Home | Contact Us | FAQ | Search & Site Map | Link to Us
Sign In | Join | Other 45 Sites in Network
Home
Discussion Groups
DB Engine
SQL ServerMSDESQL Server CE
Services
Analysis (Data Mining)Analysis (OLAP)DTSIntegration ServicesNotification ServicesReporting Services
Programming
CLRConnectivitySQLXML
Other Technologies
ClusteringEnglish QueryFull-Text SearchReplicationService Broker
General
Data WarehousingPerformanceSecuritySetupSQL Server ToolsOther SQL Server Topics
DirectoryUser Groups
Related Topics
MS AccessOther DB ProductsMS Server Products.NET DevelopmentVB DevelopmentJava DevelopmentMore Topics ...

SQL Server Forum / Other Technologies / Clustering / November 2006

Tip: Looking for answers? Try searching our database.

Looking for suggestions on specs

Thread view: 
Enable EMail Alerts  Start New Thread
Thread rating: 
Jake Smythe - 16 Nov 2006 00:37 GMT
Hello,

   We are wanting to setup a sql2k5 std cluster and are curious if we have
good specs drawn out.

SAN has 3 arrays

SQL Group has the following resources:
Network name
IP Address
physical disks for data (RAID 5)
physical disk for logs (RAID 1)

Cluster Group has the following resources:
Network Name
IP Address
physical disk for quorum (RAID 1)
MS DTC resource

Is this a valid setup or a better suggestion? Thanks in advance.

Jake
Tom Moreau - 16 Nov 2006 00:42 GMT
If you can avoid RAID5 and go with RAID10, that would be an improvement.  As
for logs, if you have just one DB, then RAID1 is likely sufficient.
However, if you have many DB's with update activity, consider putting the
logs on RAID10.

Signature

  Tom

----------------------------------------------------
Thomas A. Moreau, BSc, PhD, MCSE, MCDBA, MCITP, MCTS
SQL Server MVP
Toronto, ON   Canada
.

Hello,

   We are wanting to setup a sql2k5 std cluster and are curious if we have
good specs drawn out.

SAN has 3 arrays

SQL Group has the following resources:
Network name
IP Address
physical disks for data (RAID 5)
physical disk for logs (RAID 1)

Cluster Group has the following resources:
Network Name
IP Address
physical disk for quorum (RAID 1)
MS DTC resource

Is this a valid setup or a better suggestion? Thanks in advance.

Jake
Jake Smythe - 16 Nov 2006 00:48 GMT
Tom,

   Thanks for the reply. I was curious if I have my cluster groups setup
correctly. Is it ok to have the quorum and the MS DTC within the same group
that share the same physical disk?

Jake

> If you can avoid RAID5 and go with RAID10, that would be an improvement.
> As
[quoted text clipped - 24 lines]
>
> Jake
Tom Moreau - 16 Nov 2006 00:57 GMT
It depends.  ;-)  If you use a lot of distributed transactions, consider
putting the DTC on its own disk.  If you use little or no distributed
transactions, then you can put it on the same disk as the quorum.

Signature

  Tom

----------------------------------------------------
Thomas A. Moreau, BSc, PhD, MCSE, MCDBA, MCITP, MCTS
SQL Server MVP
Toronto, ON   Canada
.

Tom,

   Thanks for the reply. I was curious if I have my cluster groups setup
correctly. Is it ok to have the quorum and the MS DTC within the same group
that share the same physical disk?

Jake

> If you can avoid RAID5 and go with RAID10, that would be an improvement.
> As
[quoted text clipped - 24 lines]
>
> Jake
Edwin vMierlo - 16 Nov 2006 10:29 GMT
As a best practice I never put DTC in the Cluster Group (and therefore not
on the quorum disk), my logic is that when you cluster, you want to omit as
much as possible anything which can terminate the cluster.

Example:
if for some odd reason the DTC fails, this immediately prompt the Cluster
Group to move nodes. If again for some odd reason the DTC fails to com
online on the other node, you might run the risk of loosing your cluster
completely.

In my opinion I would leave the cluster group the cluster group, and create
a seperate group with an IP, NetName, Disk(small) and DTC. This can then
move independently from your cluster group, and in case the DTC has
problems, you are able to work and troubleshoot this without effecting
anything in the cluster group.

