[ENBD] Re: NBD

[Peter T. Breuer]

"A month of sundays ago Thompson.Todd wrote:"
> I'm setting up a critical Linux file server that will contain 80 to 120 Gig
> of 1 to 50 MB image files for my company, and I ran across your article in
> LinuxJournal discussing NBD usage.   I want to have a situation in which one
> piece of hardware can fail, and the backup will keep chugging along with all
> data intact until I get the first one fixed, or vice versa, but I don't want
> to spend $2,000 for a commercial solution unless I have to, as I believe
> that true open source products are better tested and a good move in the long
> run.   NBD seems like it could fit the bill, but I have some questions.  

It really sounds like you should try it and then tell me the results! I
don't have those GB to test with.

> If I had two PC's with 120gb of disk space each, would the NBD solution work
> for me?  Would it matter if I was using RAID 5 across 3 or 4 disks on each
> machine, or if I used an Ultra ATA controller for hardware RAID?  Or could I

Gaaaaah. You are talking about RAID over NBD devices. In theory it
works. People tell me it works (some kinks on failover, reconnect,
etc., but all possibly part of their learning curves).

> set up a volume group across several disks on one machine and mirror that?
> I guess I'm just trying to get a "best practices" for my particular setup.
> I would imagine after reading your article that multiple NIC's would be a
> great plus as well.  Thanks so much for working on a project like this in
> the first place, as it seems to be just what a lot of people need, and
> thanks for any help you can give on these questions.

Well, there is continuing development.  As far as I can tell the basic
driver is rock stable under 2.2.* but possibly not under 2.4.0t1, which
is the latest kernel I have to test under.  The driver is the same in
all the nbd versions, apart from minor bugfixes and improvements.  The
user layer stuff above the driver has undergone considerable development
recently, and has not yet reached a stable state, though it probably
works a lot better than the old stable series does, except in some
end-conditions that are progressively becoming known and eliminated as
the code becomes understood.

 Cc:ed to mailing list in case better balanced views can be heard
here.

Peter