This post is more than 5 years old
64 Posts
0
7655
SmartConnect Advanced config for SMB
Hi there.
I had a little discussion about configuring the Isilon for SMB access.
What i learned from trainings and white papers is, that i put 1 static IP per node/interface. Config Done!
The other opinion was to put 5-20 IPs per node and set SmartConnect Advanced to "Dynamic".
As far as i know, "dynamic" is bad for SMB and good for NFS.
So, in the end i am confused and wanted to ask you about your opinions/experience on that.
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
July 3rd, 2014 08:00
Alex,
that is correct, you want to use different pools for different protocols. Here is an example from my cluster
I have two pools in my subnet, one is used for NFS clients (dynamic IP allocation) and one is used for CIFS clients (static IP allocation)
Does it help ?
zerothehero
64 Posts
0
July 4th, 2014 00:00
What would happen if i run SMB with the "Dynamic" option?
Of course every client connected to a failing node would lose its connection. In this case, there should be no difference between Static and Dynamic.
But what about the "failback", when the IPs get redistributed?
Will the IP stay on a node as long as there is an active connection? Or will it simply failback to some node and the client loses his connection again?
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
1
July 4th, 2014 06:00
nothing will happen until you try to "rebalance" a dynamic pool or move IP addresses around (for example way too many sessions connected to one node or you want to move some NFS connection pro-actively). CIFS clients will get disconnected, NFS clients will continue to be connected. When i first configured my cluster 4 years ago i thought "well I would never rebalance in my environment so i don't need to setup a dedicated pool for CIFS clients". I finally decided to go ahead and set it up anyway, that's why you see i have two pools, one for NFS and one for CIFS. I am glad i set it up this way not because i was worried about CIFS clients being disconnected but because of dealing with Isilon support. If you ever have any issues with CIFS clients getting disconnected, i can guarantee you 100% that support will look at your cluster and say "well you are not using static IP allocation for your CIFS clients, that's that problem". It probably has nothing to do with your problem but they will latch on to that and won't do anything else ..they will continue to blame that configuration for all your problems, bad market, horrible weather outside. Trust me i have been down that road with another customer, set it up per best practices from the get go.
there is no failover/failback for IP in static pool, that IP address is not available until that node come back online. For dynamic pool, when node becomes available again (let's say you were doing OneFS upgrade and node rebooted), IP address will failback to that node and not necessarily the same IP address that were on that node originally.
swi1
33 Posts
0
February 5th, 2015 12:00
Hi Dynamox,
How did you use the interfaces for static and dynamic? Is it one static and one dynamic on two interfaces of a node?
Thanks
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
February 5th, 2015 14:00
i am not sure i follow you. An aggregated interface can be used in multiple subnets (via VLAN tagging) and multiple IP Pools.
swi1
33 Posts
0
February 5th, 2015 14:00
That's interesting, can we configure the aggregated ports for static as well as dynamic? Does it make any change or impact on work function of static and dynamic?
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
February 5th, 2015 14:00
I am using 2 x 10G per node in an LACP configuration. The same interface (10gige-agg-1) is member of static and dynamic pools.
osaddict
110 Posts
4
February 5th, 2015 15:00
NFS clients would move to an different node because the IP would move and the stateless nature allows the client to move without issue.
Warning, this part gets a little long:
The CIFS clients attached to that node/interface would get no response from that IP since it is down. Assuming a Windows client, the client itself will timeout the connecting and do a DNS lookup again for the SmartConnect Zone name i.e. storage.isilon.com or something like that. Then the DNS request will go to the DNS server in the environment, which will be forwarded to the DNS server running on the cluster's SmartConnect Service IP, it will then return one of the IPs for the nodes and interfaces that are still on line. That will then be forwarded to the DNS server, then the client. The client with then reconnect with a new session different node. This all happens in about 30-60 seconds. The only download side is that any data inflight from the client will be lost. However, Windows bubbles that message up the application layer well. The end user will get a message like, "Word failed to save file". Then the user can try again and it will work because the reconnect generally happens very quickly.
Hope that helps. A lot of these details are on this whitepaper.
http://www.emc.com/collateral/hardware/white-papers/h8316-wp-smartconnect.pdf
swi1
33 Posts
0
February 5th, 2015 15:00
Perfect. I gotcha, Thanks
swi1
33 Posts
0
February 5th, 2015 15:00
Sorry for the confusion. If a node fails, dynamic pool moves the ip to other nodes and static remains on the interface.So my question is - How it works in your configuration, since you are using the same interfaces for both pools(dmz_3039_ipool, dmz_3039_Cipool)?