11 Replies Latest reply on Jun 16, 2010 10:08 AM by 807557

    7310 - Problem with CIFS and Roaming Profiles since upgrade to 2010.Q1.0.2

    807557
      We have developed a strange problem with our environment which I'm pretty sure is down to the upgrade to 2010.Q1.0.2 from 2009.Q3.4.1 on our 7310 (all the previous 2009 releases had been fine) since nothing else has changed in our environment. I suspect some changes to the underlying CIFS server causing this?

      We have virtual Windows servers hosted on a VMWare VSphere cluster which are stored on the 7310 via iSCSI LUNs and also CIFS shares on the 7310 for home directories and separate CIFS shares for roaming profiles - all paths are correct in AD for each user - we also use folder redirection for XP Pro clients to force things like "Application Data", "My Documents" etc. onto the Homedir share.

      What we've been seeing recently (which only started happening after the upgrade) is a lot of failed logons to the domain for users. It looks like the usual corrupted profile problem that has plagued Windows forever ...the usual messages that it cannot log the user on with a copy of their roaming profile, and that it will use a temporary one. Some folder redirections (that are initiated via Group Policy) also don't get applied. Users don't see errors when logging off from a "good" profile, and NTUSER.DAT etc. seemingly gets written correctly - the next time they log on, around half the time the users will get these errors as described below:

      Event viewer logs show "cannot find the file specified" errors for NTUSER.DAT, along with "directory name is invalid" errors for some of the folder redirections.

      More worrying (and what I think might be the real reason for these failures) are the "offline caching is enabled on the roaming profile share" errors. I think that the client-side caching might not be working - possibly the profiles aren't getting flushed and written correctly upon logout?

      Now, unfortunately the MMC snap-in for managing shares doesn't seem to support changing the behaviour for client-side caching on the CIFS shares (as confirmed in the latest 7000-series Admin Guide on page 198).

      I've been thinking about unchecking the "Enable Oplocks" box which from the CIFS side would completely stop all client-side caching I presume?

      Is this likely to be the culprit here, or is there any other known behaviour that could be causing these errors? Is it also worth disabling "Cache device usage" altogether for the Profiles share itself?

      Can anyone help? It's a bit of a strange problem, and something I don't want to raise with Sun on our support contract just yet, since at first glance it looks like a Windows problem, but I suspect the storage could well be to blame...