As you are in a SAN, it should be possible to present a small disk to this
cluster for DTC purposes.

... my two cents ...
Rgds,
Edwin.

> It depends.  ;-)  If you use a lot of distributed transactions, consider
> putting the DTC on its own disk.  If you use little or no distributed
[quoted text clipped - 36 lines]
> >
> > Jake
Jake Smythe - 16 Nov 2006 15:20 GMT
Guys,

   Thanks for the input I truly appreciate it. One last question, hopefully
:), in the documentation I see that the quorum can be a relatively small
size (something like 50 MB) what is your recommendation? I am assuming you
guys have had a box in production so you'd know better than the docs I am
reading. Thanks again for the input.

Jake

> As a best practice I never put DTC in the Cluster Group (and therefore not
> on the quorum disk), my logic is that when you cluster, you want to omit
[quoted text clipped - 65 lines]
>> >
>> > Jake
Edwin vMierlo - 16 Nov 2006 15:58 GMT
Jake,

Microsoft still (since a long time) recommends 500MB for its quorum, see KB
280345

That recommendation might be larger than actually needed, however with 500MB
I have never seen any problems !
I have alway deployed a quorum disk of 500MB minimum.

Rgds,
Edwin.

> Guys,
>
[quoted text clipped - 75 lines]
> >> >
> >> > Jake
Jake Smythe - 16 Nov 2006 16:03 GMT
Edwin,

   Thanks.

Jake
> Jake,
>
[quoted text clipped - 101 lines]
>> >> >
>> >> > Jake
Russ Kaufmann - 16 Nov 2006 17:25 GMT
> Guys,
>
[quoted text clipped - 3 lines]
> I am assuming you guys have had a box in production so you'd know better
> than the docs I am reading. Thanks again for the input.

500MB is recommended. The reasoning is that 500 MB is the size when NTFS
becomes more efficient.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Rodney R. Fournier [MVP] - 16 Nov 2006 19:09 GMT
Though I use 1 GB and Geoff does 2 GB (even math or maths for those in the
UK).

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner

>> Guys,
>>
[quoted text clipped - 6 lines]
> 500MB is recommended. The reasoning is that 500 MB is the size when NTFS
> becomes more efficient.
Russ Kaufmann - 16 Nov 2006 15:24 GMT
> As a best practice I never put DTC in the Cluster Group (and therefore not
> on the quorum disk), my logic is that when you cluster, you want to omit
> as
> much as possible anything which can terminate the cluster.

I agree with this to a point. However, the cost of the disk for the MSDTC is
pretty high when you consider other applications in larger clusters that can
cause you to run out of drive letters.

Also, it is wasted resources in my opinion if it isn't ever used.

Since the Cluster team has extensively tested MSDTC in the default cluster
group and have given it their blessing, I don't see any need for it having
its own resources. Except, when the application uses it heavily.

> Example:
> if for some odd reason the DTC fails, this immediately prompt the Cluster
> Group to move nodes. If again for some odd reason the DTC fails to com
> online on the other node, you might run the risk of loosing your cluster
> completely.

Set the Affect Group to disabled for the MSDTC resource and that concern
goes away.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Edwin vMierlo - 16 Nov 2006 16:10 GMT
Russ,

I have considered all that before I posted, and must say you are absolutely
right, and KB 301600 details the microsft view on this.
Where on one hand they say "leave it in the cluster group" and in the same
article they say "create a sepereate group".

For drive letters, and the "cost" of a small disk, I think each should look
at their own config of SAN and cluster to determine if an additional disk is
feasible.
Then again... if you likely to run out of drive-letters, you might be
pushing a solution with mountpoints already.

In regards to disabling the "affect group" property, you should always do
that if you add the MSDTC into cluster group, however I have seen people
forget this, or even admins who noticed this disabled and enabled it,
because they did not understand the purpose of disabling.

in my opinion, a seperate group for MSDTC will never jeaporadise resources
in the cluster group, while MSDTC in the cluster group might (in some
circumstances)... its about minimising risk... and uptime !

Rgds,
Edwin.

