http://bugzilla.novell.com/show_bug.cgi?id=405246
User haveblue@us.ibm.com added comment
http://bugzilla.novell.com/show_bug.cgi?id=405246#c9
Dave Hansen
Just an other point: Is http://marc.info/?l=linux-mm&m=120496901605830&w=2 wrong in your opinion, if not I'd like to close this without any change.
Yes, I believe that is wrong. As I said in the post above, it changes long-established behavior. There is absolutely no reason to be adding 'kb_swap_cached' into the calculation at this point. It has been in there since at least the 2.5 fork and surely there have been changes to procps in the last seven years, even if it is a bit slow. There is also documentation out there that tells people who are having memory issues to go look at the output from procps commands. They might be told 'if "cached" decreases or increases too fast under memory pressure, go adjust /proc/sys/vm/vfs_cache_pressure accordingly'. (BTW, that adjusts how much we go after page cache vs dentries/inodes) Don't get me wrong, I think it is wonderful that someone is looking after procps. I'd love to see more information about slab in those tools. It is one of the things they're lacking the most. However, I don't want to see it mixed in with the rest of the established information in there. Perhaps some extra fields somewhere? I'd *love* to see 'vmstat -x' or something for extended information. I also desire consistency. I'd hate to have to start trying to consider what particular versions of distributions a customer is running. For instance, I've had customers believe 'cached' is at too high a value. Now, if they went to test SLES10 from SLES11 they might falsely believe that the behavior has changed and been fixed. I'd be happy if one of the following was done: * Add this as optional behavior that is not the default * Get this accepted into the upstream procps distribution * Convince several other distros to take the same patch * Revert the patch -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.