New NFS Feature - Looking for Beta Customers

[Dave Olker]

Hi all,

The NFS lab is getting ready to release a new feature for 11i v2 (11.23) and we're looking for interested customers to beta test it for us.

This feature will replace the biod daemons on the NFS client with pools of kernel threads. There are many benefits to such a design:

1) Each NFS mount point will have a separate pool of service threads to handle requests

2) When an NFS server becomes unavailable only the threads associated with the mount points for that server will be affected, not the entire client system

3) Client threads will service different I/O requests in a round-robin fashion. This means no more read starvation (when a reading process and writing process are competing for buffer cache resources the writing process always wins).

Anyone interested in trying this new feature please contact me directly at: dave.olker@hp.com.

Thanks,

Dave

[Geoff Wild]

So, in a SAP Service Gard environment - when say the trans package goes away (NFS package exporting /usr/sap/trans to the landscape) - then the clients will function somewhat normally?

IE - a bdf won't hang?

Rgds...Geoff

[Dave Olker]

Hi Geoff,

> IE - a bdf won't stops working?

I guess it depends on why bdf is stopped working. If it's stops working because it's trying to call statfs() on the mount point and the server is down it will likely stops working.

The difference here is that with biods, if one server goes down there's a good possibility that all the biods will get hung on that dead mount point and *ALL* NFS activity will stop workingto *ALL* mount points, regardless of whether the NFS server is alive and responding for the other mount points. This is where the separate pools of threads per mount point works its magic. With separate pools of threads, if serverA goes down then only the threads associated with mount points from serverA will be affected. Threads servicing requests for mount points associated with serverB and serverC will be unaffected.

Now, separate thread pools won't change the fact that if serverA is down and we have filesystems mounted from serverA and we do a bdf, it will likely stop workingtrying to call statfs() on serverA's filesystems.

One thing to keep in mind - if you don't like bdf stops workingwhen the server goes down you have a couple options:

1) AutoFS - if the filesystem isn't in use for 10 minutes we'll unmount it and then if the server goes down we're not affected

2) Forcible Unmount - if the server goes down we can forcibly unmount the filesystem from that server and then no more stops working bdf commands.

With this in mind, do you have any interest in testing this new feature?

Dave

[John Payne_2]

I'm hooked into it already, but I havent' had time to get the installation this week. Will they be upset if it's another week before I can get to work on it?

John

[Dave Olker]

> Will they be upset if it's another week
> before I can get to work on it?

Horribly...but we'll live. :)

Seriously, thank you very much for participating in this beta program. Any feedback you can provide will be greatly appreciated by myself and the NFS lab.

Regards,

Dave

[Dave Olker]

- bump -

Still looking for any interested customers...

[A. Clay Stephenson]

Hi Dave,

I just emailed you. I can load this stuff on my 11.23 sandbox.

[Dave Olker]

Hi Clay,

Beautiful. I'll get the bits sent to you.

Dave