> > As a best practice I never put DTC in the Cluster Group (and therefore not
> > on the quorum disk), my logic is that when you cluster, you want to omit
[quoted text clipped - 19 lines]
> Set the Affect Group to disabled for the MSDTC resource and that concern
> goes away.
Russ Kaufmann - 16 Nov 2006 17:28 GMT
> in my opinion, a seperate group for MSDTC will never jeaporadise resources
> in the cluster group, while MSDTC in the cluster group might (in some
> circumstances)... its about minimising risk... and uptime !

So, tell me, what are those risks that you are trying to mitigate? I am not
aware of circumstances (other than performance reasons) that would encourage
it.

I would love to hear more.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Anthony Thomas - 18 Nov 2006 18:01 GMT
The problem is that, for SQL Server, if MS DTC does not have a dedicated
network name, and fails over for any reason, if the Virtual DTC shares the
quorum group, it MUST RUN ON THE SAME NODE as the SQL Server resource group,
or SQL Server cannot connect to it.

In a dedicated resource group, DTC is always available to SQL Server.

Now, as a better alternative, if you are only running a single-instanced SQL
Server installation, then allow the DTC resource to be assigned to the SQL
Server resource group.  Moreover, make the SQL Server resource depend on the
DTC resource.  This way, SQL Server will always be able to acquire the DTC
handle on startup.

If you are running multiple SQL Server instances, then a dedicated DTC group
is the only way to ensure each SS instance will be able to connect.

Sincerely,

Anthony Thomas

> > in my opinion, a seperate group for MSDTC will never jeaporadise resources
> > in the cluster group, while MSDTC in the cluster group might (in some
[quoted text clipped - 5 lines]
>
> I would love to hear more.
Russ Kaufmann [MVP] - 20 Nov 2006 17:14 GMT
> The problem is that, for SQL Server, if MS DTC does not have a dedicated
> network name, and fails over for any reason, if the Virtual DTC shares the
> quorum group, it MUST RUN ON THE SAME NODE as the SQL Server resource
> group,
> or SQL Server cannot connect to it.

It should be set to not affect the group.

> In a dedicated resource group, DTC is always available to SQL Server.

It is always available as well in the default cluster group.

> Now, as a better alternative, if you are only running a single-instanced
> SQL
[quoted text clipped - 3 lines]
> DTC resource.  This way, SQL Server will always be able to acquire the DTC
> handle on startup.

It will always be able to access it whether it is on the same node or not.
The D is for Distributed, not Dedicated.

> If you are running multiple SQL Server instances, then a dedicated DTC
> group
> is the only way to ensure each SS instance will be able to connect.

Citation?

You can only have one MSDTC per cluster, and you can have up to 16 instances
per cluster. Are you saying that all of the 16 instances will not work
properly unless they are all on the same node as the MSDTC? That would be
incorrect.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Anthony Thomas - 21 Nov 2006 04:48 GMT
There's the way it "should be" and the way "it is."

The original KB article used to suggest you do exactly as you propose: to
assign the DTC resource to the quorum drive in the cluster resource group.

However, along the way, DTC was modified, and the KB was updated to suggest
that DTC needed to be in its own resource group.

The fact is that in single-instanced and multi-instanced configurations, we
constantly had bind errors for one instance or another until we followed the
new KB recommendations and created dedicated resource groups for it.

You would have to ask the DTC developers what changed to cause this to
occur, but it has, nevertheless.

I do agree, however, that unless distributed queries are being executed,
there should not be any need to create a dedicated resource group...but
then, why not just run DTC on each clustered node, unclustered?  If you are
not going to follow the recommendation, why bother clustering the resource
at all?

Sincerely,

Anthony Thomas

> > The problem is that, for SQL Server, if MS DTC does not have a dedicated
> > network name, and fails over for any reason, if the Virtual DTC shares the
[quoted text clipped - 29 lines]
> properly unless they are all on the same node as the MSDTC? That would be
> incorrect.
Russ Kaufmann [MVP] - 21 Nov 2006 15:37 GMT
> The fact is that in single-instanced and multi-instanced configurations,
> we
> constantly had bind errors for one instance or another until we followed
> the
> new KB recommendations and created dedicated resource groups for it.

The latest KB recommendation is MSDTC in the default cluster group.

> You would have to ask the DTC developers what changed to cause this to
> occur, but it has, nevertheless.

I think there is something else in your configuration that might have
issues. I have never heard of a single issue with MSDTC in the default
cluster group.

