Windows 2008 Server Usage

[warhed]

I apologize ahead of time if I am completely using Starwind iSCSI in an incorrect way, I am very new to this.

We have two Windows 2008 R2 SP2 Enterprise Servers. Our desire is to have real time (which I don't believe is possible with our environment) or at a minimum quick manual failover for mission critical apps and services.

Mission Critical Apps:
1. A 200gig database system with about 75-100 users using it at a time. In the event of the primary server that hosts this database and client, we would want the database to be available and current up to the point of failure. Then we could simply redirect users to the secondary server via DNS name change on the Admin level or by simply providing a new name to the secondary server, and have users log in again.

2. Similar to the above situation, we are also using Hyper-V on Top of Windows. I would like to place all the VM's within a Starwind iSCSI container such as RAID1 or HA so that the data is replicated in near real time between two servers. When the primary server fails, I could simply connect to the secondary server, connect the iSCSI RAID1/HA Partner drives and start them up with Hyper-V on the 2nd Wind2008 server. I am aware that you cannot connect two servers at the same time to the same iSCSI NTFS drive, but a manual failover and startup is ok with us.

Is there a best practice for this? Seems like everyone is using vSphere or Xen on here but that is just not something we can afford at this time.

Thanks for any suggestions on the best way to implement this.

[mooseracing]

I'm not sure which license of StarWind you have but have you looked into setting up Windows Clusters? Everyone thing you talk of about wanting HA and failover, I would go that route.

I know it is pricey. We are small company, just under 100 people, but they didn't want downtime. All of our virtual servers in Hyper V except for a couple that aren't real important are On a Hyper V cluster shared across 2 StarWind HA Servers.

IIRC if you don't want to go that big there is also SQL Clustering. This stuff is not as hard to setup as the old clustering days, just make sure you do it like they suggest and test well.

[venom]

Thank for the reply I appreciate it. However from my reading about Windows Clustering, the actual application itself has to be Cluster Aware such as Hyper-V, Exchange and SQL. You cannot just cluster anything, such as our ancient but highly critical database (Sybase).

What I have come across by accident is this and I have yet to test it this week.

Two servers, we will call them ServerA and ServerB

ServerA - Starwind - Setup as RAID1 mirror. The Source of the RAID1 mirror is itself, ServerA
ServerB - Starwi9nd - Basic CDP disk.
ServerA sets the second RAID1 mirror (Partner) to Server B's basic CDP disk)

So what I hopefully have accomplished is Server B being able to snapshot every 30/60 minutes, and restore the database in near real time or with very low downtime, and if ServerA physically dies, we can now iSCSI Target to ServerB with the database and get on with work while ServerA gets fixed.

I have a lot of testing to do on this obviously, but I have not seen any documentation on this method that I am trying. Not sure if it is supported.

[venom]

One thing I am worried about that will hopefully be answered after testing is the utter failure and restoration of the Source RAID1 Starwind server, ServerA.

If that comes back online after ServerB has been providing the database to end users, will ServerA and Starwind recognise the being out of sync (initial testing says yes) but will it realize the Partner of the RAID1 is now the primary and sync back from ServerB to ServerA.

Also the fact that snapshotting may not work with this databse. Initial tests using a compressed .zip file as a "database" and adding files to it, then sealing the zip file again, has resulted in snapshots being the entire size of the zip file. So if we had a 1 gig zip files, snapshotted, it would be 1gig. Snapshot again and it would be zero. Add 100mbs to that zip file, and close it, the snapshot is now 1.1gigs rather than the 100mb difference. This is probably a poor test, but I am coming from a Delta type background mentality which only backs up at the bit level. More real world testing must be done, but our database is 200gigs big!

[Max (staff)]

Venom, unfortunately your idea will not work as expected, snapshots taken this way may be inconsistent and overall speed will be too slow.
On the other hand we have a better solution with utilizing servers both as DB and HA nodes: you just need to install StarWind on both servers and configure their own storage as HA and present it to Sybase.

[clayton@mcc911.org]

WORD!

Venom, unfortunately your idea will not work as expected, snapshots taken this way may be inconsistent and overall speed will be too slow.
On the other hand we have a better solution with utilizing servers both as DB and HA nodes: you just need to install StarWind on both servers and configure their own storage as HA and present it to Sybase.

[anton (staff)]

In the future versions of StarWind the whole idea of using mirror with a single point of failure (primary controller head) would be suspended. We'll proceed with full HA (2+ nodes) and async on-line and off-line replication only. So as Max had suggested you're recommended to stick with HA model for now.

One thing I am worried about that will hopefully be answered after testing is the utter failure and restoration of the Source RAID1 Starwind server, ServerA.

If that comes back online after ServerB has been providing the database to end users, will ServerA and Starwind recognise the being out of sync (initial testing says yes) but will it realize the Partner of the RAID1 is now the primary and sync back from ServerB to ServerA.

Also the fact that snapshotting may not work with this databse. Initial tests using a compressed .zip file as a "database" and adding files to it, then sealing the zip file again, has resulted in snapshots being the entire size of the zip file. So if we had a 1 gig zip files, snapshotted, it would be 1gig. Snapshot again and it would be zero. Add 100mbs to that zip file, and close it, the snapshot is now 1.1gigs rather than the 100mb difference. This is probably a poor test, but I am coming from a Delta type background mentality which only backs up at the bit level. More real world testing must be done, but our database is 200gigs big!

[chop]

Hi Anton

Is there any timeline for the the mirror version will be replaced with HA? The reason I ask is that we have only just paid for the mirrored edition.

[anton (staff)]

V6 and it's Fall 2011.

You'll be upgraded to HA version (more expensive one) free of charge.

Mirror one is fine except we would like to provide better service to our customers.

Hi Anton

Is there any timeline for the the mirror version will be replaced with HA? The reason I ask is that we have only just paid for the mirrored edition.

[chop]

Cannot wait, am looking forward to the new version and services.

Cheers

[anton (staff)]

Don't shoot the piano player he's doing his best (c) ...

Cannot wait, am looking forward to the new version and services.

Cheers