normal mode and starve mode of sync.Mutex and runtime mutex?
178 views
Skip to first unread message
xie cui
Jan 31, 2021, 10:15:23 AM1/31/21
to golang-nuts
why mutex in runtime do not need starve mode?
jake...@gmail.com
Jan 31, 2021, 11:38:08 AM1/31/21
to golang-nuts
What makes you think it does not? Have you looked at Mutex.lockSlow()? I have not really worked through the algorithm, but the comments do mention starvation mode.
xie cui
Jan 31, 2021, 9:56:55 PM1/31/21
to golang-nuts
mutex in sync package and runtime is different,
mutex in runtime package is here, https://github.com/golang/go/blob/master/src/runtime/lock_futex.go#L46
i do not see there is a starving mode.
robert engels
Jan 31, 2021, 10:07:10 PM1/31/21
to xie cui, golang-nuts
Not sure the runtime needs it - as all lock usages are known - so the users in the runtime can ’negotiate’ fairness through other means for a particular mutex.
--
You received this message because you are subscribed to the Google Groups "golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/golang-nuts/0cc1fae8-6c22-4e40-9645-425d12bb909dn%40googlegroups.com.
Ian Lance Taylor
Feb 1, 2021, 12:22:19 AM2/1/21
to robert engels, xie cui, golang-nuts
On Sun, Jan 31, 2021 at 7:07 PM robert engels <ren...@ix.netcom.com> wrote:
>
> Not sure the runtime needs it - as all lock usages are known - so the users in the runtime can ’negotiate’ fairness through other means for a particular mutex.
Yes. The point of starvation mode in sync.Mutex is to avoid high tail
>
> Not sure the runtime needs it - as all lock usages are known - so the users in the runtime can ’negotiate’ fairness through other means for a particular mutex.
latency for highly contended mutexes, in which one set of goroutines
acquiring the mutex can starve a different set of goroutines waiting
for the mutex. The runtime does not have highly contended mutexes.
When we discover a highly contended mutex in the runtime, we change
the code to use a different approach.
That is, sync.Mutex has to work reasonably well for any Go program.
The runtime.mutex type only has to work well for the runtime package.
Ian
> On Jan 31, 2021, at 8:56 PM, xie cui <cuiw...@gmail.com> wrote:
>
> mutex in sync package and runtime is different,
> mutex in runtime package is here, https://github.com/golang/go/blob/master/src/runtime/lock_futex.go#L46
> i do not see there is a starving mode.
>
> On Monday, February 1, 2021 at 12:38:08 AM UTC+8 jake...@gmail.com wrote:
>>
>> What makes you think it does not? Have you looked at Mutex.lockSlow()? I have not really worked through the algorithm, but the comments do mention starvation mode.
>>
>> On Sunday, January 31, 2021 at 10:15:23 AM UTC-5 cuiw...@gmail.com wrote:
>>>
>>> why mutex in runtime do not need starve mode?
>
>
> --
> You received this message because you are subscribed to the Google Groups "golang-nuts" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/golang-nuts/0cc1fae8-6c22-4e40-9645-425d12bb909dn%40googlegroups.com.
>
>
> --
> You received this message because you are subscribed to the Google Groups "golang-nuts" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages