NLB or MS Cluster Service used for an HA pair?

[dynac0mp]

Use Case: Highly available iSCSI LUN to be offered to a MySQL Cluster. Require that the DB always see the Quorum and Data LUNs even in the event of failure of a single node with automatic / transparent failover.

Two main questions:

1. If I want to setup an HA pair of identical servers on the same VLAN, what is the network failover method based upon? NLB? Windows Cluster Service with a floating VIP?

2. Is there any requirement for the OS that connects to be Windows in order for automatic failover to work?

Thanks in advance for any assistance.

[anton (staff)]

0) For guaranteed uptime you'd better use 3-way rather then 2-way cluster (3 nodes and not 2). This will ensure airspace grade protection (triplication).

1) Unlike MS target StarWind does not rely on Windows clustering and does its own. So we can do active-active or active-passive (or triple active) HA clusters on non-clustered and different Windows versions.

2) Absolutely. Any server-grade OS shoud work (in theory desktop versions work as well we just don't test releases versions on them to save on support).

Use Case: Highly available iSCSI LUN to be offered to a MySQL Cluster. Require that the DB always see the Quorum and Data LUNs even in the event of failure of a single node with automatic / transparent failover.

Two main questions:

1. If I want to setup an HA pair of identical servers on the same VLAN, what is the network failover method based upon? NLB? Windows Cluster Service with a floating VIP?

2. Is there any requirement for the OS that connects to be Windows in order for automatic failover to work?

Thanks in advance for any assistance.

[dynac0mp]

Very good information, thanks so much for the quick reply! I have some further questions:

3) Can Linux nodes connect to the cluster IP (of the 'triplicated' storage cluster that you recommend) and get the same auto-fail-over functionality as Windows nodes?

4) In your experience what is the maximum latency that can be tolerated by members of a triplicated cluster before some type of halt/fault is experienced? 0.5 ms? 1ms? 5ms? (In this datacenter environment, for example, I cannot guarantee provisioning of servers on the same switch or even row, but I can be sure that they are in the same server room, this may cause one router hop. The network *IS* passing jumbo frames end-to-end BTW).

5) What is the best practice for offsite monitoring of the health of the storage cluster?

Also, feel free to contact me, Chris, directly at 214-243-4728 if you are interested in a service-provider level opportunity to partner.

[anton (staff)]

3) StarWind uses MPIO interface rather then shared IP *nix style cluster. Any software being compatible with MPIO will work with StarWind.

4) With Windows it's default 30 seconds (unless hacked in registry to make it bigger or smaller) before Windows I/O Manager will mark pending
request failed and use another MPIO path to retry request.

5) StarWind has built-in alert notification but if you want to have absolute current updates I'd suggest you to use RDP on StarWind Management Console running box.

[anton (staff)]

Yes, we're interested Guys will contact you on Monday (was CRAZY week for sales). Thank you!

[dynac0mp]

Anton,
I'm free after 1PM Central time if you want to try and line up a call. I'm certainly interested in speaking with you today if possible.
Chris.

[anton (staff)]

OK, looks like you've talked to Jack already

Let us know if you'd need anything else.