On Mon, Jul 15, 2013 at 10:56:17AM +0800, Bob Liu wrote: > As my test results showed in this thread. > 1. Zswap only useful when total ram size is large else the performance > was worse than disabled it! I have not observed this. In my kernbench runs, I was using VMs with ~512MB of RAM and saw significant improvement from zswap. > > 2. Zswap occupies some memory but that's unfair to file pages, more > file pages maybe reclaimed during memory pressure. This is true. It remains to be explored how the policies that balance anon reclaim and page cache reclaim can be respected by zswap. Until then though, the growth of the zswap pool does add memory pressure which causes more reclaim in general, both anon and page cache. > I think that's why the performance of the background io-duration was > worse than disable zswap. The I/O load during the parallelio-memcached test shouldn't be effected by page cache reclaim since it is not re-reading anything. Again, I say that that test is not a good and repeatable (across different systems and kernel versions) to test zswap. parallelio-memcached is designed to test suboptimal page reclaim decisions, not swap performance. Have you tried running kernbench in a memory environment restricted enough to cause swapping with zswap enabled? I think that would be a better test. Seth -- 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>