Hi Philipp,

Thanks very much for such valuable response, it helps me a lot.


BTW, is the proprietary add-on module for IB (wit rdma) from LINBIT a open 
source or sold with LINBIT production?
Could you please provide some details for it? e.g. the URL link.


Best regards,
Gu




------------------ ???????? ------------------
??????: "Philipp Reisner";<[email protected]>;
????????: 2015??8??31??(??????) ????3:57
??????: "drbd-user"<[email protected]>; 
????: "????"<[email protected]>; 
????: Re: [DRBD-user] [Questions] Many questions about DRBD



Am Mittwoch, 26. August 2015, 10:36:16 schrieb ????:
> Hi guys,
> 
> I am a new comer to DRBD, as we want to use drbd for data backup, and I
> have some questions about drbd, so hope you guys can help me a lot.
> 
> 1. Does drbd support rdma for IB (InfiniBand)? as we need to use IB for
> high performance network rather tcp/ip.

Yes, it does. By a proprietary add-on module from LINBIT. 

> 2. How is the local I/O performance drop degradation if compared without
> drbd used?

Very depended on your hardware. Under optimal circumstances (network link
strong enough, CPU powerfull to handly IO requests) it can be below 3% in
throughput.

> 3. How many replicate modes does drbd support? e.g. sync, async or other
> ones.

C - synchronous, B - memory synchronous, A - asynchronous

> 4. Does it support intermittently replicate ? I mean, sometime we turn-off
> copying data to remote backup device, sometime we can turn-on it again, and
> them force them to be synchronous via some manual intervention (e.g. rsync).

Yes it support temporal network failures or/and administratively imposed 
pauses of the synchronization process.
As soon as the connection is re-established (or re-enabled) DRBD does a
"resync" by itself. I.e. it copies over the 4k blocks  that was touched on
the primary node during the pause of the mirroring.

Best regards,
 phil
_______________________________________________
drbd-user mailing list
[email protected]
http://lists.linbit.com/mailman/listinfo/drbd-user

Reply via email to