Re: [PATCH v2] l2tp: use per-cpu variables for u64_stats updates

[Rick Jones <rick.jones2@hp.com>]

On 06/27/2012 02:35 PM, Eric Dumazet wrote:
> On Wed, 2012-06-27 at 14:31 -0700, Ben Greear wrote:
>
>> For an example, see the VLAN code. rx-errors and tx-dropped are only 32-bit
>> counters.  Now, in the real world, we wouldn't expect those counters to
>> increase at high rates, but they are still 32-bit counters masquerading
>> as 64, and they could wrap after a while, so any code that expected a wrap
>> to mean a 64-bit wrap would be wrong.
>>
>> At the time I was last complaining, there were lots more cases
>> of this that I was fighting with, but I don't recall exactly what they
>> were.  Once my user-space code got paranoid enough, it was able to
>> at least mostly deal with 32 and 64 wraps.
>
> Good, you now know how to deal correctly with these things.
>
> Using 64bit fields for tx_dropped is what I call kernel bloat.

Today, sure, generalizing to packet counters in general, that bloat is 
likely on its way.  At 100 Gbit/s Ethernet, that is upwards of 147 
million packets per second each way.  At 1 GbE it is 125 million octets 
per second.  So, if 32 bit octet counters were insufficient for 1 GbE, 
32 bit packet counters likely will be insufficient for 100GbE.

Or, I suppose, 3 or more bonded 40 GbEs or 10 or more bonded 10 GbEs 
(unlikely though that last one may be) assuming there is stats 
aggregation in the bond interface.

rick jones