KSM

KSM(Kernel Samepage Merging). Try to improve it. Based on PKSM. Kernel v3.18-rc7

Possible plans

delay, 延迟去重..
hash adjustable. 尝试调整部分哈希的强度. 来自 UKSM
hash table instead of rbtree. 将红黑树改为哈希表.
test UKSM, 移植 UKSM.
same virtual address, 合并相同虚拟地址的页面.
page cache, Page Cache 中是否有去重的潜力.

Background

KSM

Plans

delay

adaptive partial hash

aph

Others

COW

    struct BookKeeper *bk = BookKeeper_init();
    alloc_pages_write(bk, 100, 1);
    sleep(10);

    pid_t pid = fork();
    int r;
    if (pid == 0) {
            write_pages(bk, 100, 2);
            sleep(10);
    } else {
            wait(&r);
    }

    free_all_pages(bk);

期望结果是, 子进程 write_pages() 导致 COW, 产生新的页面, 这些页面再次被合并, sharing 增加 100, 实际上, 没有增加. 可能原因:

这个结果和之前在手机上测的不一致, 手机上 fork 后, sharing 增加了.

    struct BookKeeper *bk = BookKeeper_init();
    alloc_pages_write(bk, 100, 1);
    sleep(10);

    pid_t pid = fork();
    int r;
    if (pid == 0) {
            write_pages(bk, 100, 2);
            sleep(10);
    } else {
            write_pages(bk, 100, 3);
            wait(&r);
    }

    free_all_pages(bk);

期望结果, write_pages() 会导致 COW, sharing 减少 100, 然后增加 200, 实际结果, 在 write_pages() 后, sharing 减少 100.

这个程序移到手机上测一测?

如何使用 email 收发 patch

reveriel/ksm-doc