Starwind Two Node Setup



  • Assuming both nodes in a two node setup are the same. Is it better to split the VM load between the servers, meaning Starwind has to replicate in both directions, or is it better to run them all from a single host replicating in one direction?



  • To me, it would make more sense to split up the load between the two servers.



  • @Dashrender said:

    Assuming both nodes in a two node setup are the same. Is it better to split the VM load between the servers, meaning Starwind has to replicate in both directions, or is it better to run them all from a single host replicating in one direction?

    Are they on the same site? What is the end goal? Would it make more sense to do two-way replication and just have the hypervisor setup in a cluster automatically handle the resource provisioning?



  • Pinging @KOOLER


  • Vendor

    @Dashrender said:

    Assuming both nodes in a two node setup are the same. Is it better to split the VM load between the servers, meaning Starwind has to replicate in both directions, or is it better to run them all from a single host replicating in one direction?

    You create two virtual LUNs and you split your VMs between two hosts having own virtual LUN replicated to "shadow" one on a partner. Within this scenario there's minimal "fight" for any of LUNs ownership so performance is better.


  • Vendor

    @mlnews said:

    Pinging @KOOLER

    Replied! Thank you for brining me in 🙂



  • @coliver said:

    Are they on the same site? What is the end goal? Would it make more sense to do two-way replication and just have the hypervisor setup in a cluster automatically handle the resource provisioning?

    This question of mine came out of the discussion that Hubtech stated last week. Assuming a design where the two hosts (local to each other) are designed for failover, I was curious if it was better to have all VMs running on a single host or split.



  • @KOOLER said:

    @Dashrender said:

    Assuming both nodes in a two node setup are the same. Is it better to split the VM load between the servers, meaning Starwind has to replicate in both directions, or is it better to run them all from a single host replicating in one direction?

    You create two virtual LUNs and you split your VMs between two hosts having own virtual LUN replicated to "shadow" one on a partner. Within this scenario there's minimal "fight" for any of LUNs ownership so performance is better.

    How do you recommend the setup of the underlying disk? Would it be a One Big RAID 10? or two smaller RAID 10s?



  • Also, since I've never set something like this up before, considering it's only a two node setup, it is best to skip the network switch for the replication ports between hosts?

    And is the recommendation for ports two 10 Gb ports? or is something like 2-4 8 Gb ports normally good enough? (per server of course)



  • @Dashrender said:

    @coliver said:

    Are they on the same site? What is the end goal? Would it make more sense to do two-way replication and just have the hypervisor setup in a cluster automatically handle the resource provisioning?

    This question of mine came out of the discussion that Hubtech stated last week. Assuming a design where the two hosts (local to each other) are designed for failover, I was curious if it was better to have all VMs running on a single host or split.

    Split for performance and minimal impact during an "event." All on one for licensing. One more spot that Windows licensing often makes for a small, but impactful problem that makes Linux that much more beneficial in a case like this.



  • @Dashrender said:

    How do you recommend the setup of the underlying disk? Would it be a One Big RAID 10? or two smaller RAID 10s?

    Typically RAID 6 or even RAID 0. Remember that there is Network RAID 1 going on. RAID 10 is certainly an option, but far less common given the RAID 1 that exists on top.



  • I prefer to err on the side of caution... Knowing that a drive can blow out for any reason at all... I'd do at least RAID 6 on the hosts themselves and let Starwind handle the Network RAID 1.



  • I generally would do the same. Needing to rebuild a full node is a pain that I would like to avoid.



  • @scottalanmiller said:

    @Dashrender said:

    How do you recommend the setup of the underlying disk? Would it be a One Big RAID 10? or two smaller RAID 10s?

    Typically RAID 6 or even RAID 0. Remember that there is Network RAID 1 going on. RAID 10 is certainly an option, but far less common given the RAID 1 that exists on top.

    Wow, really? even considering the write penalties? AND the fact that you're mirroring to another server?



  • I can see doing RAID 6 if you can afford the performance penalty.



  • @Dashrender said:

    @scottalanmiller said:

    @Dashrender said:

    How do you recommend the setup of the underlying disk? Would it be a One Big RAID 10? or two smaller RAID 10s?

    Typically RAID 6 or even RAID 0. Remember that there is Network RAID 1 going on. RAID 10 is certainly an option, but far less common given the RAID 1 that exists on top.

    Wow, really? even considering the write penalties? AND the fact that you're mirroring to another server?

    You would think just RAID 0? Remember that with the mirroring you are already adding a lot of write penalty, the RAID 6 penalty probably will not be noticed.



  • @Dashrender said:

    I can see doing RAID 6 if you can afford the performance penalty.

    Which you would assume that you can if you can wait for a distant node to write as well.



  • @scottalanmiller said:

    @Dashrender said:

    I can see doing RAID 6 if you can afford the performance penalty.

    Which you would assume that you can if you can wait for a distant node to write as well.

    By distant, you mean local, as in the same rack?



  • @Dashrender said:

    @scottalanmiller said:

    @Dashrender said:

    I can see doing RAID 6 if you can afford the performance penalty.

    Which you would assume that you can if you can wait for a distant node to write as well.

    By distant, you mean local, as in the same rack?

    Yes, distant meaning outside of the chassis connected over a slow Ethernet link.