Messages in this thread | | | Date | Mon, 6 Jul 2015 22:48:50 +0900 | From | Sergey Senozhatsky <> | Subject | Re: [PATCH v5 6/7] zsmalloc: account the number of compacted pages |
| |
On (07/06/15 22:22), Minchan Kim wrote: > On Mon, Jul 06, 2015 at 09:17:49PM +0900, Sergey Senozhatsky wrote: > > Compaction returns back to zram the number of migrated objects, > > which is quite uninformative -- we have objects of different > > sizes so user space cannot obtain any valuable data from that > > number. Change compaction to operate in terms of pages and > > return back to compaction issuer the number of pages that > > were freed during compaction. So from now on `num_compacted' > > column in zram<id>/mm_stat represents more meaningful value: > > the number of freed (compacted) pages. > > Fair enough. > > The main reason I introduced num_migrated is to investigate > the effieciency of compaction. ie, num_freed / num_migrated. > However, I didn't put num_freed at that time so I can't get > my goal with only num_migrated. > > We could put new knob num_compacted as well as num_migrated > but I don't think we need it now. Zram's compaction would be > much efficient compared to VM's compaction because we don't > have any non-movable objects in zspages and we can account > exact number of free slots. > > So, I want to change name from num_migrated to num_compacted > and maintain only it which is more useful for admin, you said. > > It's not far since we introduced num_migrated so I don't think > change name wouldn't be a big problem for userspace,(I hope). >
Hello,
Yes, num_migrated rename patch was on my table. I was thinking about two variants:
struct zs_pool atomic_long_t pages_allocated; unsigned long num_compacted;
or
struct zs_pool atomic_long_t pages_allocated; unsigned long pages_compacted;
the latter looks even better I think. But I didn't come up with a sane API name to get these stats-- zs_get_pages_compacted() is a bit misleading. So I decided to keep num_compacted to make the patch set smaller.
Hm, exporting a new `struct zs_pool_stat' to zram is probably a good way to go.
-ss
| |