[DRBD-user] master/master over WAN with GFS

Brian R. Hellman brian at linbit.com
Tue Jul 14 19:20:22 CEST 2009

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


I could be done, but again like Lars said, performance will be very bad
(unusable) and the likelihood/risk of split brain high.  As soon as your
Internet providers link goes down, and it will, you'll have two
different data sets without a way to merge them. Then you get the fun
chore of deciding who gets to loose their emails :)



LINBIT - Your Way to High Availability
8152 SW Hall Blvd., Suite #209 : Beaverton, OR 97008

http://www.linbit.com


Mikel Jimenez wrote:
> Ok Thnaks for the answers!!
> 
> So... there is not a way to have to machines synchronized over WAN and
> the two machines with the filesystem mounted in RW ?
> 
> Thanks
> 
> Lars Ellenberg wrote:
>> On Tue, Jul 14, 2009 at 05:58:14PM +0200, Lars Marowsky-Bree wrote:
>>  
>>> On 2009-07-06T20:16:26, Mikel Jimenez <mikel at irontec.com> wrote:
>>>
>>>    
>>>> The users in LAN of A would access to server A and the users in LAN
>>>> of B would access to server B.
>>>> When I user access to the webmail or A looks the same that looks
>>>> from B.
>>>>
>>>> We have a simetrical VPN between two sites.
>>>>
>>>> How does DRBD result    in this type of environment?
>>>>
>>>> I think that the best is "Protocol A" and Im looking for use
>>>> DRBD-Proxy for compression, or Openvpn + LZO.
>>>>
>>>> Opinions?
>>>>       
>>> The latency overhead - both DRBD and at the locking layer - will kill
>>> the file system performance.
>>>     
>>
>> appart from that, with dual-primary you _MUST_ replicate syncronously,
>> thus: DRBD protocol C.
>> for (hopefully) obvious reasons.
>>
>> I still have this idea of implementing a replication aware
>> cluster file system locking layer mode, where lock _requests_ would be
>> inserted into (or at least in order with) the replicated data stream, so
>> it could be replicated asynchronously in theory.
>>
>> IFF it can be made to work, that is probably several manyears
>> developement, though -- roughly estimated ;)
>>
>>   
> 
> _______________________________________________
> drbd-user mailing list
> drbd-user at lists.linbit.com
> http://lists.linbit.com/mailman/listinfo/drbd-user



More information about the drbd-user mailing list