Re: Oracle Parallel Server Option

Shirley

You're partially correct. FailSafe is one instance, one database on one node of an NT cluster. If the node dies, both the instance and the database are restarted on the remaining node. Parallel Server can be set up in one of two ways. Both of these have one shared database and two instances, one on each node (I'm using the simplest example here - OPS can run multiple instances on more than one node, but I'll leave that out for now).

1. Active - standby is where only one instance is being accessed by users. The other is started but not being used (by virtue of the fact that the active one is the first one in our address_list in the tnsnames.ora file). For disaster recovery purposes, this is the easiest approach to take (no pinging), but it's also the most expensive in terms of hardware (one node sitting idle all the time). The other problem is trying to stop management deciding to use the "wasted" node!
2. Active - active is where both instances are being accessed by users. The main advantage of this is it splits the load. The main disadvantage is the requirement to minimize pinging. Using this approach without forethought can sometimes cause performance degradation as locks and blocks switch backwards and forwards between nodes. Cache fusion in 8i is starting to address this problem in the read - write scenario, with a future release supposed to address write - write.

Now, back to the question about failover. Transparent application failover (TAF) requires the use of the OCI version 8 libraries. I believe (though I could be wrong) that the only tools that use the OCI v8 libraries now are OCI (of course), SQL*Plus and maybe ODBC. If you have an application using these, you can use TAF in one of three ways:

1. NONE - This option does not use any failover functionality. This can be explicitly specified to prevent failover from happening. This is the default behavior if no failover type is specified.
2. SESSION - In this option, if a user’s connection is lost, a second session is automatically created for the user on the surviving instance. This type of failover does not attempt to recover selects. This option should be selected if you do not want to automatically start the select after the failover and also, if you do not want the overheads involved during the normal select operations.
3. SELECT - This option allows users with open cursors to continue fetching on them after failure. This process is automatic and the user does not have to do anything for the SELECT to continue. The user sees it as if one SELECT operation has been processed. Internally, it involves re-executing the SELECT statement at the surviving instance and then fetching the same number of rows as had been fetched earlier. After this, the next fetch starts and is visible to the user. This mode involves some overhead on the client side in normal select operations.

OK, now we've covered the types of failover, which can be used for FailSafe and OPS? If you're using the v8 OCI libraries, the answer is either.

Regarding the single hostname question, I'm not sure how you've implemented this. All the failover setup is specified in the tnsnames.ora file, as are the different node names. Where does the single hostname come into the picture? If you have a single hostname, which points to a single IP address, then the only configuration you can use for OPS is active - standby, with IP address switching handled at the OS level if a node dies. FOr FailSafe, it only has one virtual hostname (provided by MSCS), so it's not a problem there.

HTH. Pete

Shirley Kwan wrote:

