Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Friday 29 October 2004 17:49, Renaud Guerin wrote: > Hi, > > I'm trying to build a fairly large drbd setup, and I'd like to have your > advice on some design choices. > > I currently have 3 "master" Linux 2.6 NFS file servers with 5-7 500GB > volumes on each, and 2 "slave" backup machines, the first one holding 15 > volumes, and the other one the remaining 3 (for now). > So I want to use DRBD 0.7.5 to replicate the 15-20 "master" volumes from > the 3 NFS servers, to the 2 backup machines (that will also serve as NFS > failover servers) > > All the volumes on the master machines and the backup machines are setup > with LVM2, on top of external RAID arrays, so I'd like to do: > NFS server -> reiserfs -> DRBD -> LVM2 -> RAID arrays -> physical disks > > * First question : is this the right layer ordering, performance-wise > and maintenance-wise ? I've seen most people on the list do LVM2 on top > of DRBD-replicated devices, not the other way round like I plan to do. I > find the way I want to do it much simpler, but are there any drawbacks I > overlooked ?? It is just a matter of taste, and people have different tastes > * Is NFS over DRBD over LVM2 stable, are there any reported problems > with these components working together ? > I am not aware of any known issue. > > * The first backup machine will have15 drbd devices (and consequently > kernel threads) running in parallel. Was DRBD designed to handle > synchronization of so many different devices at the same time, or has > anybody tried this before ? I understand 255 is the theoretical limit, > but I'm nervous about stability/performance with this setup, is there a > good reason to be ? Deadlocks ? > No known issues. > > * I'm not sure I understand the 0.7.x al-extents parameter : the > default value of 257 says it will only resync 1GB after a primary node's > crash. > My question is simple : WHEN does the rest of my 500GB get replicated, > so that I actually have 2 copies of my data for safety, which is the > whole point of DRBD ? :) > Can you point me to an email discussion where this whole "active set" > concept is explained ? Thanks. > Initially everything will be resynced. Then you have the online replication. The AL allowes DRBD to sync only the 500GB and the blocks marked by the bitmap after a primary crash. You might read one of the papers about it see drbd.org / Publications. > * Do jumbo frames really help (I have tg3/bcm5700 Gigabit network cards) ? > The improve performance. > * Anything else I should be careful about ? > The software is free (GPL), the support [more than this e-mail] from me is not, consider to get into one of LINBIT's support offers. Then I will regard your problems as mine. > > Thanks in advance for any help ! You are welcome. -Philipp -- : Dipl-Ing Philipp Reisner Tel +43-1-8178292-50 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Schönbrunnerstr 244, 1120 Vienna, Austria http://www.linbit.com :