[DRBD-user] Is it worth considering eth bonding in this setup?

Luca Lesinigo luca at lm-net.it
Mon Dec 22 19:13:38 CET 2008

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


[Eric, sorry for the direct reply. I'm resending to the list now. I  
totally hate lists without the list address in reply-to :)]

Il giorno 22/dic/08, alle ore 18:48, Robinson, Eric ha scritto:
> What would be your main goal in considering bonding, bandwidth or  
> availability?
Any incremental advantage over the two single links setup would be ok.  
I haven't much experience with bonded setups.

I don't know if a single disk is going to saturate a single direct  
jumbo gigabit so I don't know what kind of performance improvement I'm  
going to get. I also don't know if the bonding will introduce some  
additional penalties (more code path = more latency, buy I'd think  
it'll be so little as to be unmeasurable).

I'd prefer availability since this is a long running, remotely  
colocated setup, if it doesn't come at the cost of something else. The  
setup shouldn't be too complex since I don't have anything else on  
those links, so I haven't any problem of bridging (for xen domains, on  
the lan and wan links), firewalling (ditto, and I'd only have drbd  
talking on those IPs), traffic accounting, additional failure points  
(no switch, only crossover cable), etc.

As far as I know the bonding driver shouldn't have problems with jumbo  
frames and it should be rock solid. The documentation of the e1000  
driver explicitly states it supports channel bonding.

In my setup (two crossover cables, no switches) the balance-rr bonding  
method will probably be the best one provinding both improved  
bandwidth and improved availability.

--
Luca Lesinigo
LM Networks Srl




More information about the drbd-user mailing list