> So........FailSafe is work on one instance and OPS is work on at least two
> instances, right ? Which one will failover automatically without
> reconnection ?
>
> In our application, only one hostname can be set for database access. So is
> there only one hostname is visible for all clients if we use either OPS or
> FailSafe ?
>
> Thanks for your help !!!
>
> Shirley......
>
> Pete Sharman wrote in message <3767CA5E.DFFE42B0_at_us.oracle.com>...
> >OK. One of the things that is difficult to gauge in a newsgroup is whether
> >others are interested or not, so what I try and do after a few posts
> >backwards and forwards between myself and one other person, I take them
> >offline. Anyway, here's the details of what's being going on between
> >Shirley and I via email.
> >
> >FailSafe is an NT only product that allows a single instance database to
> >automatically failover from one node of an NT cluster to the other node, if
> >the first node fails. I think it's on a separate CD to the server CD, but
> >as I rarely use the CD's I'm not 100% sure of that.
> >
> >If you want to set up SQL*Net for the ability to connect to a primary node,
> >then a secondary node if the primary is down, all you need is two ADDRESS
> >lists for your alias, like this:
> >
> >ORCL.US.ORACLE.COM =
> > (DESCRIPTION =
> > (ADDRESS_LIST =
> > (ADDRESS = (PROTOCOL = TCP)(HOST = psharman-node1)(PORT = 1521))
> > (ADDRESS = (PROTOCOL = TCP)(HOST = psharman-node2)(PORT = 1521))
> > )
> > (CONNECT_DATA =
> > (SERVICE_NAME = orcl)
> > )
> > )
> >
> >If node 1 is not up, you'll connect to node 2. Note this is used for OPS,
> >not FailSafe. Also, this is NOT transparent application failover, that
> came
> >in 8.0. If the first node dies, the user will need to reconnect.
> >
> >HTH.
> >
> >Pete
> >
> >Nathan Phan wrote:
> >
> >> In article <3766754D.66E549E3_at_us.oracle.com>,
> >> Pete Sharman <psharman_at_us.oracle.com> wrote:
> >> > This is a multi-part message in MIME format.
> >> > --------------C24A24207264DA1264C5DE1E
> >> > Content-Type: text/plain; charset=us-ascii
> >> > Content-Transfer-Encoding: 7bit
> >> >
> >> > For those interested in this thread, I've taken it offline with
> >> Shirley. Let me
> >> > know if you're interested and we'll bring it back to the newsgroup
> >> again.
> >>
> >> Yes Pete, at least I am interested, the discussion just enter into the
> >> best part, don't kill me with that. Please bring the discussion on line
> >> again. Thanks !
> >>
> >> >
> >> > Pete
> >> >
> >> > Shirley Kwan wrote:
> >> >
> >> > > What is Oracle FailSafe ???
> >> > >
> >> > > For the Oracle OPS, you said that I need to set the tnsnames.ora to
> >> the
> >> > > active node first, then to the standby. Can you please give me an
> >> example on
> >> > > how to configure the tnsnames.ora ?
> >> > >
> >> > > Is it only one virtual host with shared media is visble to all
> >> client
> >> > > machine ???
> >> > >
> >> > > Thanks for your help !!!
> >> > >
> >> > > Shirley..........
> >> > >
> >> > > Pete Sharman wrote in message <375FE049.C9E9B762_at_us.oracle.com>...
> >> > > >Shirley
> >> > > >
> >> > > >If you have two servers A and B, the database is located on shared
> >> disks
> >> > > which
> >> > > >is visible to both A and B when you're using Parallel Server.
> >> Which SID
> >> > > and
> >> > > >host you need to point to is determined by what you're trying to
> >> achieve.
> >> > > Are
> >> > > >you using OPS as a hot standby (i.e. one node active, the other
> >> standby)?
> >> > > If
> >> > > >that's the case, your tnsnames.ora should have a connection string
> >> that
> >> > > points
> >> > > >to the active node first, then to the standby. If you have OPS in
> >> > > >active-active configuration, then point the users to the node they
> >> need to
> >> > > use
> >> > > >to minimize pinging etc.
