[ENBD] Problem with mkfs on enbd

[Peter T. Breuer]

"Also sprach Stephan von Krawczynski:"
> Sorry for this dumb question:
> where do I set the blocksize parameter? enbd-client or enbd-server ?

Dunno! I was hoping you could tell me that! It's -b, so I guess I have
to look in the manual to see which one has that option ... the server
has it ... and the client ... therefore the answer is "both". I think
the larger of the two offers will be agreed upon. In retrospect, that
is not such a good idea.

All vote for me to change "max" to "min" in the code? Say "Aye"!

> Currently I have enbdclient -b 4096 ...

Well, you want 512, I think.

> 
> > > flushing buffers..done
> > > reading and
> > > checking....5%....10%....15%....20%....25%....30%....35%....40%....45%....
> > > 50%....55%....60%....65%....70%....75%....80%....85%....90%....95%....
> > > (1000KB written in 5.100177s, 0.191476 MB/s, 392.143253 seeks/s) (1000KB
> > > read in 0.003492s, 279.657073 MB/s)done
> > 
> > That's strange though. I'll have to check that code. How long did that
> > test take?
> 
> Well, you got me. Actually I don't know exactly. I remember though that it was
> a matter of 10-20 minutes.

OK, so long as it wasn't the "5s" stated above! That must just be the
seek test. If it took 5m (300s) for each test, and you said, I think, you
used 10GB, then that would be 30MB/s, which seems correct to me.

> > > enbd-test on /dev/md0 went nuts. The box got unusable, no login, no shell
> > > action possible. I had to cold-reboot.
> > 
> > Well, sounds like md0 somehow does not work, but the ndX do.
> > 
> > Care to tell anything more about the box? Arch and so on.
> 
> Linux mail-a01 2.4.30 #2 SMP Mon Apr 11 12:56:03 CEST 2005 i686 i686 i386 GNU/Linux

Yes, but what arch? The above is the kernel string.  Is it 64 bit or 32
bit, I am asking. The string says "SMP" - is that really the case?

> 
> Hardware is an Intel box with Intel Raid Controller:

Raid? And you are using softraid.

> processor       : 0
> vendor_id       : GenuineIntel
> cpu family      : 15
> model           : 2
> model name      : Intel(R) Xeon(TM) CPU 2.80GHz

Aaaaaaaaaaaaarggggg..  Xeon. I don't suppose that would be 64 bit,
would it? yess it is. And I don't have a hope of emulating that
platform.


> processor       : 1
> 

Well, I have had reports of enbd working on amd 64 bit, but I honestly
don't know about the xeon 64 bit stuff - I can honestly say that I have
never even tried compiling for it. I suspect that there is "a bug".

Mind you, it's a bug that only shos up when softraid is run over enbd,
and I can't think .... could it be the ioctls? enbd sends ioctls to and
fro from the raid array. An it expects to register a callback function. 
I wonder if some of that could be 32 bit specific?

If you could comment out the raid call  stuff in the driver, that would
be helpful. It must all be in enbd_md.c ... yes, hmmm, in enbd_base.c
there are lines like 


     err = enbd_md.dec(&enbd_md);
     if (err < 0)
          return err;

     err = enbd_md.inc(&enbd_md);
     if (err < 0)
          return err;

     enbd_md.reg(&enbd_md, (int(*)(dev_t, int))arg);


Could you comment those out? Anything that looks like it references
"enbd_md". And any calls to "enbd_notify_md_devices" (the latter may be
enough on its own). 

See if that helps.

Peter