UNSOLVED
P
philippspohr
1 Rookie
•
107 Posts
0
3678
January 15th, 2015 08:00
SmartConnect DNS Cache / query time threshold
Hello,
When a client starts a FTP connection to an Isilon cluster through SmartConnect the following is happening:
- Establish the connection: Client > DNS-Server > Isilon SmartConnect with round robin > Node IP address
- Client <> Node IP address
But what happens if the client starts a lot of FTP sessions in a very short time?
If DNS caching is completely disabled at the client side I would suppose that the client gets for every connection an different IP address from the SmartConnect service.
But we see in an installation that the answers from the SmartConnect service are coming in blocks. After a few same IP addresses the client gets an different IP address for the connections. Because of this behaviour there is no real IP/Node distribution in some kind of workflows.
How long must the DNS requests of the client apart to make SmartConnect work properly? Is there a certain threshold?
Peter_Sero
6 Operator
•
1169 Posts
2429
0
Posted January 15th, 2015 10:00
Yep... remembering the old thread
It's a known bug from Windows DNS delegation. The cache is 1 sec and cannot be remove with ttl=0. You can setup a linux DNS to bypass that bug. At this time, we didn't find a better solution with fast multiple connection to a Isilon cluster.
philippspohr
1 Rookie
•
107 Posts
2429
0
Posted January 16th, 2015 01:00
Ah, thank you very much. My mistake.
I thought the Windows DNS is no longer involved on the way back. And I did not know the 'bug' with the lowest TTL of 1s. I suspected the Windows DNS is configured with a TTL of 0.
What other solutions are possible without setting up an Linux DNS?
Can I configure the Client (Windows) to use the SmartConnect DNS only for a few hostnames (= the SmartConnect Names) and for all other name resolutions the normal Windows DNS? The rule would look like:
- all addresses except xxx.isilon.xxx >> ask Windows DNS
- xxx.isilon.xxx >>> ask Isilon SmartConnect DNS
Peter_Sero
6 Operator
•
1169 Posts
2429
1
Posted January 15th, 2015 09:00
From the Isilon's side, the round-robin is "perfect", each query returns a different IP, with a TTL 0 = supposed to be uncachable.
It's those DNS servers in the middle, and some clients (OS, but also down to the apps) that might cache even "uncachable" replies, at least for one second or so. Check the docs for the systems involved at your site, in particular for Windows Server DNS...
hth
-- Peter