On 11/15/2013 12:21 AM, Seth Jennings wrote: > On Wed, Nov 13, 2013 at 08:00:34PM -0800, Hugh Dickins wrote: >> On Wed, 13 Nov 2013, Minchan Kim wrote: >> ... >>> >>> Hello Andrew, >>> >>> I'd like to listen your opinion. >>> >>> The zram promotion trial started since Aug 2012 and I already have get many >>> Acked/Reviewed feedback and positive feedback from Rik and Bob in this thread. >>> (ex, Jens Axboe[1], Konrad Rzeszutek Wilk[2], Nitin Gupta[3], Pekka Enberg[4]) >>> In Linuxcon, Hugh gave positive feedback about zram(Hugh, If I misunderstood, >>> please correct me!). And there are lots of users already in embedded industry >>> ex, (most of TV in the world, Chromebook, CyanogenMod, Android Kitkat.) >>> They are not idiot. Zram is really effective for embedded world. >> >> Sorry for taking so long to respond, Minchan: no, you do not misrepresent >> me at all. Promotion of zram and zsmalloc from staging is way overdue: >> they long ago proved their worth, look tidy, and have an active maintainer. >> >> Putting them into drivers/staging was always a mistake, and I quite >> understand Greg's impatience with them by now; but please let's move >> them to where they belong instead of removing them. >> >> I would not have lent support to zswap if I'd thought that was going to >> block zram. And I was not the only one surprised when zswap replaced its >> use of zsmalloc by zbud: we had rather expected a zbud option to be added, >> and I still assume that zsmalloc support will be added back to zswap later. > > Yes, it is still the plan to reintroduce zsmalloc as an option (possibly > _the_ option) for zswap. > > An idea being tossed around is making zswap writethrough instead of > delayed writeback. > > Doing this would be mean that zswap would no longer reduce swap out > traffic, but would continue to reduce swap in latency by reading out of > the compressed cache instead of the swap device. > > For that loss, we gain a benefit: the compressed pages in the cache are > clean, meaning we can reclaim them at any time with no writeback > cost. This addresses Mel's initial concern (the one that led to zswap > moving to zbud) about writeback latency when the zswap pool is full. > Agree! > If there is no writeback cost for reclaiming space in the compressed > pool, then we can use higher density packing like zsmalloc. > But zsmalloc will compact several 0-order pages together as a zpage which cause it not easy to reclaim one 0-order page directly from it. Especially if we want to make zswap pool can be dynamically managed in future. > Making zswap writethough would also make the difference between zswap > and zram, both in terms of operation and application, more apparent, > demonstrating the need for both. > -- Regards, -Bob -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>