|
|
|
date: Thu, 15 Feb 2007 02:33:26 +0100,
group: microsoft.public.exchange.design
back
Re: 99.9 service availability
In the CCR model, you'd use log shipping to send the logs to the passive
node where they are replayed. When a log file fills to 1M, it is shipped,
validated, and replayed. That may be every few minutes or less on a very
active SG, or a lot longer if the SG is not so active (yes, there is the
dumpster on the hub, but there are issues as to placement as well). MS
recommends VSS backups on the Seconday node due to the IO intensity of copy
on write snapshots and the IO activity associated with validating the backup
(per kb822896), so backups potentially lag as well. There's clearly a gap
here depending on how busy the storage group is. If it's right for you
really depends on what your SLA is. At a minimum, with Netapp you get space
utilzation similar to RAID 5 with no write penalty. You also get to
leverage the low IO impact of snapshots on the platform and prerform them on
both the active and inactive nodes at least eliminating this part of the
gap.
In the designs Mark refers to, you would you the good old shared storage
cluster and snapmirror replication to a DR site using standby clusters.
Locally, The RTO is as long as it take the cluster to failover (a couple of
minutes) and the RTO is up to the minute. In the event of site failure, the
RPO can be as low as 5 minutes (it's dependent of the frequency of log
snapmirror updates) or so and the RTO is a few hours (because you implement
a standby cluster). The point with the RPO in this case is that it is
expressed in terms of time - the way business objectives are expressed.
This is very similar to designs for Exchange 2003 out there and proven in
the real world today. It falls within the support policy for replication of
Exchange data because snapshots, not live data, are what is replicated.
http://www.netapp.com/go/techontap/matl/three-tradeoffs.html
http://www.microsoft.com/technet/technetmag/issues/2006/10/FailoverClusters/default.aspx
John
"Jesus Martin" wrote in message
news:ev4I%23KVUHHA.5060@TK2MSFTNGP06.phx.gbl...
>
> Mark, I don't know so much about SnapMirror but my as far as I know NetApp
> SnapMirror provides Data Availability copying the DB and logs to a
> secondary location using a cheap storage (if you want of course) what I
> don't see here is the service availability support. CCR enables you to use
> a passive node to support your users when the active node is down. Using
> the NetApp solution you should create a new storage group pointing to the
> replicated DB, this process is not immediate so the service can be down
> for a while
>
> What would you recommend here?
>
> thanks
>
> "Mark Arnold [MVP]" escribi en el mensaje
> news:7o49t29q9ds9k4gng4k93n504ajbs7p53s@4ax.com...
>> On Thu, 15 Feb 2007 18:02:34 +0100, "Jesus Martin"
>> wrote:
>>
>>>They are talking about using NetApp Snapmirror for Exchange
>>>
>>>thanks
>>>
>> If you have SnapMirror then you can put away all thoughts of LCR and
>> CCR as you just do not need it.
>> Sit down with your storage admin or call the people who sold you the
>> Filer and talk to them about Exchange 2007 configuration. They may not
>> be up to speed on it it's the same as 2003 so they will be able to
>> help you fully.
>>
>> Take a look at: http://www.netapp.com/library/tr/3407.pdf which will
>> help you understand SnapMirror if you're not already familiar with the
>> product.
>
>
date: Thu, 15 Feb 2007 16:58:01 -0800
author: John Fullbright [MVP] fjohn@donotspamnetappdotcom
Re: 99.9 service availability
Thanks John, They have 1000 mailboxes in each Datacenter so, to achieve what
you say below, I see ...
they would need 2 servers in site 1 working in a SCC model replicating the
data using snapmirror replication to other 2 nodes stand by cluster and
other 2 servers in site 2 with the mailboxes from site 2 replicating the
data to site A using snapmirror to other 2 nodes stand by cluster
this ends up with 8 servers and 8 enterprise editions of Exchange + all the
other server roles what seems to be too much for them, They want to minimise
the number of servers used but keeping in mind high availability and
contingency topics as a main goal, So I would see couple of chances here:
Using CCR ( 4 servers)
Site 1 - 1 Active Node with 1000 mailboxes from Site 1 and 1 passive node
with the 1000 mailboxes from the Site 2 replicating the data between DC
using snapmirror
Site 2 - 1 Active Node with 1000 mailboxes from Site 2 and 1 passive node
with the 1000 mailboxes from the Site 1 doing the same than above
or
Using 1 SCC (8 servers) in each location and replicate the data to the other
datacenter using netapp solution
what do you think?
thanks
Jesus
"John Fullbright [MVP]" <fjohn@donotspamnetappdotcom> escribi en el mensaje
news:uoV1FWWUHHA.2256@TK2MSFTNGP02.phx.gbl...
