Best practice for SQL cluster and domains

Hi All,
We have a critical 24x7 SQL cluster (W2K), which is a member of an NT 4
domain. As hardware is getting old, and NT4 domain is going to disappear in
the near future, the cluster has to be re-newed. There is also a trusted
Active Directory domain, which holds about all user accounts and groups.
These accounts and groups have been assigned appropriate rights to SQL and
application generated reports.
What makes this a bit more difficult, is that the company is also going to
split, as is network and AD. The split will take place within few months,
but the new AD (where the users finally will be located) is expected to be
in place and fully functional within one year. However, the new cluster
should be up and running within two months. The cluster will be built on
Windows server 2003 Enterprise.
What I should do, is to provide best scenario for implementing new cluster,
so that it minimizes work when AD domains in question change.
As far as I am concerned, if you change a cluster domain membership, you
need to rebuild the whole cluster. This is not what we want to do. We are
prepared to re-assign all appropriate user right and roles as users' domain
changes.
I see following scenarios:
1. join new cluster to present AD domain
2. install new cluster nodes as domain controllers for new "domainlet" or
domain and create trust relationships as needed
3. install separate domain controllers, and join cluster to this domain,
create trust relationships as needed
4. something else?
In scenario 1 I see most work; rebuilding the whole cluster within a year or
so. About scenarios 2 and 3 I'd like to have comments, especially about
using domainlets
(http://www.microsoft.com/windows2000...cluster/domain
lets.asp). Or, there might be a lot better option, which I have not come to
think about.
Please share your opinions and comments,
John
Great questions.
I like option 1, have you read http://support.microsoft.com/?id=319016, no
need to rebuild the cluster and start all over. Pretty easy actually.
Have you read http://support.microsoft.com/?id=298570, so option 2 is not
looking good.
Option 3 will work, but I hate extra trusts, if I can avoid them.
Go with number 1, that is what I would do
Cheers,
Rod
MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering
http://msmvps.com/clustering - Blog
"John" <someone@.microsoft.com> wrote in message
news:41f50ef2@.usenet01.boi.hp.com...
> Hi All,
> We have a critical 24x7 SQL cluster (W2K), which is a member of an NT 4
> domain. As hardware is getting old, and NT4 domain is going to disappear
> in
> the near future, the cluster has to be re-newed. There is also a trusted
> Active Directory domain, which holds about all user accounts and groups.
> These accounts and groups have been assigned appropriate rights to SQL and
> application generated reports.
> What makes this a bit more difficult, is that the company is also going to
> split, as is network and AD. The split will take place within few months,
> but the new AD (where the users finally will be located) is expected to be
> in place and fully functional within one year. However, the new cluster
> should be up and running within two months. The cluster will be built on
> Windows server 2003 Enterprise.
> What I should do, is to provide best scenario for implementing new
> cluster,
> so that it minimizes work when AD domains in question change.
> As far as I am concerned, if you change a cluster domain membership, you
> need to rebuild the whole cluster. This is not what we want to do. We are
> prepared to re-assign all appropriate user right and roles as users'
> domain
> changes.
> I see following scenarios:
> 1. join new cluster to present AD domain
> 2. install new cluster nodes as domain controllers for new "domainlet" or
> domain and create trust relationships as needed
> 3. install separate domain controllers, and join cluster to this domain,
> create trust relationships as needed
> 4. something else?
> In scenario 1 I see most work; rebuilding the whole cluster within a year
> or
> so. About scenarios 2 and 3 I'd like to have comments, especially about
> using domainlets
> (http://www.microsoft.com/windows2000...cluster/domain
> lets.asp). Or, there might be a lot better option, which I have not come
> to
> think about.
> Please share your opinions and comments,
> John
>
|||Rod,
thanks really, this was great information. I'll investigate the options
again in the light of your recent information, the scenario 1 looks now
actually quite good. If you have something to add, please do not hesitate to
share it
Cheers, John
"Rodney R. Fournier [MVP]" <rod@.die.spam.die.nw-america.com> wrote in
message news:%23z6TlkjAFHA.3592@.TK2MSFTNGP11.phx.gbl...[vbcol=seagreen]
> Great questions.
> I like option 1, have you read http://support.microsoft.com/?id=319016, no
> need to rebuild the cluster and start all over. Pretty easy actually.
> Have you read http://support.microsoft.com/?id=298570, so option 2 is not
> looking good.
> Option 3 will work, but I hate extra trusts, if I can avoid them.
> Go with number 1, that is what I would do
> Cheers,
> Rod
> MVP - Windows Server - Clustering
> http://www.nw-america.com - Clustering
> http://msmvps.com/clustering - Blog
> "John" <someone@.microsoft.com> wrote in message
> news:41f50ef2@.usenet01.boi.hp.com...
and[vbcol=seagreen]
to[vbcol=seagreen]
months,[vbcol=seagreen]
be[vbcol=seagreen]
are[vbcol=seagreen]
or[vbcol=seagreen]
year[vbcol=seagreen]
(http://www.microsoft.com/windows2000...cluster/domain
>

Best Performance?

