Few questions on Active/Active configuration (SQL 2000):
-Any articles on setting up a two node Active/Active configuration (ideally
step by step)? Almost everything out there is on Active/Passive.
-Since there are at minimum 2 instances running at the same time on a two
node active/active configuration, is there still 1 virtual IP address? If so,
how does sql server knows which instance an application/user is refering to?
-On an active/passive configuation, microsoft recommends two networks
(public/private) but it can be done with only 1 (not recommended but
possible). Does this also apply on an Active/Active configurtion? We are
using blade servers and only 1 network can be configured so I want to make
sure that Active/Active can work with only 1 network.
-We have 2 main databases and we are planning to have one database on each
node. Would'nt this have a better performance instead of having both database
on one server (active/passive). I am aware that clustering is for
availability only but it seems like we could get a better performance in
active/active. Is this a valid assumption?
Thank you all
--
System Eng.
MSc in CS, MCSE4, IBM Certified MQ 5.3 Administrator
"I saw it work in a cartoon once so I am pretty sure I can do it."
> -Any articles on setting up a two node Active/Active configuration (ideally
> step by step)? Almost everything out there is on Active/Passive.
Setting up another instance in a cluster is no different than setting up the
first instance. Books Online has detailed steps.
> -Since there are at minimum 2 instances running at the same time on a two
> node active/active configuration, is there still 1 virtual IP address? If so,
> how does sql server knows which instance an application/user is refering to?
You need a separate IP addresse for each clustered instance.
> -On an active/passive configuation, microsoft recommends two networks
> (public/private) but it can be done with only 1 (not recommended but
> possible). Does this also apply on an Active/Active configurtion? We are
> using blade servers and only 1 network can be configured so I want to make
> sure that Active/Active can work with only 1 network.
Still applies. But since the purpose of a failover cluster is to provide
high availability, you really shouldn't share a single network. That kind of
defeats the HA purpose.
> -We have 2 main databases and we are planning to have one database on each
> node. Would'nt this have a better performance instead of having both database
> on one server (active/passive). I am aware that clustering is for
> availability only but it seems like we could get a better performance in
> active/active. Is this a valid assumption?
You may get better performance by virute of running each database on its own
host, thus using separate processors, memory, network cards, HBA's, etc. That
is, if your database can take advantage of these separate hardware
components. So, if one of your databases is cpu bound, for instance, having
it on its own server would certainly help.
Linchi
"Sas" wrote:
> Few questions on Active/Active configuration (SQL 2000):
> -Any articles on setting up a two node Active/Active configuration (ideally
> step by step)? Almost everything out there is on Active/Passive.
> -Since there are at minimum 2 instances running at the same time on a two
> node active/active configuration, is there still 1 virtual IP address? If so,
> how does sql server knows which instance an application/user is refering to?
> -On an active/passive configuation, microsoft recommends two networks
> (public/private) but it can be done with only 1 (not recommended but
> possible). Does this also apply on an Active/Active configurtion? We are
> using blade servers and only 1 network can be configured so I want to make
> sure that Active/Active can work with only 1 network.
> -We have 2 main databases and we are planning to have one database on each
> node. Would'nt this have a better performance instead of having both database
> on one server (active/passive). I am aware that clustering is for
> availability only but it seems like we could get a better performance in
> active/active. Is this a valid assumption?
> Thank you all
> --
> --
> System Eng.
> MSc in CS, MCSE4, IBM Certified MQ 5.3 Administrator
> "I saw it work in a cartoon once so I am pretty sure I can do it."
|||Linchi, just make sure your application is active/active ready. Will
both nodes assign duplicate IDs such as customer or order numbers? Is
common application state stored in shared memory? Will a response get
returned to a wrong node?...and so on. www.availabilitydigest.com
On Jan 19, 3:38 pm, Linchi Shea <LinchiS...@.discussions.microsoft.com>
wrote:[vbcol=seagreen]
> first instance. Books Online has detailed steps.
