Success!
A combination of RSS disable, and the -a option on the pfcount command line, is
consistently yielding 100% packet reception at 10 Gbits/second.
Disregard my previous comment about -a -- this has to be present always.
It wasn't obvious to me that RSS and ZC shouldn't be used together, and I'd
like to understand why not if possible.
Regards,
A.
--
Andrew Howard
[email protected]
----- Reply message -----
From: "Andrew Howard" <[email protected]>
To: <[email protected]>
Subject: [Ntop-misc] pfcount stats on ZC interface not as expected
Date: Tue, Aug 18, 2015 09:54
Apologies for the badly formatted message.
I'm posting from an HTC One and its proving problematic.
Regards,
A.
----- Reply message -----
From: "Andrew Howard" <[email protected]>
To: <[email protected]>
Subject: Re: [Ntop-misc] pfcount stats on ZC interface not as expected
Date: Tue, Aug 18, 2015 09:47
Hi Alfredo,
> please make sure you disabled multiqueue (RSS=1,1,1,1), please provide cat
> /proc/net/pf_ring/dev/eth3/info
The output from the requested cat command is as follows:-
Name: eth3
Index: 186
Address: 8C:DC:D4:B4:F5:61
Polling Mode: NAPI/ZC
Type: Ethernet
Family: Intel ixgbe 82599
Max # TX Queues: 1
# Used RX Queues: 1
Num RX Slots: 32768
Num TX Slots: 32768
Why do we specify four numbers on the RSS option, and not RSS=1,1 ?
* POSSIBLE PROGRESS *
After a fresh ./load_driver.sh command, I’m unable to get ‘pfcount –I zc:eth3’
to receive all packets as previously described. However, if I run this command
ONCE with the –a option, every run of ‘pfcount –I zc:eth3’ thereafter with or
without the –a option appears to receive 100% packets or very close to it.
I’m now running more tests to explore this observation, and looking at code to
understand what is occurring.
Regards,A.
--
Andrew Howard
[email protected]
_______________________________________________
Ntop-misc mailing list
[email protected]
http://listgateway.unipi.it/mailman/listinfo/ntop-misc