> In the CCR model, you'd use log shipping to send the logs to the passive
> node where they are replayed. When a log file fills to 1M, it is shipped,
> validated, and replayed. That may be every few minutes or less on a very
> active SG, or a lot longer if the SG is not so active (yes, there is the
> dumpster on the hub, but there are issues as to placement as well). MS
> recommends VSS backups on the Seconday node due to the IO intensity of
> copy on write snapshots and the IO activity associated with validating the
> backup (per kb822896), so backups potentially lag as well. There's
> clearly a gap here depending on how busy the storage group is. If it's
> right for you really depends on what your SLA is. At a minimum, with
> Netapp you get space utilzation similar to RAID 5 with no write penalty.
> You also get to leverage the low IO impact of snapshots on the platform
> and prerform them on both the active and inactive nodes at least
> eliminating this part of the gap.
>
> In the designs Mark refers to, you would you the good old shared storage
> cluster and snapmirror replication to a DR site using standby clusters.
> Locally, The RTO is as long as it take the cluster to failover (a couple
> of minutes) and the RTO is up to the minute. In the event of site
> failure, the RPO can be as low as 5 minutes (it's dependent of the
> frequency of log snapmirror updates) or so and the RTO is a few hours
> (because you implement a standby cluster). The point with the RPO in this
> case is that it is expressed in terms of time - the way business
> objectives are expressed. This is very similar to designs for Exchange
> 2003 out there and proven in the real world today. It falls within the
> support policy for replication of Exchange data because snapshots, not
> live data, are what is replicated.
>
> http://www.netapp.com/go/techontap/matl/three-tradeoffs.html
> http://www.microsoft.com/technet/technetmag/issues/2006/10/FailoverClusters/default.aspx
>
>
> John
>
> "Jesus Martin" wrote in message
> news:ev4I%23KVUHHA.5060@TK2MSFTNGP06.phx.gbl...
>>
>> Mark, I don't know so much about SnapMirror but my as far as I know
>> NetApp SnapMirror provides Data Availability copying the DB and logs to a
>> secondary location using a cheap storage (if you want of course) what I
>> don't see here is the service availability support. CCR enables you to
>> use a passive node to support your users when the active node is down.
>> Using the NetApp solution you should create a new storage group pointing
>> to the replicated DB, this process is not immediate so the service can be
>> down for a while
>>
>> What would you recommend here?
>>
>> thanks
>>
>> "Mark Arnold [MVP]" escribi en el mensaje
>> news:7o49t29q9ds9k4gng4k93n504ajbs7p53s@4ax.com...
>>> On Thu, 15 Feb 2007 18:02:34 +0100, "Jesus Martin"
>>> wrote:
>>>
>>>>They are talking about using NetApp Snapmirror for Exchange
>>>>
>>>>thanks
>>>>
>>> If you have SnapMirror then you can put away all thoughts of LCR and
>>> CCR as you just do not need it.
>>> Sit down with your storage admin or call the people who sold you the
>>> Filer and talk to them about Exchange 2007 configuration. They may not
>>> be up to speed on it it's the same as 2003 so they will be able to
>>> help you fully.
>>>
>>> Take a look at: http://www.netapp.com/library/tr/3407.pdf which will
>>> help you understand SnapMirror if you're not already familiar with the
>>> product.
>>
>>
>
>
date: Fri, 16 Feb 2007 04:01:47 +0100
author: Jesus Martin
Re: 99.9 service availability
In the SCC model, if the source is A/P, then the standby cluster can be a
single node. This will save a couple of hosts.
The CCR model would certainly be a lower cost solution and it may well work
for them. You'll need to sit down and have an RPO/RTO discussion centered
around their business requirements to find out.
"Jesus Martin" wrote in message
news:%23vTiHaXUHHA.496@TK2MSFTNGP06.phx.gbl...
> Thanks John, They have 1000 mailboxes in each Datacenter so, to achieve
> what you say below, I see ...
