Mirroring 1U twin servers

Discussion in 'Business & Enterprise Computing' started by ex4n, Nov 5, 2012.

  1. ex4n

    ex4n Member

    Joined:
    Oct 5, 2011
    Messages:
    1,998
    Location:
    Perth
    Hi guys,

    Looking at these 1U supermicro twin servers, would it be possible to configure these servers in some kind of mirrored config, similar to raid1 for HDDs.

    I assume this is simply not possible at a hardware level, but correct me if I am wrong, and so I am wondering if there is any software that could do something like this?

    Effectively both servers would be doing the exact same thing and if one fails, the other takes over until it can be replaced.

    Sorry for the fairly vague question, any help or suggestions would be appreciated.

    Cheers
     
  2. gords

    gords Thread Killer

    Joined:
    Aug 3, 2001
    Messages:
    6,585
    Location:
    Sydney, Australia
    You probably want to provide details of intended use, OS etc.
     
  3. ex4n

    ex4n Member

    Joined:
    Oct 5, 2011
    Messages:
    1,998
    Location:
    Perth
    thanks gords, that's probably a good idea. it is something a potential client has asked about. He is likely to be running a windows server, not sure what he will be doing with it yet.

    I have looked at drdb for nix + heartbeat which could potentially do this well, but I think he's a windows man..

    Edit: Will update with more info when I can get it, was just looking for ideas for now.
     
    Last edited: Nov 5, 2012
  4. EvilGenius

    EvilGenius Member

    Joined:
    Apr 26, 2005
    Messages:
    9,157
    Location:
    _Rocky Status:_Folding!
    Possible simple solution could be:


    Install similar apps on each server.

    Copy on hobocopy.

    Set a scheduled task to hobocopy one server over top of the other.


    Suggesting hobocopy as it creates a snapshot before copying, so you can copy appdata etc of running programs. Basically, create a .bat to copy the entire user profile over from one server to another.
     
  5. 3t3rna1

    3t3rna1 Member

    Joined:
    Dec 24, 2001
    Messages:
    1,452
    Location:
    Perth
    It's the exact same as setting up redundancy on any two servers, just because they're in the same case doesn't make it any easier. They also share the same (non-redundant) power supply so they will both go down with a power related fault.
     
  6. ex4n

    ex4n Member

    Joined:
    Oct 5, 2011
    Messages:
    1,998
    Location:
    Perth
    thanks guys, i've never looked at this kind of redundancy before, usually i would spec a single machine not to fail, such as 2x psu, 2x rc, mirrored disks etc.

    I probably could have worded the OP better, I appreciate the options to look into :)
     
  7. DavidRa

    DavidRa Member

    Joined:
    Jun 8, 2002
    Messages:
    2,902
    Location:
    NSW Central Coast
    Sounds like Windows Clustering could help, depends on what you're actually trying to do. Little known fact - you can cluster a heck of a lot more stuff than just the normal suspects (Exchange, VMs, SQL). As long as it's Windows Services or you can script it in VBScript/JScript, you can cluster it with health checks etc.

    We need to know more about what's what. Data replication or shared storage? Maybe two-phase commit? Intended apps? RPO requirements (how much data on your "mirrored" server can you afford to lose)?

    Is it mirroring for availability or performance? If it's availability, what happens when the place catches fire? How do you plan to get a consistent backup if the data is in flux?
     
  8. ex4n

    ex4n Member

    Joined:
    Oct 5, 2011
    Messages:
    1,998
    Location:
    Perth
    When the client came to me today, I had assumed he was asking as he had seen/heard/read about this being done somewhere, upon further investigation, it seems he just assumed that it would be simple, like mirroring hdd's. I posed the question to you guys in case there was actually a simple piece of software capable of doing this.

    This was for redundancy only, so if one server failed, the other would take over and there would not be a second of interruption. I have explained to him the complexity of achieving something like this, and that there are other, better options for achieving the same thing.. so he is going to get back to me about his intentions.

    Sorry I can't be much more specific as he never told me exactly what he was going to be doing with them, will update again if I get more info or any follow up.

    It is most likely the idea will be dropped for now.. in favour of a more practical solution :)

    Appreciate the input all.
     
  9. Gecko

    Gecko Member

    Joined:
    Jul 3, 2004
    Messages:
    2,716
    Location:
    Sydney
    The "simple" option is to use virtual machines on top of your twin servers. There will still be some downtime if one host unexpectedly falls over while you boot up the VM on the other node, but if you know it is coming in advance, you can live migrate your VM to the other piece of hardware.

    We have a couple of small clusters set up in this way where 1 node can (just) take all of the load (we suspend any unimportant jobs if we're stuck on 1 node), but usually the two nodes have the load spread evenly across them.

    If your application is cluster-aware (or can be made to be cluster-aware), that is usually a better option. For example, if you're doing web-hosting duties, the best option would be to stick two web servers behind a load balancer and synchronise the web files between the two nodes.

    Of course, if your budget is unlimited, specialised hardware is available such as the Stratus Continuum products. Even then, that requires some thought at the application level to properly leverage the capabilities of the box.
     
  10. elvis

    elvis Old school old fool

    Joined:
    Jun 27, 2001
    Messages:
    29,973
    Location:
    Brisbane
    Sounds like another case of someone reading some buzzwords in CIO Magazine while waiting in the Qantas lounge, and then telling IT in an offhand comment to "make it happen".