> I do agree, however, that unless distributed queries are being executed,
> there should not be any need to create a dedicated resource group...but
> then, why not just run DTC on each clustered node, unclustered?

Because it doesn't work that way.

> If you are
> not going to follow the recommendation, why bother clustering the resource
> at all?

The recommendation is MSDTC in the default cluster group.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Anthony Thomas - 22 Nov 2006 05:34 GMT
I respectfully disagree.  And, as the MSDN article that Tom provided
suggests, for a single-instanced SQL Server, leaving the MSDTC resource in
the "SQL Server resource group" is viable.

The most recent article, however, has striped the Cluster Resource group
"alternative" and has made compelling arguments for a dedicated resource
group.  Article dated November 1, 2006, revision 20.4.

Where do you get your information?

How to configure Microsoft Distributed Transaction Coordinator on a Windows
Server 2003 cluster
http://support.microsoft.com/kb/301600/en-us

Note When creating the MSDTC, moving the resource group into a group other
than SQL Server or Exchange Server group is highly recommended. Creating the
MSDTC resource in its own resource group and assigning it to a separate
cluster group keeps the resource highly available.

Note If you create the resource in the same cluster group, and if Resource A
has failed over, Resource B will also failover. When you create the resource
in a different cluster group, even when Resource A has failed over, the
resource B does not failover. Creating Cluster Resources in the different
cluster groups decreases unnecessary failovers of the Cluster Resources.

Important After the new group for MSDTC is created you must create the
Network Name, IP address and Physical Disk for Msdtc. You may already have
the physical disk resource that can be moved into the new group. After the
MSDTC Group is created you must create the IP address resource, Network Name
resource and the Physical disk resources. You may already have a physical
disk resource that must be moved into the new MSDTC group.

Sincerely,

Anthony Thomas

> > The fact is that in single-instanced and multi-instanced configurations,
> > we
[quoted text clipped - 22 lines]
>
> The recommendation is MSDTC in the default cluster group.
Russ Kaufmann [MVP] - 22 Nov 2006 16:00 GMT
>I respectfully disagree.  And, as the MSDN article that Tom provided
> suggests, for a single-instanced SQL Server, leaving the MSDTC resource in
> the "SQL Server resource group" is viable.

I really don't see it as a major issue. I wouldn't recommend it, but I
wouldn't scoff at it either. If you are happy and it works, that is good
news.

Have a great weekend! I can't wait for mine to start in 7 hours. <G>

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Jake Smythe - 20 Nov 2006 18:54 GMT
Hello,

   Ok now I am little confused. :) For best practices it seems that some
are saying the DTC should be on its own cluster correct? In our
configuration we are using two nodes and a san with three arrays. One for
the logs, one for the data and one for the quorum and DTC (as it is
configured now). Since we rarely use DTC is this a best practice scenario or
should it be configured another way? If this is good what happens later? Can
we move the DTC to another array should we need to? Thanks in advance.

Jake

> The problem is that, for SQL Server, if MS DTC does not have a dedicated
> network name, and fails over for any reason, if the Virtual DTC shares the
[quoted text clipped - 32 lines]
>>
>> I would love to hear more.
Russ Kaufmann [MVP] - 20 Nov 2006 21:12 GMT
> Hello,
>
[quoted text clipped - 6 lines]
> later? Can we move the DTC to another array should we need to? Thanks in
> advance.

The MSDTC is not used in the vast majority of implementations. Unless you
are using two phase commits in SQL, you will not use the MSDTC. Since you
don't use it, you can put it in the default cluster group and save resources
(i.e. IP Address and Disk resources) that you may need in other cluster
groups.

See http://msmvps.com/blogs/clusterhelp/archive/2005/08/05/61718.aspx for
more detail.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Edwin vMierlo - 21 Nov 2006 09:51 GMT
Russ,

I am not questioning your advice, who am I to do that, however please
imaging the following:

1) an Admin person not disabling the "affect group" property ... that might
affect the cluster (when MSDTC fails)
2) an Admin person enabling the "affect group" property ... that might
affect the cluster (when MSDTC fails)

See, not everyone in the world run a "perfect" cluster, and there will be
people with less experience then you ( ! that is an understatement ! )...
hence "small mistakes" will be made. We are not perfect, we all do make
small mistakes, especially if you are looking at new technologies, or
technologies you have not implemented before, it is about minimising these
risks.