> >> > > >
> >> > > >I'm not too sure of what you're trying to achieve, but it may be
> >> that
> >> > > Oracle
> >> > > >FailSafe is what you need rather than OPS?
> >> > > >
> >> > > >Pete
> >> > > >
> >> > > >Shirley Kwan wrote:
> >> > > >
> >> > > >> So, how can I setup the sqlnet configuration for client ???
> >> > > >> Which SID I use and which host I need to point to ???
> >> > > >>
> >> > > >> Say, if I have two server, A and B with SID A and B respectively.
> >> Where
> >> > > is
> >> > > >> the database locate ??? Is it locate in the share media in server
> >> C) ???
> >> > > >>
> >> > > >> For the sqlnet configuration for client, which one is correct ???
> >> > > >> (1) SID = A, host = A & SID = B, host = B (set both or either one
> >> ???)
> >> > > >> (2) SID = A, host = C & SID = B, host = C (set both or either one
> >> ???)
> >> > > >>
> >> > > >> Actually, we would like to use Oracle Parallel Server in NT
> >> clustering
> >> > > >> environment.
> >> > > >> In NT clustering environment, only one virtual host is visible
> >> for all
> >> > > >> client although we have two server. Is Oracle Parallel Server
> >> Option have
> >> > > >> the same concept ???
> >> > > >>
> >> > > >> Thank you very much for your help !!!
> >> > > >>
> >> > > >> Shirley Kwan......
> >> > > >>
> >> > > >> markp7832_at_my-deja.com wrote in message
> >> <7jj69e$kts$1_at_nnrp1.deja.com>...
> >> > > >> >No. You have one shared database, but two or more separate
> >> instances,
> >> > > >> >each with a distinct SID. Some hardware vendors provide an
> >> automatic
> >> > > >> >failover feature so that if one instance crashes the users are
> >> > > >> >automatically re-routed to another instance when they log back
> >> in. How
> >> > > >> >well failover works depends in part on how your users log in.
> >> That is
> >> > > >> >do they log into an application server or directly to a database
> >> server
> >> > > >> >and so on.
> >> > > >> >
> >> > > >> >There are data partitioning issues with OPS in that you want all
> >> > > >> >updaters for a heavy hit table to come from one machine. But
> >> that is
> >> > > >> >getting well beyond the basics. Note with Oracle 8 there is an
> >> > > >> >init.ora parameter that you must code to run parallel that does
> >> not
> >> > > >> >exist in ver 7.0 - 7.3. When this parameter is coded the
> >> database
> >> > > >> >always starts as if you entered: 'startup parallel'. There is a
> >> > > >> >Parallel Server Concepts manual on the documentation CD.
> >> > > >> >
> >> > > >> >
> >> > > >> >In article <AC6JM7Us#GA.153_at_pet.hiwaay.net>,
> >> > > >> > "Shirley Kwan" <sskwan_at_ingr.com> wrote:
> >> > > >> >> So it means that I need to setup two machines with two
> >> instance of
> >> > > >> >the same
> >> > > >> >> name. And both instances are running at the same time. A
> >> shared disk
> >> > > >> >needed
> >> > > >> >> to setup to contain the actual tablespace and all client
> >> machine need
> >> > > >> >to
> >> > > >> >> point to the shared disk. If one machine fails, all client can
> >> still
> >> > > >> >access
> >> > > >> >> Oracle through the instance of another machine.
> >> > > >> >>
> >> > > >> >> Is that correct ???
> >> > > >> >>
> >> > > >> >> Thanks in advance !!!
> >> > > >> >>
> >> > > >> >> Shirley Kwan
> >> > > >> >>
> >> > > >> >> markp7832_at_my-deja.com wrote in message
> >> > > >> ><7jgh7i$nfj$1_at_nnrp1.deja.com>...
> >> > > >> >> >In article <nd1ohZKs#GA.200_at_pet.hiwaay.net>,
> >> > > >> >> > "Shirley Kwan" <sskwan_at_ingr.com> wrote:
> >> > > >> >> >> Hi all,
> >> > > >> >> >>
> >> > > >> >> >> Does anyone have information about Parallel Server Option
> >> ???
> >> > > >> >> >> How can I implement it ?
> >> > > >> >> >> What's the basic system requirement ?
> >> > > >> >> >> What's the advantage of using Parallel Server Option ?
