Re: [PATCH v2] sched/fair: Replace random newidle_balance with Bresenham accumulator

[Christian Loehle]

On 10/27/26 09:31, Qing Wang wrote:
> Hi Peter,
> Gentle ping on this. I know you're busy, but I'd really appreciate your thoughts.
> 
>> I took the most sensitive workload I have for newidle balance (tbench)
>> and took it for a spin with these changes. Following are the results:
>>
>>     Clients:        tip                  bresenham_accm
>>         1      321.65 (0.00 pct)       313.97 (-2.38 pct)
>>         2      641.92 (0.00 pct)       638.74 (-0.49 pct)
>>         4     1245.65 (0.00 pct)      1237.26 (-0.67 pct)
>>         8     2435.80 (0.00 pct)      2442.23 ( 0.26 pct)
>>        16     4717.66 (0.00 pct)      4688.19 (-0.62 pct)
>>        32     9303.53 (0.00 pct)      9390.71 ( 0.93 pct)
>>        64    18002.57 (0.00 pct)     17911.56 (-0.50 pct)
>>       128    27729.26 (0.00 pct)     27621.95 (-0.38 pct)
>>       256    47134.77 (0.00 pct)     46137.36 (-2.11 pct)
>>       512    43179.41 (0.00 pct)     43277.53 ( 0.22 pct)
>>      1024    40339.30 (0.00 pct)     40176.49 (-0.40 pct)
>>
>> The %diff is in noise range which is a good indication that there
>> shouldn't be any surprises. I'll queue a run overnight to see if
>> there are other benchmarks that like / dislike these changes.
> 
> This benchmarks looks good from Prateek.
> Link: https://lore.kernel.org/all/da1896b3-6bec-4b56-bf6d-9ce775b00b8d@xxxxxxx/
> Cc: K Prateek Nayak <kprateek.nayak@xxxxxxx>

Might make sense to resubmit the patch as standalone.
Also minor suggestions inline:

+				 * Base on accumulator of Bresenham's line algorithm
+				 *
+				 * Accumulate ratio each time and trigger balance
+				 * when sum ≥ 1024.
>= Is more common I guess
+				 *
+				 * Ensures exactly ratio balances per 1024 calls
Ensures the same ratio balances per 1024 invocations
+				 * and distributes ratio balance operations uniformly
+				 * across every 1024 invocations.
across them.
 				 */