"Page/swap file: use at most" setting can't be set to 0

[adamradocz]

Hi,

The "Page/swap file: use at most" setting must be between 1 and 100 and the 0 value is not allowed. Some cases I don't want to allow BOINC to use page/swap file and I can't set 0 value. It doesn't make sense especially, when I don't have page/swap file.

[ChristianBeer]

Is this setting even used at all? And if yes what effect does it have?

[adamradocz]

Deleting this feature and let the OS do its job is perfectly make sense to me. Plus it'd simplify the manager.

[davidpanderson]

Each task has an address space of a certain size. Part of this - the "working set" - is kept in RAM. The rest is kept on disk, in an area called swap space. If the sum of the address spaces exceeds swap space, the OS kills processes and/or dynamically expands the swap space. We don't want BOINC to cause this.

The client keeps track of each task's size, and there is logic that tries to avoid exhausting swap space. Currently this logic isn't correct. It should be fixed. The feature should be kept.

[SETIguy]

We should also be sure the behavior is correct on machines without swap.

…

On Tue, Sep 26, 2017 at 11:27 PM, David Anderson ***@***.***> wrote: Each task has an address space of a certain size. Part of this - the "working set" - is kept in RAM. The rest is kept on disk, in an area called swap space. If the sum of the address spaces exceeds swap space, the OS kills processes and/or dynamically expands the swap space. We don't want BOINC to cause this. The client keeps track of each task's size, and there is logic that tries to avoid exhausting swap space. Currently this logic isn't correct. It should be fixed. The feature should be kept. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1880 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKXcsldpUpikv7DhmRj15AnU-q17NXFzks5smerSgaJpZM4NFXKk> .

-- Eric Korpela korpela@ssl.berkeley.edu AST:7731^29u18e3

[davidpanderson]

In that case swap-size detection would return 0 and the logic would be skipped.
But like I say the current code is wrong, and probably a no-op.
See cpu_sched.cpp.