>
> high availability, you really shouldn't share a single network. That kind of
> defeats the HA purpose.
> host, thus using separate processors, memory, network cards, HBA's, etc. That
> is, if your database can take advantage of these separate hardware
> components. So, if one of your databases is cpu bound, for instance, having
> it on its own server would certainly help.
> Linchi
>
> "Sas" wrote:
>
|||> Linchi, just make sure your application is active/active ready.
I don't know what you meant by 'active/active ready'. There is nothing you
need to ready your application for.
> Will both nodes assign duplicate IDs such as customer or order numbers?
The SQL instance is running on one node at a time. At no time, it would run
on both nodes, and no database can be accessed by more than one node at a
time.
> Is common application state stored in shared memory?
There is no shared memory.
> Will a response get returned to a wrong node?...and so on.
It doesn't matter what node is running the SQL instance. The instance is
accessed via a virtual IP address that is not the same as the node IP address.
Linchi
"jreederhi@.nac.net" wrote:
> Linchi, just make sure your application is active/active ready. Will
> both nodes assign duplicate IDs such as customer or order numbers? Is
> common application state stored in shared memory? Will a response get
> returned to a wrong node?...and so on. www.availabilitydigest.com
> On Jan 19, 3:38 pm, Linchi Shea <LinchiS...@.discussions.microsoft.com>
> wrote:
>
|||Linchi. It looks as if we are using different definitions of active/
active. To me, an active/active configuration is one in which two or
more nodes are executing the same application using synchronized
database copies. If one node goes down, its users can be switched to
a surviving node in seconds, thus virtually eliminating downtime due
to node failures. If you are using a single common database, there
should be no application issues. www.availabilitydigest.com
On Jan 27, 11:58 pm, Linchi Shea
<LinchiS...@.discussions.microsoft.com> wrote:[vbcol=seagreen]
> need to ready your application for.
> on both nodes, and no database can be accessed by more than one node at a
> time.
>
> accessed via a virtual IP address that is not the same as the node IP address.
> Linchi
>
> "jreede...@.nac.net" wrote:
>
>
>
>
|||That type of clustering is not supported by SQL Server 2005. Each instance
of SQL can run on a single node at a time. Other than a read-only shared
database model for reporting, there is no provision in SQL Server for
multiple SQL instances runing on different nodes to access the same data
store.
This is the major reason I dislike the Active/Passive and Active/Active
nomenclature. It is a holdover from SQL 7.0 clustering and does not
accurately reflect the capabilities of the current product. And 7.0 didn't
do this either, it just had a functional distinction between the primary
node and its failover partner that led to this unfortunate choice of names.
Geoff N. Hiten
Senior Database Administrator
Microsoft SQL Server MVP
<jreederhi@.nac.net> wrote in message
news:1170095155.783463.3200@.h3g2000cwc.googlegroup s.com...
> Linchi. It looks as if we are using different definitions of active/
> active. To me, an active/active configuration is one in which two or
> more nodes are executing the same application using synchronized
> database copies. If one node goes down, its users can be switched to
> a surviving node in seconds, thus virtually eliminating downtime due
> to node failures. If you are using a single common database, there
> should be no application issues. www.availabilitydigest.com
> On Jan 27, 11:58 pm, Linchi Shea
> <LinchiS...@.discussions.microsoft.com> wrote:
>
|||<jreederhi@.nac.net> wrote in message
news:1170095155.783463.3200@.h3g2000cwc.googlegroup s.com...
> Linchi. It looks as if we are using different definitions of active/
> active. To me, an active/active configuration is one in which two or
> more nodes are executing the same application using synchronized
> database copies.
Your definition is wrong. High Availability clustering is not about
horizontal scaling.
> If one node goes down, its users can be switched to
> a surviving node in seconds, thus virtually eliminating downtime due
> to node failures.
This is what Server Clustering, or MSCS, is - it is a high availability
platform that helps reduce downtime. It does not provide horizontal scaling.
Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp
The next ClusterHelp classes are:
Denver starting Feb 12th
NYC starting Feb 19th
London starting March 6th
|||Geoff: I suspect that the inability for two instances of a database
manager to access the same physical database is common. When we speak
of active/active, we mean that two nodes, each with its own database,
are cooperating in the same application. The two database copies are
kept in synchronism by some means, typically data replication. By
keeping replication latency small (hundreds of milliseconds with
today's products), the problem of data loss following a node failure
or of data collisions is minimized. Furthermore, the two nodes can be
geographically separated for disaster tolerance. (See http://
availabilitydigest.com/public_articles/what_is_active-active.pdf).
There are several products that support SQL Server that can provide
this capability. www.availabilitydigest.com
On Jan 29, 1:33 pm, "Geoff N. Hiten" <SQLCrafts...@.gmail.com> wrote:[vbcol=seagreen]
> That type of clustering is not supported by SQL Server 2005. Each instance
> of SQL can run on a single node at a time. Other than a read-only shared
> database model for reporting, there is no provision in SQL Server for
> multiple SQL instances runing on different nodes to access the same data
> store.
> This is the major reason I dislike the Active/Passive and Active/Active
> nomenclature. It is a holdover from SQL 7.0 clustering and does not
> accurately reflect the capabilities of the current product. And 7.0 didn't
> do this either, it just had a functional distinction between the primary
> node and its failover partner that led to this unfortunate choice of names.
> --
> Geoff N. Hiten
> Senior Database Administrator
> Microsoft SQL Server MVP
> <jreede...@.nac.net> wrote in messagenews:1170095155.783463.3200@.h3g2000cwc.goog legroups.com...
>
>
>
>
>
>
>
>
|||<jreederhi@.nac.net> wrote in message
news:1170181441.248151.311950@.j27g2000cwj.googlegr oups.com...
> Geoff: I suspect that the inability for two instances of a database
> manager to access the same physical database is common.
It is part of the shared nothing model for MSCS.
> When we speak
> of active/active, we mean that two nodes, each with its own database,
> are cooperating in the same application.
No, when you speak of active/active, you mean two nodes, each with its own
separate database in its own virtual server. Two instances of SQL, one on
each node.
> The two database copies are
> kept in synchronism by some means, typically data replication.
You are not talking about Microsoft clustering here.
Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp
The next ClusterHelp classes are:
Denver starting Feb 12th
NYC starting Feb 19th
London starting March 6th
|||<jreederhi@.nac.net> wrote in message
news:1170181441.248151.311950@.j27g2000cwj.googlegr oups.com...
> Geoff: I suspect that the inability for two instances of a database
> manager to access the same physical database is common. When we speak
> of active/active, we mean that two nodes, each with its own database,
> are cooperating in the same application. The two database copies are
> kept in synchronism by some means, typically data replication. By
> keeping replication latency small (hundreds of milliseconds with
> today's products), the problem of data loss following a node failure
> or of data collisions is minimized. Furthermore, the two nodes can be
> geographically separated for disaster tolerance. (See http://
> availabilitydigest.com/public_articles/what_is_active-active.pdf).
> There are several products that support SQL Server that can provide
> this capability. www.availabilitydigest.com
>
I disagree with this, in relation to Microsoft Cluster Service.
The PDF file referenced here is NOT describing Active/Active for Microsoft
Clustering
Please do NOT refer to this PDF in relation to MSCS
The technologies represented on your website, are not in relation to
Microsoft, and terminology you use is conflicting with standard terminology
within the Microsoft scope. I understand that you have some patented
technology to sell. Please do not use newsgroups to advertise, as this is
considered spam.
Please see Microsofts own definition on Active/Active :
http://technet2.microsoft.com/WindowsServer/en/library/8846a72b-0882-4a24-8eee-a768e52925281033.mspx?mfr=true
(scroll down to the "Q. What is active/active verses active/passive?")
Rgds,
Edwin.