When MSDTC is in its own group, it will never affect the cluster resources
in the cluster group, I hope you agree on this ?

> > in my opinion, a seperate group for MSDTC will never jeaporadise resources
> > in the cluster group, while MSDTC in the cluster group might (in some
[quoted text clipped - 5 lines]
>
> I would love to hear more.
Edwin vMierlo - 21 Nov 2006 11:10 GMT
Funny, right now I am working on a cluster issue where the admin has the
"affect group" property enabled...

... it does exists... I guess this is the difference between "it should not"
and "what is out there"

rgds,
Edwin.

> Russ,
>
[quoted text clipped - 28 lines]
> >
> > I would love to hear more.
Tom Moreau - 21 Nov 2006 12:03 GMT
This article has some useful bits:

http://msdn.microsoft.com/library/default.asp?url=/library/en-us/cossdk/html/cde
18bcb-ce2b-4d6f-9ada-baf658a7f87f.asp


Signature

  Tom

----------------------------------------------------
Thomas A. Moreau, BSc, PhD, MCSE, MCDBA, MCITP, MCTS
SQL Server MVP
Toronto, ON   Canada
.

Funny, right now I am working on a cluster issue where the admin has the
"affect group" property enabled...

... it does exists... I guess this is the difference between "it should not"
and "what is out there"

rgds,
Edwin.

"Edwin vMierlo" <EdwinvMierlo@discussions.microsoft.com> wrote in message
news:erlIGQVDHHA.1748@TK2MSFTNGP02.phx.gbl...
> Russ,
>
> I am not questioning your advice, who am I to do that, however please
> imaging the following:
>
> 1) an Admin person not disabling the "affect group" property ... that
might
> affect the cluster (when MSDTC fails)
> 2) an Admin person enabling the "affect group" property ... that might
[quoted text clipped - 11 lines]
>
> > "Edwin vMierlo" <EdwinvMierlo@discussions.microsoft.com> wrote in
message
> > news:unUvxmZCHHA.5012@TK2MSFTNGP06.phx.gbl...
> >
[quoted text clipped - 10 lines]
> >
> > I would love to hear more.
Russ Kaufmann [MVP] - 21 Nov 2006 15:42 GMT
> Funny, right now I am working on a cluster issue where the admin has the
> "affect group" property enabled...
>
> ... it does exists... I guess this is the difference between "it should
> not"
> and "what is out there"

There are a bazillion examples of people incorrectly implementing
technology. Are you going to try to configure your systems so that you can
have all the errors you want?

High availability is much more than putting together a cluster. High
availability requires proper training, proper experience, well documented
processes, and well documented DR plans among the many requirements.

To me, it sounds like you are making excuses for the technology, which isn't
the point at all.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Russ Kaufmann [MVP] - 21 Nov 2006 15:40 GMT
> Russ,
>
[quoted text clipped - 9 lines]
> See, not everyone in the world run a "perfect" cluster, and there will be
> people with less experience then you ( ! that is an understatement ! )...

People make mistakes all the time, which is why we offer training and which
is why there is so much documentation out there regarding implemention of
clustering and the various properties.

> hence "small mistakes" will be made. We are not perfect, we all do make
> small mistakes, especially if you are looking at new technologies, or
> technologies you have not implemented before, it is about minimising these
> risks.

Even so, you make it sound like the MSDTC will fail on a regular basis. That
is not correct. If properly installed (and that is not difficult), it will
run just fine.

> When MSDTC is in its own group, it will never affect the cluster resources
> in the cluster group, I hope you agree on this ?

If you do it right, it won't be a problem putting it into the default
cluster group.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Edwin vMierlo - 21 Nov 2006 18:18 GMT
Russ,

you are right, lets stop this conversation, lets stick to KB 301600
http://support.microsoft.com/kb/301600

In this article Microsoft states:

"We recommend that you create the MSDTC resource in its own resource group,
such as you would for Microsoft SQL Server, where you can do so"

and if this is not an option then Microsoft tells us in this article:

"If you cannot do this, or if you have a light-use DTC scenario such as a
cluster dedicated to Exchange Server, proceed and use the Cluster group"

Actually, while reading this again, I suppose the truth is in the middle,
for Exchange clusters --> cluster group, however the recommendation is still
its own resource group where you can do so.

