>Whether msync/fsync can make data persistent depends on ADR feature on
>memory controller, if it exists everything works well, otherwise, we need
>to have another interface that is why 'Flush hint table' in ACPI comes
>in. 'Flush hint table' is particularly useful for nvdimm virtualization if
>we
>Whether msync/fsync can make data persistent depends on ADR feature on
>memory controller, if it exists everything works well, otherwise, we need
>to have another interface that is why 'Flush hint table' in ACPI comes
>in. 'Flush hint table' is particularly useful for nvdimm virtualization if
>we
>> Platforms supporting NVDIMMs are now required to provide persistence
>> guarantees once pmem stores are accepted by the memory subsystem.
>
>Can you point us to a precise definition of what exactly constitutes
>stores being "accepted by the memory subsystem"? Back when pcommit
>was a thing
>> Platforms supporting NVDIMMs are now required to provide persistence
>> guarantees once pmem stores are accepted by the memory subsystem.
>
>Can you point us to a precise definition of what exactly constitutes
>stores being "accepted by the memory subsystem"? Back when pcommit
>was a thing
>
>And when the filesystem says no because the fs devs don't want to
>have to deal with broken apps because app devs learn that "this is a
>go fast knob" and data integrity be damned? It's "fsync is slow so I
>won't use it" all over again...
...
>
>And, please keep in mind: many application
>
>And when the filesystem says no because the fs devs don't want to
>have to deal with broken apps because app devs learn that "this is a
>go fast knob" and data integrity be damned? It's "fsync is slow so I
>won't use it" all over again...
...
>
>And, please keep in mind: many application
>> The takeaway is that msync() is 9-10x slower than userspace cache management.
>
>An alternative viewpoint: that flushing clean cachelines is
>extremely expensive on Intel CPUs. ;)
>
>i.e. Same numbers, different analysis from a different PoV, and
>that gives a *completely different
>> The takeaway is that msync() is 9-10x slower than userspace cache management.
>
>An alternative viewpoint: that flushing clean cachelines is
>extremely expensive on Intel CPUs. ;)
>
>i.e. Same numbers, different analysis from a different PoV, and
>that gives a *completely different
8 matches
Mail list logo