    For infinite use cases there are infinite solutions at infinite price points. Specific requirements will give your client a specific answer.
     
  11. Mystro1

    Mystro1 Member

    Joined:
    Jun 27, 2001
    Messages:
    684
    Location:
    Victoria Point, QLD
    Depends on the $$$ we are talking, for simple Mirroring with Auto fail-over you can do this a couple of ways, Active/passive style (one way replication) using something like Double Take Availability as a physical option or using Hyper-V replica to replicate the Virtual Machine to a hot spare (trans log style shipping so potential small data loss as with most/any replica's) but both options have automatic fail-over. Both these options will support Linux OS's (however the Hyper-V option is limited somewhat) (Yes you can use the Free Hyper-V 2012 Core version as its supports Replica too)

    As with all replica style services and defiantly with Hyper-V replica there is a I/O overhead that needs to be accounted for but if your replicating a single VM just for the HA factor then you wont need to worry about that.

    Next up you have the Clustering for Active/passive or Active/active HA depending on the application type.

    Just my 2 cents...
     
  12. [eXo]

    [eXo] Member

    Joined:
    Jul 31, 2001
    Messages:
    987
    Location:
    Sydney
    If he has some kind of shared storage, you could put in a failover cluster, however a 2 node cluster is troublesome. Something like HyperV or VMware could help if you want to have multiple autonomous systems running on it.

    It depends if you want to just replicate the data for redundancy, or actually run applications on the two servers. If you had some kind of plan and costs in mind we might be able to help out more.

    Data replication between two servers is certainly possible, with cli things like robocopy, but they are never 100% 'mirrored'. With these though, large file copies are a pain as it needs to replicate the entire file with a minor file change (think email PSTs). Block level replication is better for this kind of thing. Options for this could be DFS-R (Free with Windows 2008) or there is also another product that used to be Veritas but I believe they were bought by another company so the product might be rebranded. It was called VDR or something (been a few years since I used it) It established a relationship between 2 server disks and replicated changes only.
     
  13. h-90

    h-90 Member

    Joined:
    Aug 19, 2002
    Messages:
    880
    Windows Server 2012 should make this type of clustering easier with SMB 3, SAN like replication between servers.

    Also server 2012 then can do things like clustering updates, where it copies services over to one server whilst it performs updates on the other.
     
  14. Iceman

    Iceman Member

    Joined:
    Jun 27, 2001
    Messages:
    6,647
    Location:
    Brisbane (nth), Australia
    Since we're throwing out crazy ideas I'm surprised nobody mentioned vmware's product called "fault tolerance"..

    Rather than try and explain it, I'll just paste the blurb:

    vSphere Fault Tolerance (FT) provides continuous availability for applications in the event of server failures by creating a live shadow instance of a virtual machine that is in virtual lockstep with the primary instance. By allowing instantaneous failover between the two instances in the event of hardware failure, FT eliminates even the smallest chance of data loss or disruption

    Haven't seen anyone try and gold plate their solution to this degree .. yet.
     
  15. aza2001

    aza2001 Member

    Joined:
    Sep 14, 2002
    Messages:
    2,015
    Location:
    Northmead
    I would suggest something like ESX and Veeam backup and replication.

    it would just snapshot and replicate the data between the servers every say 15min-30min so you would only loose that much per sync...
     
  16. Onthax

    Onthax Member

    Joined:
    Nov 5, 2003
    Messages:
    424
    I think the killer is the requirements. Alot of requirements + enterprise plus licensing if i'm not mistaken.
     
  17. NSanity

    NSanity Member

    Joined:
    Mar 11, 2002
    Messages:
    16,190
    Location:
    Canberra
    Really depends on the Apps, what outage the client is prepared to take and how much money he wants to spend.

    Having a duplicate cold server sitting there, that you can yank some Raid1 drives out of the previous server, insert into the cold one, import the foreign raid config and boot is probably the cheapest way.

    Then you move to using things like Windows/Linux clustering if your app supports it. This is just 2x Win Server 2012 Std licences (2x Win Server 2008R2 Enterprise though if you can't go to 2012) + SQL Enterprise or whatever. Obviously Linux+Postgres/mysql is free.

    Then you move to VMWare/Hyper-V/Xen/KVM Guests in a cluster. Depending on tech, you may need shared storage.

    Then you move to VMWare/Hyper-V/Xen/KVM Guests in FT mode. Depending on tech, you may need shared storage.
     
  18. PabloEscobar

    PabloEscobar Member

    Joined:
    Jan 28, 2008
    Messages:
    9,914
    Why am I only finding out about this now?
     
  19. Iceman

    Iceman Member

    Joined:
    Jun 27, 2001
    Messages:
    6,647
    Location:
    Brisbane (nth), Australia
    I just meant because the whole discussion is pie in the sky. When people are presented with the cost of a even a basic clustered solution (that still has down for the time of failover) they usually turn green and decide it won't be the end of the world if their SBS box is down for an hour to replace a failed drive :p
     
  20. Gecko

    Gecko Member

    Joined:
    Jul 3, 2004
    Messages:
    2,716
    Location:
    Sydney
    And then proceed jump up and down screaming when something does break...
     

Share This Page