Sorry for all the confusion, it was not my intention to have such a long
discussion on this, I guess just follow the Microsoft recommendation is the
route to go.

rgds,
Edwin.

> > Russ,
> >
[quoted text clipped - 28 lines]
> If you do it right, it won't be a problem putting it into the default
> cluster group.
Russ Kaufmann [MVP] - 21 Nov 2006 19:32 GMT
> you are right, lets stop this conversation, lets stick to KB 301600
> http://support.microsoft.com/kb/301600

Who said anything about stopping this conversation. You are entitled to an
opinion.

The facts are simple, though, and in large clusters, drive letters are a
major constraint. Wasting one on a resource that doesn't get used (except in
rare cases) is not efficient. Even if your cluster starts off small, it may
grow, and part of HA is having an eye towards the future.

Let's use a simple analogy. You own an SUV. It has two rows of seats. You
have two kids and a wife. So, you need four seats for four passengers. Once
or twice a year, your dog (a small pomeranian) rides with you to the vet.
Are you going to buy a third row of seats for that extremely rare occassion
or are you going to have somebody hold him in their lap?

> "If you cannot do this, or if you have a light-use DTC scenario such as a
> cluster dedicated to Exchange Server, proceed and use the Cluster group"

Absolutely. If you have a light use scenario (which is almost always the
case), then you should use the default cluster group and not waste resources
for the MSDTC. That is Microsoft's recommendation as you pointed out. Do you
want to discuss "language" in the article now?

SQL is also a light use scenario unless using two phase commits. In those
cases, you should do as stated above and use the default cluster group. The
only time you should dedicate resources to the MSDTC is IF there is a
performance requirement like heavy use of two phase commits or, in the case
of Exchange, heavy use of work flows.

The horse is dead, Edwin. No sense beating it anymore.

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Geoff N. Hiten - 21 Nov 2006 22:12 GMT
Just to play devil's advocate, there are some SQL scenarios that make heavy
use of DTC without using linked server queries or two-phase commits.  I
refer to ODBC transactions.  Since ODBC doesn't have pass-through
transaction support (remember, it was intended to abstract Jet/Access) it
uses the Microsoft Transaction Server interfaces, which are implemented via
DTC on a cluster.  So, if you have lots of ODBC clients, you can
inadvertantly have a high DTC throughput without meaning to.

Having said that, the original SAN problem that caused the KB article change
only happens in rare cases on certain hardware platforms, but to retain
hardware neutrality, MS made it a general recommendation.  They later backed
down and put the "light duty" words in the article.

And SQL 2005 supports mount points so disk drive letters are no longer sold
by the each and thus in limited supply

Play nice.

Signature

Geoff N. Hiten
Senior Database Administrator
Microsoft SQL Server MVP

>> you are right, lets stop this conversation, lets stick to KB 301600
>> http://support.microsoft.com/kb/301600
[quoted text clipped - 28 lines]
>
> The horse is dead, Edwin. No sense beating it anymore.
Russ Kaufmann [MVP] - 21 Nov 2006 22:44 GMT
> Just to play devil's advocate, there are some SQL scenarios that make
> heavy use of DTC without using linked server queries or two-phase commits.
[quoted text clipped - 3 lines]
> via DTC on a cluster.  So, if you have lots of ODBC clients, you can
> inadvertantly have a high DTC throughput without meaning to.

True...

> Having said that, the original SAN problem that caused the KB article
> change only happens in rare cases on certain hardware platforms, but to
> retain hardware neutrality, MS made it a general recommendation.  They
> later backed down and put the "light duty" words in the article.

Performance reasons have been fully acknowledged as an extremely good reason
to put MSDTC in its own group with its own resources.

> And SQL 2005 supports mount points so disk drive letters are no longer
> sold by the each and thus in limited supply

Mount points, in Exchange at least, have been known  to be troublesome. Is
the same true of SQL?

From my experience, I just don't see mount points being used for SQL 2005.
Perhaps it is lack of knowledge or just fear of the unknown. <G>

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Anthony Thomas - 22 Nov 2006 05:42 GMT
As an alternative to mount points (yes, they can be troublesome, even for
SQL Server), the use of a complete logical drive manager utility (note
Veritas Software Foundation, the same as the LDM "lite" licensed by
Microsoft) supports a better implementation for Volume Management, for
stand-alones as well as clustered configurations.

