On Tue, Jan 28, 2025 at 10:04:27AM -0600, Seth Jennings wrote: > On Tue, Jan 28, 2025 at 9:37 AM Yosry Ahmed <yosry.ahmed@xxxxxxxxx> wrote: > > > > On Tue, Jan 28, 2025 at 11:21:10AM +0100, Vlastimil Babka wrote: > > > On 1/28/25 00:58, Yosry Ahmed wrote: > > > > The zbud compressed pages allocator is rarely used, most users use > > > > zsmalloc. zbud consumes much more memory (only stores 1 or 2 compressed > > > > pages per physical page). The only advantage of zbud is a marginal > > > > performance improvement that by no means justify the memory overhead. > > > > > > > > Historically, zsmalloc had significantly worse latency than zbud and > > > > z3fold but offered better memory savings. This is no longer the case as > > > > shown by a simple recent analysis [1]. In a kernel build test on tmpfs > > > > in a limited cgroup, zbud 2-3% less time than zsmalloc, but at the cost > > > > of using ~32% more memory (1.5G vs 1.13G). The tradeoff does not make > > > > sense for zbud in any practical scenario. > > > > > > > > The only alleged advantage of zbud is not having the dependency on > > > > CONFIG_MMU, but CONFIG_SWAP already depends on CONFIG_MMU anyway, and > > > > zbud is only used by zswap. > > > > > > > > Following in the footsteps of [2], which deprecated z3fold, deprecated > > > > zbud as planned and remove it in a few cycles if no objections are > > > > raised from active users. > > > > > > > > Rename the user-visible config options so that users with CONFIG_ZBUD=y > > > > get a new prompt with explanation during make oldconfig. Also, remove > > > > CONFIG_ZBUD from defconfig. > > > > > > > > [1]https://lore.kernel.org/lkml/CAJD7tkbRF6od-2x_L8-A1QL3=2Ww13sCj4S3i4bNndqF+3+_Vg@xxxxxxxxxxxxxx/ > > > > [2]https://lore.kernel.org/lkml/20240904233343.933462-1-yosryahmed@xxxxxxxxxx/ > > > > > > > > Signed-off-by: Yosry Ahmed <yosry.ahmed@xxxxxxxxx> > > > > > > Seems weird not to Cc the folks listed in MAINTAINERS for ZBUD? Unless their > > > addresses are known to bounce? > > > > Ugh I had them in the CC list, but I played around with it and probably > > lost them :/ > > > > Seth, Dan, apologies. > > I haven't contributed in this space for quite some time so I defer to > more active contributors on this. > > If it is true that zsmalloc has no/negligible deficiencies compared to > zbud, it seems reasonable. Zsmalloc has been the default allocator and is widely used, I am unaware of any zbud users in prod settings. > > but at the cost of using ~32% more memory > > I do push back here as this % could be highly variable depending on > the compressibility of the data. 100% agreed, this is just an example from this specific testing scenario. > If the average compressibility is ~50%, zbud should be very memory efficient. > > If the average compressibility is very high or very low, it is less > memory efficient. That being said, the average compression ratio hovers around 30% for workloads I am familiar with. I think I have seen similar ratios from others as well but I will let others chime in.
