[OpenAFS] MR-AFS [was: Re: OpenAFS Project List ]

[Hartmut Reuter]

Esther Filderman wrote:
> 
> > | Large            |              | High, but     | Peter Honeyman       |
> >| files/volumes (> |              | difficult     | (UMich).  Note:      |
> >| 2 GB files, > 8  |              |               | Peter will get work  |
> >| GB volumes)      |              |               | related to this      |
> >|                  |              |               | effort from Antonelli|
> >|                  |              |               | & his group at UMich |
> >|                  |              |               | .  Hartmut Reuter    |
> >|                  |              |               | (Max Planck) may also|
> >|                  |              |               | have some related    |
> >|                  |              |               | work.                |
> 
> NOTE:  These are just my words and my take on the situation.   Please DO NOT
> take this as an official statement from the PSC.  Anyone trying to do so, or
> passing this along without this warning, will be hunted down and turned into a
> frog.  Just try me if you think I'm joking!
> 
> Anyway.  The work that has been done at UMich and by Hartmut is all part of, or
> based upon MR-AFS.  Multiple Resident AFS was originally built at the
> Pittsburgh Supercomputing Center.  We still hold the rights to it.
> 
> With the Open-Sourcing of AFS there has been a lot of call [and even some
> screaming and yelling] to Open Source MR-AFS.  Unfortunately, the HigherUps
> here do not want it Open Sourced.  It's now my job to get it properly licensed,
> one way or another.
> 
> On the one hand, the bosses [currently] want some financial restrictions placed
> on the code.  [Mainly, as I see it, I believe that they don't want MR-AFS being
> turned around and marketed as part of a product.]  On the other hand, I've
> pointed out the critical importance of parts of the MR-AFS code, such as the
> larger file sizes, to become part of OpenAFS.
> 
> At the last LISA conference I had a long talk with Hartmut Reuter about MR-AFS.
>  Certainly Hartmut has put in tons of hard work and a few gallons of blood into
> it.  He had some wise advice, which I'm taking.
> 
> Globally, OpenAFS should add a hook so people can use MR-AFS if they desire.
> Then folks who want to use MR-AFS would get a license from us and be able to use it.
> 
> More specifically, I do want to completely free up some areas of the code, such
> as the larger file size stuff, and make it available to add directly into
> OpenAFS.

The original MR-AFS code doesn't provide large file support at all. 
Also, the difficult part of large file support sits in the client side
not in the server side. 

It was policy of MR-AFS to run with the from the shelf Transarc AFS
client without requiring modifications (bypassing the kernel code for
MR-AFS specific functions). So my large file support (presently in beta
test on AIX 4.3.3) is not really part of MR-AFS. However, I wrote it to
make use of MR-AFS features I built into the server code. 

In so far I feel free to contribute the client part of large file
support to OpenAFS without breaking any license agreement with PSC.

I think to include large file support into OpenAFS servers is easy, if
you use the NAMEI-interface instead of abusing inode fields. You then
also have the advantage to be completely independent of fsck and you can
dump /vicep-partitions. (The only problem is to find a place in the
vnode for the upper 32 bits of the file length.)

Hartmut

> 
> By the way, if you need someone to yell at about this, please yell at me.
> Yelling at my higher ups will do nothing but piss them off [in my unhumble
> opinion].  Right now they don't consider MR-AFS a priority.  I, on the other
> hand, think this is very high priority. However, since I will need their
> sign-off to get this completed, it won't help me if you cheese them off.
> 
> Thanks,
> Esther
> 
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo.cgi/openafs-info

-- 
-----------------------------------------------------------------
Hartmut Reuter                           e-mail reuter@rzg.mpg.de
					   phone +49-89-3299-1328
RZG (Rechenzentrum Garching)               fax   +49-89-3299-1301 
Computing Center of the Max-Planck-Gesellschaft (MPG) and the
Institut fuer Plasmaphysik (IPP)
-----------------------------------------------------------------