3 Node HA Configuration Recommendations

[Slingshotz]

This is what I currently have to work with:

Two new Dell R520 servers
One older Dell R510 server

Each server has exactly the same configuration except the R510 which has slightly slower processors. Each server otherwise has:
1. Dual 10Gig NIC ports and dual 1Gig NIC ports.
2. Raid 1 146GB SAS OS disks (two)
3. Raid 0 1.2TB SAS (three)
4. 32GB of RAM
5. The 10Gig NIC ports are all configured with jumbo packets of 9014

I want to create the 3 node Starwind cluster over all three servers but I will also host the Hyper-V cluster on these servers. What I've done so far is connected each server directly with the 10Gig links as per the Starwind PDF 3-node diagram (Sync 1, 2 and 3). Since the Hyper-V cluster is local to the servers, I have teamed the 1Gig NIC ports on each server to our one switch (we do not have redundant switches yet).

So my questions are:

1. I have read that NIC teaming is not recommended for the cluster sync but I assume that it should be fine if it is only Hyper-V traffic going over it?
2. Is there is any issue or benefits for running a 2-node vs 3-node Hyper-V cluster right on top of the 3-node Starwind cluster? If running a 3-node Hyper-V is better due to redundancy, can it be configured to keep the R510 as the last possible server for live migration?
3. I can configure the 10Gig ports for jumbo packets of 9614, is there any benefit of this?
4. I configured the Raid 0 with the default stripe element size of 64KB, read ahead, write back and disk cache enabled. Are those settings ideal or is it better to have different options to cut down on overhead?
5. When I format the disk in Windows, should the allocation unit size be left at default or changed to match the stripe size?
6. Since the Sync and Heartbeat will be going over the 10gig links, is it fine to have the heartbeat only channel network connection shared with the Hyper-V connections?

Thanks in advance.

[Anatoly (staff)]

1. I have read that NIC teaming is not recommended for the cluster sync but I assume that it should be fine if it is only Hyper-V traffic going over it?

We can give our recommendation only regarding our software, so this question should be be better to ask at MSFT side. In my personal opinion, I wouldn`t said that NIC Bounding has some other benefits except easiness in management, while bringing a bunch of negative aspects.

2. Is there is any issue or benefits for running a 2-node vs 3-node Hyper-V cluster right on top of the 3-node Starwind cluster? If running a 3-node Hyper-V is better due to redundancy, can it be configured to keep the R510 as the last possible server for live migration?

You are getting better reliability with 3 nodes obviously. What is unobvious is that you are getting better perforamance with 3 nodes, since more "heads" of HA means more paths, and that is equal to better performance if you are using Round Robin.

3. I can configure the 10Gig ports for jumbo packets of 9614, is there any benefit of this?

That choice should be test based. Sometimes you can get better perfromance even without using JF.

4. I configured the Raid 0 with the default stripe element size of 64KB, read ahead, write back and disk cache enabled. Are those settings ideal or is it better to have different options to cut down on overhead?

Recommended RAID for implementing an HA are RAID 1, 0 or 10, RAID 5 or 6 are not recommended due to low write performance ((btw the last two ones are not recommended to put into production by LSI, RAID vendor).
The performance of a RAID array directly depends on the Stripe Size used. There are no exact recommendations of which stripe size to use. It is a test-based choice. As best practice we recommend at first step to set value recommended by vendor and run tests. Then set a bigger value and run tests again. In third step set a smaller value and test again. These 3 results should guide you to the optimal stripe size value to set. In some configuration smaller stripe size value like 4k or 8k give better performance and in some other cases 64k, 128k or even 256k values will give better performance.
Performance of the HA will depend on the performance of the RAID array used. It’s up to the customer to determine the optimal stripe size.

5. When I format the disk in Windows, should the allocation unit size be left at default or changed to match the stripe size?

the 32k and 64k are prefered.

6. Since the Sync and Heartbeat will be going over the 10gig links, is it fine to have the heartbeat only channel network connection shared with the Hyper-V
connections?

It should work.

[Slingshotz]

Thank you very much for the clear answers. With regards to your answer 4, how do you actually test the performance? Is there a recommended utility, perf mon or manual test of some sort that can give me some clear numbers to compare?

[Anatoly (staff)]

I`m sure that those docs should guide you really well:
http://www.starwindsoftware.com/starwin ... ice-manual
http://www.starwindsoftware.com/images/ ... _guide.pdf