Received-SPF: pass (google.com: domain of linux-kernel+bounces-24252-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Message-ID: <e117f6e0-a8bc-4068-8bce-65a7c4e129cf@kernel.dk>
Date: Thu, 11 Jan 2024 19:58:12 -0700
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v6] io_uring: Statistics of the true utilization of sq
 threads.
Content-Language: en-US
To: Xiaobing Li <xiaobing.li@samsung.com>
Cc: asml.silence@gmail.com, linux-kernel@vger.kernel.org,
 io-uring@vger.kernel.org, kun.dou@samsung.com, peiwei.li@samsung.com,
 joshi.k@samsung.com, kundan.kumar@samsung.com, wenwen.chen@samsung.com,
 ruyi.zhang@samsung.com
References: <b0c67327-5131-4cde-a8bd-df69b1f300e5@kernel.dk>
 <CGME20240112012013epcas5p38c70493069fb14da02befcf25e604bc1@epcas5p3.samsung.com>
 <20240112011202.1705067-1-xiaobing.li@samsung.com>
From: Jens Axboe <axboe@kernel.dk>
In-Reply-To: <20240112011202.1705067-1-xiaobing.li@samsung.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

On 1/11/24 6:12 PM, Xiaobing Li wrote:
> On 1/10/24 16:15 AM, Jens Axboe wrote:
>> On 1/10/24 2:05 AM, Xiaobing Li wrote:
>>> On 1/5/24 04:02 AM, Pavel Begunkov wrote:
>>>> On 1/3/24 05:49, Xiaobing Li wrote:
>>>>> On 12/30/23 9:27 AM, Pavel Begunkov wrote:
>>>>>> Why it uses jiffies instead of some task run time?
>>>>>> Consequently, why it's fine to account irq time and other
>>>>>> preemption? (hint, it's not)
>>>>>>
>>>>>> Why it can't be done with userspace and/or bpf? Why
>>>>>> can't it be estimated by checking and tracking
>>>>>> IORING_SQ_NEED_WAKEUP in userspace?
>>>>>>
>>>>>> What's the use case in particular? Considering that
>>>>>> one of the previous revisions was uapi-less, something
>>>>>> is really fishy here. Again, it's a procfs file nobody
>>>>>> but a few would want to parse to use the feature.
>>>>>>
>>>>>> Why it just keeps aggregating stats for the whole
>>>>>> life time of the ring? If the workload changes,
>>>>>> that would either totally screw the stats or would make
>>>>>> it too inert to be useful. That's especially relevant
>>>>>> for long running (days) processes. There should be a
>>>>>> way to reset it so it starts counting anew.
>>>>>
>>>>> Hi, Jens and Pavel,
>>>>> I carefully read the questions you raised.
>>>>> First of all, as to why I use jiffies to statistics time, it
>>>>> is because I have done some performance tests and found that
>>>>> using jiffies has a relatively smaller loss of performance
>>>>> than using task run time. Of course, using task run time is
>>>>
>>>> How does taking a measure for task runtime looks like? I expect it to
>>>> be a simple read of a variable inside task_struct, maybe with READ_ONCE,
>>>> in which case the overhead shouldn't be realistically measurable. Does
>>>> it need locking?
>>>
>>> The task runtime I am talking about is similar to this:
>>> start = get_system_time(current);
>>> do_io_part();
>>> sq->total_time += get_system_time(current) - start;
>>
>> Not sure what get_system_time() is, don't see that anywhere.
>>
>>> Currently, it is not possible to obtain the execution time of a piece of 
>>> code by a simple read of a variable inside task_struct. 
>>> Or do you have any good ideas?
>>
>> I must be missing something, because it seems like all you need is to
>> read task->stime? You could possible even make do with just logging busy
>> loop time, as getrusage(RUSAGE_THREAD, &stat) from userspace would then
>> give you the total time.
>>
>> stat.ru_stime would then be the total time, the thread ran, and
>> 1 - (above_busy_stime / stat.ru_stime) would give you the time the
>> percentage of time the thread ran and did useful work (eg not busy
>> looping.
> 
> getrusage can indeed get the total time of the thread, but this
> introduces an extra function call, which is relatively more
> complicated than defining a variable. In fact, recording the total
> time of the loop and the time of processing the IO part can achieve
> our observation purpose. Recording only two variables will have less
> impact on the existing performance, so why not  choose a simpler and
> effective method.

I'm not opposed to exposing both of them, it does make the API simpler.
If we can call it an API... I think the main point was using task->stime
for it rather than jiffies etc.

-- 
Jens Axboe