February 7, 2013 at 8:48 am
Hi there
We have a WMware host with a SQL2008R2 cluster.
Disks are raw mapped to a EVA 6400 SAN.
We have performance problems, we have located it to network problems.
To much packet los and there for resend, this is a hardware/ hardware driver issue.
We have issued it to the hardware vendor and they are aware about the problem and are looking in to it.
The vendor can't provide us with an estimated date for fixing it, there are critical systems on the cluster and we can't wait for
the fix to get ready, so we need another fix.
We have two new machines (physical) machines installed (atm windows only).
Now we need a method to migrate to the new machines.
As i se we have a couple of options.
1. Side by Side (system by system), takes much time but it's a failsafe method however it takes much time.
Pro: No need for roolbackplan, safe
Coins: Takes time
2. Install new servers, add them to the cluster, install SQL. Failover to the new nodes, shutdown the old machines.
This method sounds good, but has it ever been tested with the combination of physical and virtual machines?
Between physical only machines it should work.
Pro: Fast, minimum downtime, Roolback plan available.RollbackFeels risky, due to diskmappings. (Wmware on this lvl isn't board).
3. Install a new cluster, backup/ restore (deattach/ attach not an option since i dont want to change the enviroment due to roolback plan) from old cluster to new cluster.
Shutdown old cluster, rename new cluster and restart machines, and pray that the 30 systems will find the cluster.
Another thing is DNS, pray that windows will accept the new cluster with same name.
Pro: Prettyfast, accepted downtime.
Coins: A bit risky
Any other suggestions, or what would you do?
I allways like to go the safe way like method 1, but buissnes requirements might think other...
Sorry for any kind of bad english gramar and such...
February 7, 2013 at 9:05 am
1 and 3 are essentially the same arent they?
Knut_knut (2/7/2013)
2. Install new servers, add them to the cluster, install SQL. Failover to the new nodes, shutdown the old machines.This method sounds good, but has it ever been tested with the combination of physical and virtual machines?
Between physical only machines it should work.
Pro: Fast, minimum downtime, Roolback plan available.RollbackFeels risky, due to diskmappings. (Wmware on this lvl isn't VMWarerd).
Clustering a physical and a virtual machine is quite common and works well when done properly.
I wouldn't be in too much of a rush to just dump the virtual infrastructure just yet. Can you provide a little more info on the system setup?
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
February 7, 2013 at 9:15 am
Will try to find some more input.
Guess we need..
ESX version (know its 4 or 5 not sure though)
How the disks are mapped (what ive heard this can be done more than one way.)
Nics (versions)
Virtual servers are running Windows 2008 R2 and SQLserver2008R2
Host are HP blade servers
System installed on C disk and SQLbinarys on D disk
Datadisk 400 GB
Logdisk 200 GB
Quorum 5 GB
DTC 5 GB
February 7, 2013 at 9:32 am
Knut_knut (2/7/2013)
ESX version (know its 4 or 5 not sure though)
Important to know, there's big differences and improvements between 4.0, 4.1 and 5.1 😉
Knut_knut (2/7/2013)
How the disks are mapped
You stated Raw Device Mappings, is this correct?
Knut_knut (2/7/2013)
Nics (versions)
What virtual NIC adapter type are you using on the VM's?
How many NICs per VM?
Knut_knut (2/7/2013)
Quorum 5 GBDTC 5 GB
Keep these small, 512MB per disk is more than sufficient.
-----------------------------------------------------------------------------------------------------------
"Ya can't make an omelette without breaking just a few eggs" 😉
February 8, 2013 at 2:14 pm
In method (1) side by side , are you planning to have different cluster names?. I would assume to end the migration ,you still need to rename the cluster and let the other 30 machines recognize it
February 11, 2013 at 12:27 pm
Still no info on ESX versions :/, i'm trying to get intouch with the wm people.
Side by Side - Yepp 2 clusters and move system by system takes time but its a safe method.
However it seems like i have to do this fast (i dont like shooting fast ...) since the virtual isn't performing accaccording the SLA we have to mak a fast solution,
I think i will use the solution where i bring up a second cluster backup/restore on the new cluster, bring down the old, rename the new.
I did a similar thing a few days ago on a single sql2008r2 with 6 instances, a few small hickups due to high security thinking with accounts and so other than that it went more smooth then i expected it to do.
The only thing was the naming in active directory, where we had to delete the server and register the new again...
Thanx to all of you who respond and discuss....
I will update here when the job is done and post the results for others to read. (Job has to be done in 3-4 weeks.)
Viewing 6 posts - 1 through 5 (of 5 total)
You must be logged in to reply to this topic. Login to reply