Current TabRanker 
https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager.cc?rcl=036350672c7cc81d99da59975900b7a07331aaea&l=276

either rerank all tabs (Enabled) or none (Disabled).

But the TabManager only discards one tab very time.

So there could be the a potential problem:

(1) TabRanker rankers the tab better averagely  among first 100 tabs.

(2) TabRanker make mistakes sometimes, so the first one can be very bad.

(3) TabManager always select the first one returned by TabRanker.

A walk around is to only rerank top-n tabs returned by MRU; or use TabRanker as a correction of MRU.

 

Current TabRanker 
https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager.cc?rcl=036350672c7cc81d99da59975900b7a07331aaea&l=276

either rerank all tabs (Enabled) or none (Disabled).

But the TabManager only discards one tab very time.

So there could be the a potential problem:

(1) TabRanker rankers the tab better averagely  among first 100 tabs.

(2) TabRanker make mistakes sometimes, so the first one can be very bad.

(3) TabManager always select the first one returned by TabRanker.

A walk around is:

(1) Use LRUIndex = MRU.total - MRU.index inside TabActivityWatcher::WebContentsData to denote how old a tab is (MRU.total for the latest, and 1 for the oldest.)

(2) Only score tabs with LRUIndex <=N; return base::nullopt for the rest, which will be assigned SortKey::kMaxScore as its score eventually.

(3) When sorted by TabLifecycleUnitSource::TabLifecycleUnit::GetSortKey():
    The oldest N tabs will be sorted by its score; 
    The rest will be sorted by last_focused_time since they all have same score.
    The oldest N tabs is always in front of the rest.

This the result we want to experiment.  



 

Current TabRanker 
https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager.cc?rcl=036350672c7cc81d99da59975900b7a07331aaea&l=276

either rerank all tabs (Enabled) or none (Disabled).

But the TabManager only discards one tab very time.

So there could be the a potential problem:

Let us say the user has opened 1000 tabs. He is in memory pressure.
Suppose: 
(1) Tabs are sorted by least recent used as: 
    tab_0, tab_1, tab_2, ..., tab_1000
(2) Tabs are sorted by TabRanker score as:
    tab_1000, tab_20, tab_22, tab_23, tab_24, ...
(3) The user eventually reload page tab_1000, tab_2, tab_4, tab_5, and then 
    closed the browser.

In the case above, if TabRanker have averagely better result if TabManager discards 5 tabs every time.

But currently TabManager discards one tab very time (which is meant to be), then there is a chance the bad first one makes the TabRanker result much worse.

Currently it's hard to verify this unless we reconstruct a users' full tab list.

But it can be very helpful if we have the options to set how many tabs to rerank.


 






(1) TabRanker rankers the tab better averagely among first 100 tabs.

(2) In some extreme case, TabRanker make mistakes sometimes, gets very low score occasionally.

(3) TabManager always select the first one returned by TabRanker.

A walk around is:

(1) Use LRUIndex = MRU.total - MRU.index inside TabActivityWatcher::WebContentsData to denote how old a tab is (MRU.total for the latest, and 1 for the oldest.)

(2) Only score tabs with LRUIndex <=N; return base::nullopt for the rest, which will be assigned SortKey::kMaxScore as its score eventually.

(3) When sorted by TabLifecycleUnitSource::TabLifecycleUnit::GetSortKey():
    The oldest N tabs will be sorted by its score; 
    The rest will be sorted by last_focused_time since they all have same score.
    The oldest N tabs is always in front of the rest.

This the result we want to experiment.  



 

Current TabRanker 
https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager.cc?rcl=036350672c7cc81d99da59975900b7a07331aaea&l=276

either rerank all tabs (Enabled) or none (Disabled).

But the TabManager only discards one tab very time.

So there could be the a potential problem:

Let us say the user has opened 1000 tabs. He is in memory pressure.
Suppose: 
(1) Tabs are sorted by least recent used as: 
    tab_0, tab_1, tab_2, ..., tab_1000
(2) Tabs are sorted by TabRanker score as:
    tab_1000, tab_20, tab_22, tab_23, tab_24, ...
(3) The user eventually reload page tab_1000, tab_2, tab_4, tab_5, and then 
    closed the browser.

In the case above, if TabRanker have averagely better result if TabManager discards 5 tabs every time.

But currently TabManager discards one tab very time (which is meant to be), then there is a chance the bad first one makes the TabRanker result much worse.

Currently it's hard to verify this unless we reconstruct a users' full tab list.

But it can be very helpful if we have the options to set how many tabs to rerank.



 

Current TabRanker 
https://cs.chromium.org/chromium/src/chrome/browser/resource_coordinator/tab_manager.cc?rcl=036350672c7cc81d99da59975900b7a07331aaea&l=276

either rerank all tabs (Enabled) or none (Disabled).

But the TabManager only discards one tab very time.

So there could be the a potential problem:

Let us say the user has opened 1000 tabs. He is in memory pressure.
Suppose: 
(1) Tabs are sorted by least recent used as: 
    tab_0, tab_1, tab_2, ..., tab_1000
(2) Tabs are sorted by TabRanker score as:
    tab_1000, tab_20, tab_22, tab_23, tab_24, ...
(3) The user eventually reload page tab_1000, tab_2, tab_4, tab_5, and then 
    closed the browser.

In the case above, TabRanker have averagely better result if TabManager discards 5 tabs every time.

But currently TabManager discards one tab very time (which is meant to be), then there is a chance the bad first one makes the TabRanker result much worse.

Currently it's hard to verify this unless we reconstruct a users' full tab list.

But it can be very helpful if we have the options to set how many tabs to rerank.