[lttng-dev] [PATCH 14/16] urcu-qsbr: batch concurrent synchronize_rcu()

[Lai Jiangshan]

On 11/23/2012 02:30 AM, Paul E. McKenney wrote:
> On Thu, Nov 22, 2012 at 05:04:47PM +0800, Lai Jiangshan wrote:
>> On 11/22/2012 02:33 AM, Mathieu Desnoyers wrote:
>>> * Lai Jiangshan (laijs at cn.fujitsu.com) wrote:
>>>> Could you delay 14~16 for 40 days if I don't implement it in 40 days?
>>>
>>> I'm curious to know more about the changes you are planning. Is that
>>> another way to implement grace periods that would allow multiple threads
>>> to execute synchronize_rcu() concurrently ?
>>
>> synchronize_rcu()s in this implement share coarse-grain step(1 GP)
>> to achieve concurrence. My implement will use fine-grain step(1 check or 1 flip)
>> like SRCU. and call_rcu() is also considered in this implement to avoid
>> unneeded wait.
>>
>>>
>>> Please note that changes in these algorithms will need to go through
>>> very strict review/validation/verification. So I expect that if it takes
>>> 40 days to implement, we can plan at least 3-4 months of validation work.
>>
>> I means I don't have time. If I can't steal some time from the late 40 days,
>> this code is OK for me.
> 
> Why don't we take the current code, which would allow some academic
> projects to test on large systems in the next few months, and then
> replace it with your code when available and if appropriate?
> 


Agreed.

thanks,
Lai