Assume a heavy transaction environment.
Let's say we have the same database model,
same hardware, file usage, type of data, queries etc. on two servers.
Server A is configured with a RAID 5 on all the data disks (9 of them),
and Server B is configured with RAID 5 on the data disks (5 of them) but
transaction log and tempdb is separated to its own mirrored disks (4 of
them - 2 each).
Which SQL server will run faster in terms of query response time AND write
performance? Database in Server A or Server B?Assumimng disc io is the limiting factor then I would say B would be =most likely to outperform A. But even better would be lose the raid 5 =and use 0+1 on 4 discs. or possibly on 6 and lose the separation of =tempdb. Of course this will lose you some space over the raid 5, but =will perform better. Hardware raid vs S/W will alo make a difference. =How much separating tempdb gives you totally depends on the number of =queries generating large temporary results.
Opinions are likely to vary a lot and you haven't given much info to go =on.
Mike John
"Rick" <bob@.bob.net> wrote in message =news:3f858779$0$103$8f4e7992@.newsreader.goldengate.net...
> Assume a heavy transaction environment.
> > Let's say we have the same database model,
> same hardware, file usage, type of data, queries etc. on two servers.
> Server A is configured with a RAID 5 on all the data disks (9 of =them),
> and Server B is configured with RAID 5 on the data disks (5 of them) =but
> transaction log and tempdb is separated to its own mirrored disks (4 =of
> them - 2 each).
> > Which SQL server will run faster in terms of query response time AND =write
> performance? Database in Server A or Server B?
> >|||I completely agree with Mike John's remark to drop RAID5 completely,
certainly for a 'heavy transaction environment' as you describe. Check out
this link for some reasons why
http://www.miracleas.dk/BAARF/BAARF2.html
(OK it's Oracle/Unix, but it is as tru for the SQLServer/Win world too..)
"Mike John" <Mike.John@.knowledgepool.com> wrote in message
news:uRds7fpjDHA.2676@.TK2MSFTNGP11.phx.gbl...
Assumimng disc io is the limiting factor then I would say B would be most
likely to outperform A. But even better would be lose the raid 5 and use 0+1
on 4 discs. or possibly on 6 and lose the separation of tempdb. Of course
this will lose you some space over the raid 5, but will perform better.
Hardware raid vs S/W will alo make a difference. How much separating tempdb
gives you totally depends on the number of queries generating large
temporary results.
Opinions are likely to vary a lot and you haven't given much info to go on.
Mike John
"Rick" <bob@.bob.net> wrote in message
news:3f858779$0$103$8f4e7992@.newsreader.goldengate.net...
> Assume a heavy transaction environment.
> Let's say we have the same database model,
> same hardware, file usage, type of data, queries etc. on two servers.
> Server A is configured with a RAID 5 on all the data disks (9 of them),
> and Server B is configured with RAID 5 on the data disks (5 of them) but
> transaction log and tempdb is separated to its own mirrored disks (4 of
> them - 2 each).
> Which SQL server will run faster in terms of query response time AND write
> performance? Database in Server A or Server B?
>|||On Thu, 9 Oct 2003 11:06:17 -0500, "Rick" <bob@.bob.net> wrote:
>Assume a heavy transaction environment.
> Let's say we have the same database model,
>same hardware, file usage, type of data, queries etc. on two servers.
>Server A is configured with a RAID 5 on all the data disks (9 of them),
>and Server B is configured with RAID 5 on the data disks (5 of them) but
>transaction log and tempdb is separated to its own mirrored disks (4 of
>them - 2 each).
>Which SQL server will run faster in terms of query response time AND write
>performance? Database in Server A or Server B?
Well, like the other guys said, one would probably anticipate B
running a bit faster, but it's also going to depend on how much data
is cached, that is, on RAM size, and just how heavily tempdb gets
used, and for that matter, how heavily the log gets used. If it were
100% simple queries, A might even win, for a more usual mix of 90%
simple queries, 5% complex queries, and 5% highly localized updates,
you're probably close to the break-even point, assuming you're not
running short of disk IO bandwidth or CPU in either case A or case B.
J.