access and do Inserts in
the
> > tables, and when one postgres does an insert, the second sees what the
> first
> > has inserted.
> >
> > Thanks.
> >
> >
> >
>

"Clark, Joel" wrote:
> 
> Yeah, I was really just curious if the proposed system was at a system level
> or a network level.  Having two physical machines share (at the system i/o
> level) a storage unit is something I have never heard of.  Does such a beast
> exist? 

Yeah, they exist, though why they exist is beyond me.  Maybe because
windows systems crash all the time.  It doesn't make much sense to me,
after all it would seem that hard drive failure happens more often that
the system crashing for no reason.



-- 
Joseph Shraibman
jks@selectacast.net
Increase signal to noise ratio.  http://www.targabot.com

> "Clark, Joel" wrote:
> >
> > Yeah, I was really just curious if the proposed system was at a system level
> > or a network level.  Having two physical machines share (at the system i/o
> > level) a storage unit is something I have never heard of.  Does such a beast
> > exist?
>
> Yeah, they exist, though why they exist is beyond me.  Maybe because
> windows systems crash all the time.  It doesn't make much sense to me,
> after all it would seem that hard drive failure happens more often that
> the system crashing for no reason.
>
they exist for a variety of reasons.

most importantly however, a hardware cluster was one of the first technologies on the scene (and still a very important
onetoday)
 
that offered redundancy and load balancing.  remember, a system often crashes, not because of hardware failure, or o/s
failure
(excluding windows of course), but rather application software failure.  the need for hardware redundancy solved that
problemas
 
well as the remote problems of other non-disk related failures.  just because a failure doesn't happen often doesn't
meanit is
 
acceptable.  how do you think the banks, or airline traffic controlers would feel about that?

Digital basically pioneered the entire front, as their hardware and o/s were most often used in critical systems
demanding100%
 
uptime.  since then the technology has moved around a bit.  (arguably, one of compaq's main interest's in digital was
clustering
technology).  most unix systems have some kind of clustering available.  today, most outfits go the replication route.
however,for
 
truly mission critical systems, replication is still not an acceptable option, for the simple reason of too much
latencyand lossy
 
performance.  even with the giant leaps forward in network bandwidth, replication technology still has a long way to go
beforeit
 
will be considered for the most critical of applications.

one of my major issues with postgresql is its contractidory stance on performance:

how can you be a frontier database with a (at best) sloppy high availability/load-balancing strategy?

tonys.