MySQL Bugs: #107806: AIO systerm optimize

Bug #107806	AIO systerm optimize
Submitted:	8 Jul 2022 3:22	Modified:	12 Aug 2022 11:40
Reporter:	alex xing (OCA)	Email Updates:
Status:	Can't repeat	Impact on me:	None
Category:	MySQL Server: InnoDB storage engine	Severity:	S5 (Performance)
Version:	8.0.19	OS:	Any
Assigned to:		CPU Architecture:	Any

Description:
 For AIO systerm, slot management relies on full scan, which is inefficient.
 In order to locate the slot which is working I/O or idle in the slot array quickly, and reduce the overhead of adding/removing I/O requests to the slot array, it is better to maintain the two type slot separately

How to repeat:
just read the code

Suggested fix:
add std::set<int> slot_set in AIO, which used to manage idle slot

pseudo-code is as follows
void AIO::release(Slot *slot) {
    slot->is_reserved = false;

    int i = slot - at(0);
	//modify here
    slot_set.insert(i);

    --m_n_reserved;

    if (m_n_reserved == m_slots.size() - 1) {
        os_event_set(m_not_full);
    }

    if (m_n_reserved == 0) {
        os_event_set(m_is_empty);
    }
}

dberr_t AIO::init_slots() {
    for (ulint i = 0; i < m_slots.size(); ++i) {
        Slot& slot = m_slots[i];

        slot.pos = static_cast<uint16_t>(i);

        slot.is_reserved = false;
        //modify here
        slot_set.insert(static_cast<int>(i));
    }

    return DB_SUCCESS;
}

Slot* BosonAIO::reserve_slot(){
    //no need to full scan
    auto itup = slot_set.upper_bound(local_seg * slots_per_seg - 1);
    if (itup == slot_set.end()) {
        itup = slot_set.upper_bound(-1);
    }
    ut_a(itup != slot_set.end());
    ulint i = *itup;
    slot = at(i);
    slot_set.erase(itup);
}

Slot* AIO::reserve_slot(){
    //no need to full scan
    auto itup = slot_set.upper_bound(local_seg * slots_per_seg - 1);
    if (itup == slot_set.end()) {
        itup = slot_set.upper_bound(-1);
    }
    ut_a(itup != slot_set.end());
    ulint i = *itup;
    slot = at(i);
    slot_set.erase(itup);
}

Hi Mr. xing,

Thank you for your performance improvement request.

We have analysed latest source code from 8.0.29 and 8.0.30 and your analysis, based on 8.0.19, is outdated.

First of all, we support native AIO on Linux and Windows. We do have added code for the platforms that do not support it.

Hence, you have to be very specific about which code are you actually referring to. Using several slots is native to some of the operating systems.

Also, code has been changed so much, that it seems that your request has been taken care of, already ...

We do require your feedback on our questions ....

This optimization is for imulated-AIO

This optimization is for simulated-AIO

Hi Mr. xing,

We think that this has been addressed in the 8.0.29 and 8.0.30 (yet to be released). Hence, if you could base your analysis on those sources, it would be appreciated.

Also, your code is based on an ancient release.

No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".