Network and Storage Protocols

Automatic Node Referrals and Regarding DNS Entry


Hi all,


We have FAS 8020 , Created SVM for CIFS  and added 2 LIFS . One LIF is created on Node SA6UNS05:e0e and another LIF created on Node SA6UNS06 : e0e .It means 2 Data LIFS with Ip adresess for   SVM  .


Need to Configure DNS for  Data LIF in DNS server  so In this scenario  Do we need to add only  one Data LIF  or  two DATA LIFS to be added ?

If One Data LIF to be add then How it will be resolve thru second Data LIF ?

If 2 DATA LIFS to be added then how to create 2 Entries for single host name for 2 2 DATA LIF IP's




Need to implement  CIFS feature Automatic Node refferals with option  " Is referrals entry in CIFS Options : true "  ,  what are the pre requisite steps for this to set option TRUE




SA6CLS02::> net port show
(network port show)
Speed (Mbps)
Node Port IPspace Broadcast Domain Link MTU Admin/Oper
------ --------- ------------ ---------------- ----- ------- ------------
e0M Default up 1500 auto/1000
e0a Cluster Cluster up 9000 auto/10000
e0b Cluster Cluster up 9000 auto/10000
e0e Default up 1500 auto/1000
e0f Default up 1500 auto/1000
e0M Default up 1500 auto/1000
e0a Cluster Cluster up 9000 auto/10000
e0b Cluster Cluster up 9000 auto/10000
e0e Default up 1500 auto/1000
e0f Default up 1500 auto/1000
10 entries were displayed.





Srikanth Manchana



Your question raises two separate but related topics.


First - DNS.  When you have two LIFs for a single SVM, each on a separate node, you generally want to balance the workload for each node.  One way to do this is through DNS.  The basics are to setup two DNS records for the same SVM name, each pointing to one of the LIFs.  You depend on DNS to hand out both addresses in response to queries.  This tends to spread the client connection requests somewhat evenly.


The node that gets the request does all the CIFS work, whether the volume is local to that node or not.  The actual data work to disk (reads/writes) will traverse the cluster backplane if needed.


There are multiple ways to configure DNS - statically, dynamically, etc. - use whatever works best for you but I highly recommend putting both LIFs into DNS to resolve to the SVM name.


CIFS node referral is different and unrelated to DNS.  CIFS referrals does not depend on anything in DNS to work.  What happens when CIFS referrals are active in an SVM is that when a CIFS request arrives at a node, the node determines whether the target volume of the share is located on the same node.   If it is, nothing special happens and the request proceeds normally.  If it isn't, the first node will send a DFS-like referral back to the client using the IP address for a LIF on the node that does hold the target volume.  The client then starts a CIFS session to the other node based on IP address, not name lookup.  So DNS setup, beyond perhaps the original connection, is not a factor in CIFS referrals.


Some caveats on CIFS referrals:


1.  The referral only happens once based on the volume to which the original share is on.  If somewhere lower in the namespace you have linked a different volume within cDot, if that volume is again on a different node no future referral will take place.


2.  Security can be impacted be a referral.  The UNC reference by name is not the same as the UNC reference by IP with certain levels of SMB protocol and security options enabled.  There are KB articles at that reference subtle differences in certain circumstances.  This is not likely to happen unless you have specific security options enabled in your overall environment.  I mention it only because if you do have such options, it could be a nightmare tracking down why a referral isn't working.  I apologize that I don't have a direct reference off hand - just something I remember reading in KB's in the past few months.



I hope this helps you.


Bob Greenwald

Lead Storage Engineer

Consilio, LLC


Kudos and accepted solutions are always appreciated.



I have been reading up on both Automatic Node Referrals and on-box-DNS.  Since you mentioned they are 2 separate topics can they both be implemented for the same cluster?