[DRBD-user] Re: Suggestion to prevent split brain situation

Bas van Schaik bas at tuxes.nl
Wed Dec 5 10:17:31 CET 2007

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


Florian Haas wrote:
> On Wednesday 05 December 2007 01:37:00 Kelly Byrd wrote:
>   
>>> And is it correct that DRBD only supports a single IP per machine using
>>> the "address" directive? In that case the redundant link is only useful
>>> for HeartBeat, but not to DRBD?
>>>       
>> Use the bonded driver. Depending on the bond mode, you can get just
>> redundancy or you get increased bandwidth and redundancy.
>>     
>
> Correct, but
>
> - increased bandwidth isn't worth much for improving DRBD performance unless 
> your disk I/O subsystem can indeed write faster than 110 MB/s (unlikely in 
> the case of Bas' servers I suppose);
> - added redundancy doesn't make a huge difference in case you are using 
> crossover cables between nodes (which, when they fail due to hardware issues, 
> typically don't fail in a way that the bonding driver would detect). 
>
> If and when you're running DRBD replication over switches, you had better make 
> sure that your cables are routed through physically separate switches hooked 
> to independent power sources. Otherwise, again, bonding doesn't do you any 
> good either. 
>
> Our recommendation thus is generally to use a passive (i.e. back-to-back) 
> network connection for your DRBD replication. Crossover cable is highly 
> likely to provide the best MTBF. Add Heartbeat with that connection and a 
> separate one via your switched LAN, plus dopd, and you're good to go.
>   
Yes, but as I said: I'm still using DRBD 0.7 and dopd seems to be part
of DRBD 0.8?

  -- Bas




More information about the drbd-user mailing list