>
> they would need 2 servers in site 1 working in a SCC model replicating the
> data using snapmirror replication to other 2 nodes stand by cluster and
> other 2 servers in site 2 with the mailboxes from site 2 replicating the
> data to site A using snapmirror to other 2 nodes stand by cluster
>
> this ends up with 8 servers and 8 enterprise editions of Exchange + all
> the other server roles what seems to be too much for them, They want to
> minimise the number of servers used but keeping in mind high availability
> and contingency topics as a main goal, So I would see couple of chances
> here:
>
>
> Using CCR ( 4 servers)
>
>
>
> Site 1 - 1 Active Node with 1000 mailboxes from Site 1 and 1 passive node
> with the 1000 mailboxes from the Site 2 replicating the data between DC
> using snapmirror
>
> Site 2 - 1 Active Node with 1000 mailboxes from Site 2 and 1 passive node
> with the 1000 mailboxes from the Site 1 doing the same than above
>
>
>
>
> or
>
>
>
> Using 1 SCC (8 servers) in each location and replicate the data to the
> other datacenter using netapp solution
>
>
>
> what do you think?
>
>
>
> thanks
>
>
>
> Jesus
>
>
>
>
>
> "John Fullbright [MVP]" <fjohn@donotspamnetappdotcom> escribi en el
> mensaje news:uoV1FWWUHHA.2256@TK2MSFTNGP02.phx.gbl...
>> In the CCR model, you'd use log shipping to send the logs to the passive
>> node where they are replayed. When a log file fills to 1M, it is
>> shipped, validated, and replayed. That may be every few minutes or less
>> on a very active SG, or a lot longer if the SG is not so active (yes,
>> there is the dumpster on the hub, but there are issues as to placement as
>> well). MS recommends VSS backups on the Seconday node due to the IO
>> intensity of copy on write snapshots and the IO activity associated with
>> validating the backup (per kb822896), so backups potentially lag as well.
>> There's clearly a gap here depending on how busy the storage group is.
>> If it's right for you really depends on what your SLA is. At a minimum,
>> with Netapp you get space utilzation similar to RAID 5 with no write
>> penalty. You also get to leverage the low IO impact of snapshots on the
>> platform and prerform them on both the active and inactive nodes at least
>> eliminating this part of the gap.
>>
>> In the designs Mark refers to, you would you the good old shared storage
>> cluster and snapmirror replication to a DR site using standby clusters.
>> Locally, The RTO is as long as it take the cluster to failover (a couple
>> of minutes) and the RTO is up to the minute. In the event of site
>> failure, the RPO can be as low as 5 minutes (it's dependent of the
>> frequency of log snapmirror updates) or so and the RTO is a few hours
>> (because you implement a standby cluster). The point with the RPO in
>> this case is that it is expressed in terms of time - the way business
>> objectives are expressed. This is very similar to designs for Exchange
>> 2003 out there and proven in the real world today. It falls within the
>> support policy for replication of Exchange data because snapshots, not
>> live data, are what is replicated.
>>
>> http://www.netapp.com/go/techontap/matl/three-tradeoffs.html
>> http://www.microsoft.com/technet/technetmag/issues/2006/10/FailoverClusters/default.aspx
>>
>>
>> John
>>
>> "Jesus Martin" wrote in message
>> news:ev4I%23KVUHHA.5060@TK2MSFTNGP06.phx.gbl...
>>>
>>> Mark, I don't know so much about SnapMirror but my as far as I know
>>> NetApp SnapMirror provides Data Availability copying the DB and logs to
>>> a secondary location using a cheap storage (if you want of course) what
>>> I don't see here is the service availability support. CCR enables you to
>>> use a passive node to support your users when the active node is down.
>>> Using the NetApp solution you should create a new storage group pointing
>>> to the replicated DB, this process is not immediate so the service can
>>> be down for a while
>>>
>>> What would you recommend here?
>>>
>>> thanks
>>>
>>> "Mark Arnold [MVP]" escribi en el mensaje
>>> news:7o49t29q9ds9k4gng4k93n504ajbs7p53s@4ax.com...
>>>> On Thu, 15 Feb 2007 18:02:34 +0100, "Jesus Martin"
>>>> wrote:
>>>>
>>>>>They are talking about using NetApp Snapmirror for Exchange
>>>>>
>>>>>thanks
>>>>>
>>>> If you have SnapMirror then you can put away all thoughts of LCR and
>>>> CCR as you just do not need it.
>>>> Sit down with your storage admin or call the people who sold you the
>>>> Filer and talk to them about Exchange 2007 configuration. They may not
>>>> be up to speed on it it's the same as 2003 so they will be able to
>>>> help you fully.
>>>>
>>>> Take a look at: http://www.netapp.com/library/tr/3407.pdf which will
>>>> help you understand SnapMirror if you're not already familiar with the
>>>> product.
>>>
>>>
>>
>>
>
>
date: Fri, 16 Feb 2007 07:01:00 -0800
author: John Fullbright [MVP] fjohn@donotspamnetappdotcom
|
|
