Thanks Karl,

always nice to hear from you and get detailed insight on your usage of the system.

Rainer

Since no one has chimed in, I'll comment from my own experience.  Your setup is almost identical to ours - an NS80 primary with an NS20 secondary for DR.  We use Replicator v2 to keep a DR copy in sync on the NS20.

We're using the Celerra Home Directory service, and our lab workstations use Roaming Profiles and Folder Redirection to provide a user's desktop and files on any workstation they log into.

In our environment, we have approximately 9000 users.  Since we don't differentiate between users, we provide just one CIFS server on one VDM for all users (i.e. \\STORAGE\HOME on vdm_10).  I've seen no evidence of performance impact from this approach, so we've had no reason to change it.  We have a total of 12 VDMs for various purposes, but only one for home directories.

However, we have a number of filesystems for home dirs - 8 in total, each with about 1200 users per filesystem.  Each filesystem server one chunk of the alphabet (A-F, G-K, etc.); this makes it easy to have the home dir mapping file redirect users by the first letter of their login name - the mapping file has just 10 lines.  We use tree quotas to create each user's directory, and the original limit of 1024 tree quotas per filesystem was the limiting factor.

Given the size of your WAN link, you'll probably have to experiment a bit to right-size your filesystems.  Each of our filesystems are about 1.3TB in size; we replicate across a 10GbE backbone and the initial replication took several hours each.  After that, we've never had an issue with replications being slow.

Considering that 100Mbs is roughly 12MB/s, you'll have to decide if you can "start small" or "start behind".  By start small, I mean, create very small filesystems - say, 100GB - and start using them.  Turn on replicator and get the first 100GB caught up on the destination NAS.  Then, extend the filesystem by another 100GB and put more data there.  The destination will extend and replicate the changes.  Keep growing in small increments until you have your full data payload.  This is good, if you have an SLA that mandates that no data will be out of sync for long.

Or, you can start behind - expect that data will be out of sync on the target and just let it catch up over time.  This is more time-consuming by far - it could take weeks to replicate many TB of data to the target.

Back to comments on right-sizing, VDMs don't add much overhead to the performance of your NAS.  If you have logical divisions of users, or mandates that business units have their own resources, feel free to make whichever number of VDMs make sense.  The same holds true for CIFS servers as well.  Dividing up users, shares and CIFS servers across VDMs allows you to break down your replications into smaller and smaller chunks.  For example, if your business has the groups Marketing, Engineering and Shipping, make one VDM for each group.  Then, create whichever CIFS server(s) are needed by that group.  Then, your point of replication for that group becomes that VDM and its associated filesystems.  With your WAN link, you could reliably add your first VDM and CIFS servers, get it replicated fully, then move to your next one and so on.

However, if you don't have clear divisions, you're probably safe to do a small number of VDMs and CIFS servers - you'll want to experiment and see what fits.

Let me know if this helps!

Karl

My pleasure!  I, too, wish there was a best-practices document for VDMs, CIFS servers and filesystem counts.  There are clear cases when you can have too many files in a filesystem (trust me - 12M JPGs at 250kb each is always a bad idea!), but everything else is pretty vague.  A best practices document would certainly help.  There's a nice collection of articles and comments in the forums, but something "official" would sure be nice.

I don't think anyone is unhappy with the Clariion best practices documents - they're essential to getting great performance from your CX.  A similar document for Celerra and VNX would be no-less welcome.

Thanks - and it was great to see you at EMCWorld!

Karl

Hi Karl,

It is truly an honor to have people like you within the community. I have been researching and testing about this for weeks without a success. You just made my evening now.Thank you very much.

That said, one of the qualities of a good infrastructure product is the availability of best practices in addition to its step by step procedures, which in this case the EMC Celerra lacks. With their great product, I hope the Celerra team at EMC will work on it in the future.

Cheers,

Ashenafi

It's always a pleasure to help - this community has given much to me, and I'm happy to be able to contribute in whatever why I can!

Let's break this down into smaller chunks, and attack the pieces you still need help/advice with.  List out what you'd like to change, and we can talk about what options are available.  With a little conversation, we can get other users to chime in with their experiences and get some good ideas flowing for you.  Other users here have done things very differently from me, and with good reason: not every idea will work with everyone's environment - especially things that I do in mine!

Thanks!

Karl