On Mon, Sep 5, 2022 at 9:53 PM Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx> wrote: > > On (22/09/05 21:21), Barry Song wrote: > > > 3) HUGE pages recompression is activated by `huge` mode > > > > > > echo huge > /sys/block/zram0/recompress > > > > Thanks for developing this interesting feature. It seems reasonable for cold > > pages. But for a huge page, do you have some data to show that the hugepage > > is not compressed by lzo/lz4 so we need zstd further? i assume the size of > > the huge page you are talking about is 2MB? > > Oh, yeah, this is the lingo we use in zram. We used "huge" object and "huge" > size class in zsmalloc and the term "huge" transitioned to zram, but zram > operates with pages not objects, so huge zsmalloc object is "huge zram page". > > > on second thoughts, it seems you mean hugepage is those pages > > whose compressed data is big? if so, can you please avoid using > > "huge page" as it is quite misleading in linux. we are using hugepage > > for pages larger than 4KB. > > Yes, you are right. And I wish I could use a different term, but... > this is what zram has been using for many years: > Documentation/admin-guide/blockdev/zram.rst > > And we already accept "huge" and "huge pages", and so, in sysfs knobs > (zram device attrs), so the confusing term, unfortunately, is there > forever. make sense! thanks! i assume you will have some benchmark data to compare three cases, 1. lzo with recompress zstd 2. always use lzo 3. always use zstd such as power consumption, cpu utilization, available memory, benefits to user experiences especially to UI smoothness under memory pressure? Thanks Barry
