The dirty and non-dirty pages are checked one by one in vl.c. When the most of the memory is not dirty, checking the dirty and non-dirty pages by multiple page size should be much faster than checking them one by one. We introduced bit-based phys_ram_dirty for VGA, CODE and MIGRATION, and cpu_physical_memory_get_dirty_range() for this purpose. This patch is based on the following discussion. http://www.mail-archive.com/kvm@xxxxxxxxxxxxxxx/msg28733.html To prove our prospect, we have evaluated effect of this patch. We compared runtime of ram_save_remaining with original ram_save_remaining() and ram_save_remaining() using functions of this patch. Test Environment: CPU: 4x Intel Xeon Quad Core 2.66GHz Mem size: 96GB kvm version: 2.6.33 qemu-kvm version: commit 2b644fd0e737407133c88054ba498e772ce01f27 Host OS: CentOS (kernel 2.6.33) Guest OS: Debian/GNU Linux lenny (kernel 2.6.26) Guest Mem size: 512MB Conditions of experiments are as follows: Cond1: Guest OS periodically makes the 256MB continuous dirty pages. Cond2: Guest OS periodically makes the 256MB dirty pages and non-dirty pages in turn. Cond3: Guest OS read 1GB file, which is bigger than memory. Cond4: Guest OS write 1GB file, which is bigger than memory. Experimental results: Cond1: 1.9 ~ 61 times speed up Cond2: 1.9 ~ 56 times speed up Cond3: 1.9 ~ 59 times speed up Cond4: 1.7 ~ 59 times speed up -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