> >> > > >> >> >> .......
> >> > > >> >> >> --
> >> > > >> >> >> Shirley Kwan......
> >> > > >> >> >>
> >> > > >> >> >The Oracle Parallel Server product comes with the Enterprise
> >> edition
> >> > > >> >of
> >> > > >> >> >Oracle. In inorder to set it up you must have multiple
> >> hardware
> >> > > >> >boxes
> >> > > >> >> >from a vendor that supports shared file systems. With
> >> versions 7.0 -
> >> > > >> >> >7.3 you have to setup the hardware vendors DLM, distributed
> >> lock
> >> > > >> >> >manager, and then you have to create the database with
> >> multiple
> >> > > >> >> >instances. Oracle has taken over providing the DLM with
> >> either
> >> > > >> >version
> >> > > >> >> >8 or 8.1.
> >> > > >> >> >
> >> > > >> >> >One of the main advantages of OPS is failover. Should one
> >> machine
> >> > > >> >have
> >> > > >> >> >a hardware failure you shift the load to the other machine(s)
> >> > > >> >providing
> >> > > >> >> >you hold the load to 50% of machine capacity in a two box
> >> setup.
> >> > > >> >> >Another is if you are approaching machine capacity you can
> >> add
> >> > > >> >another,
> >> > > >> >> >in theory cheaper, machine of the same type rather than
> >> upgrade to a
> >> > > >> >> >bigger box. With recent advances in hardware capacity and
> >> reduction
> >> > > >> >in
> >> > > >> >> >cost I see the only real advantage being failover.
> >> > > >> >> >
> >> > > >> >> >
> >> > > >> >> >
> >> > > >> >> >--
> >> > > >> >> >Mark D. Powell -- The only advice that counts is the advice
> >> that
> >> > > >> >> > you follow so follow your own advice --
> >> > > >> >> >
> >> > > >> >> >
> >> > > >> >> >Sent via Deja.com http://www.deja.com/
> >> > > >> >> >Share what you know. Learn what you don't.
> >> > > >> >>
> >> > > >> >>
> >> > > >> >
> >> > > >> >--
> >> > > >> >Mark D. Powell -- The only advice that counts is the advice
> >> that
> >> > > >> > you follow so follow your own advice --
> >> > > >> >
> >> > > >> >
> >> > > >> >Sent via Deja.com http://www.deja.com/
> >> > > >> >Share what you know. Learn what you don't.
> >> > > >
> >> > > >--
> >> > > >
> >> > > >
> >> > > >Regards
> >> > > >
> >> > > >Pete
> >> > > >
> >> > > >
> >> >
> >> > --
> >> >
> >> > Regards
> >> >
> >> > Pete
> >> >
> >> > --------------C24A24207264DA1264C5DE1E
> >> > Content-Type: text/x-vcard; charset=us-ascii;
> >> > name="psharman.vcf"
> >> > Content-Transfer-Encoding: 7bit
> >> > Content-Description: Card for Pete Sharman
> >> > Content-Disposition: attachment;
> >> > filename="psharman.vcf"
> >> >
> >> > begin:vcard
> >> > n:Sharman;Peter
> >> > tel;fax:+1.650.633.1669
> >> > tel;work:+1.650.607.0109
> >> > x-mozilla-html:FALSE
> >> > url:http://www.oracle.com
> >> > org:Worldwide Internal Services Education;Oracle Corporation
> >> > adr:;;500 Oracle Parkway M/S OPL-B1024;Redwood
> >> Shores;California;94065;USA
> >> > version:2.1
> >> > email;internet:psharman_at_us.oracle.com
> >> > title:Project Manager
> >> > note;quoted-printable:SQL> select standard_disclaimer,
> >> witty_remark=0D=0A 2 from company_requirements;=0D=0A=0D=0A ****
> >> The statements and opinions expressed here are my **** =0D=0A **** own
> >> and do not necessarily represent those of **** =0D=0A ****
> >> Oracle Corporation. =20
> >> ****=0D=0A=0D=0A"Controlling application developers is like herding
> >> cats." =0D=0AKevin Loney, ORACLE DBA Handbook =0D=0A=0D=0A"Oh no it's
> >> not! It's much harder than that!" =0D=0ABruce Pihlamae, long term
> >> ORACLE DBA
> >> > x-mozilla-cpt:;29648
> >> > fn:Pete Sharman
> >> > end:vcard
> >> >
> >> > --------------C24A24207264DA1264C5DE1E--
> >> >
> >> >
> >>
> >> Sent via Deja.com http://www.deja.com/
> >> Share what you know. Learn what you don't.
> >
> >--
> >
> >
> >Regards
> >
> >Pete
> >
> >

--

Regards

Pete

Received on Fri Jun 18 1999 - 11:27:21 CDT