Anthony Thomas

> > Just to play devil's advocate, there are some SQL scenarios that make
> > heavy use of DTC without using linked server queries or two-phase commits.
[quoted text clipped - 22 lines]
> From my experience, I just don't see mount points being used for SQL 2005.
> Perhaps it is lack of knowledge or just fear of the unknown. <G>
Edwin vMierlo - 22 Nov 2006 09:16 GMT
Russ,

not sure what horse died on you, I think I do not appreciating that remark :

To follow my logic, in my opinion people should follow Microsofts
Recommendation

"We recommend that you create the MSDTC resource in its own resource group,
such as you would for Microsoft SQL Server, where you can do so."

Clearly detailed in KB-301600
I am aware that this article has changed a bit over time and probably will
change over time, but it is clearly documented and Microsofts recommendation
at this moment.

I understand that there are reasons not to follow this, and I understand
that lite use of DTC will perfectly run in harmony with cluster resources in
the cluster group. As you said before "when done right", this will cause no
problems. I am fully aware that microsoft and its cluster team has fully
tested such scenarios, and that it is what they call "supportable". And I
have come accross many clusters with the DTC in the cluster group, without
any problems.

This, however, is not the point. The point is to give advice and help to the
community, which should be based on Microsofts recommendation.

For advice to the outside world, I, for one, will stick with KB-301600 for
the moment:
1) Microsoft Recommendation : create seperate group, where you can
2) lite usage : allowed in cluster group
3) dedicated Exchange cluster : should be in cluster group

Rgds,
Edwin.

> > you are right, lets stop this conversation, lets stick to KB 301600
> > http://support.microsoft.com/kb/301600
[quoted text clipped - 28 lines]
>
> The horse is dead, Edwin. No sense beating it anymore.
Russ Kaufmann [MVP] - 22 Nov 2006 16:04 GMT
> not sure what horse died on you, I think I do not appreciating that remark

Sorry, I am not going to get in a flame war. I have much better things to
do.

MSDTC is really simple. Yes, you can put it in its own resouce group. Yes,
that is a good thing to do if it heavily utilized. Yes, you can and should
put it in the default cluster group if it isn't being utilized. Yes, by
putting it in the default cluster group, you can save precious drive letters
and it is really important when there is limited disk space like in many
SCSI attached arrays.

The problem that I see, and it is often, is that people don't take the
long-term view that they should and consider each decision and its full
impact on the cluster today and in three years.

Have a great holiday, Edwin (assuming you are in US).

Signature

Russ Kaufmann
MVP - Windows Server - Clustering
ClusterHelp.com, a Microsoft Certified Gold Partner
Web http://www.clusterhelp.com
Blog http://msmvps.com/clusterhelp

Edwin vMierlo - 22 Nov 2006 16:18 GMT
thanks, (not in US though)

> > not sure what horse died on you, I think I do not appreciating that remark
>
[quoted text clipped - 13 lines]
>
> Have a great holiday, Edwin (assuming you are in US).
Geoff N. Hiten - 22 Nov 2006 17:18 GMT
One slight disagreement here.

Often times personal experience can be at odds with Microsoft
recommendations.  Sharing both points of view is essential, otherwise we
would just be extensions of Microsoft PSS/CSS.

Community <> Microsoft

Signature

Geoff N. Hiten
Senior Database Administrator
Microsoft SQL Server MVP

> Russ,
>
[quoted text clipped - 79 lines]
>>
>> The horse is dead, Edwin. No sense beating it anymore.
Edwin vMierlo - 23 Nov 2006 08:43 GMT
point taken !

> One slight disagreement here.
>
[quoted text clipped - 87 lines]
> >>
> >> The horse is dead, Edwin. No sense beating it anymore.
 
Sign In
Join
My Latest Posts
My Monitored Threads
My Blog
My Photo Gallery
My Profile
My Homepage

Start New Thread
Enable EMail Alerts
Rate this Thread



©2009 Advenet LLC   Privacy Policy - Terms of Use
This website includes both content owned or controlled by Advenet as well as content owned or controlled by third parties.