linux – Getting Process Memory Usage Information from /proc/pid/smaps

linuxmemoryprocessvirtual-memory

For a given process in /proc/<pid>/smaps, for a given mapping entry what are:

  1. Shared_Clean
  2. Shared_Dirty
  3. Private_Clean
  4. Private_Dirty

Is Shared_Clean + Shared_Dirty the amount of memory that is shared with other processes? So it is like shared RSS?

Similarly is Private_Clean + Private_Dirty the amount of memory that is available for only one process? So it is like private RSS?

Is the PSS value = PrivateRSS + (SharedRSS / number of processes sharing it)?

Some more questions after reading this link: LWN

Now lets talk about the process as a whole, whose smaps entry we are looking at.

I noticed that if I do Shared_Clean + Shared_Dirty + Private_Clean + Private_Dirty for every smaps entry for the process I get the RSS of the process as
reported by ps, which is pretty cool. For e.g. 

ps -p $$ -o pid,rss

Will give me the (approx) same value for rss as the sum of every Shared_Clean , Shared_Dirty , Private_Clean , Private_Dirty entry in /proc/$$/smaps.

But what about PSS for the entire process? So, from the example above how do I get the PSS for $$ ? Can I just add the PSS entry for every smaps mapping and arrive at PSS for $$ ?

And what about USS for the entire process? Again taking the example above I am guessing that I can arrive at the USS for $$ by summing up only the Private_* entries for every smaps entry for $$..right?

Notes:
PSS= Proportional set size.
USS= Unique set size.

Best Answer

Clean pages are pages that have not been modified since they were mapped (typically, text sections from shared libraries are only read from disk (when necessary), never modified, so they'll be in shared, clean pages).
Dirty pages are pages that are not clean (i.e. have been modified).

Private pages are available only to that process, shared pages are mapped by other processes*.

RSS is the total number of pages, shared or not, currently mapped into the process. So Shared_Clean + Shared_Dirty would be the shared part of the RSS (i.e. the part of RSS that is also mapped into other processes), and Private_Clean + Private_Dirty the private part of RSS (i.e. only mapped in this process).

PSS (proportional share size) is as you describe. Private pages are summed up as is, and each shared mapping's size is divided by the number of processes that share it.
So if a process had 100k private pages, 500k pages shared with one other process, and 500k shared with four other processes, the PSS would be:

100k + (500k / 2) + (500k / 5) = 450k

Further readings:

Regarding process-wide sums:

  • RSS can be (approximately+) obtained by summing the Rss: entries in smaps (you don't need to add up the shared/private shared/dirty entries).

    awk '/Rss:/{ sum += $2 } END { print sum }' /proc/$$/smaps
  • You can sum up Pss: values the same way, to get process-global PSS.
  • USS isn't reported in smaps, but indeed, it is the sum of private mappings, so you can obtain it the same way too

*Note that a "share-able" page is counted as a private mapping until it is actually shared. i.e. if there is only one process currently using libfoo, that library's text section will appear in the process's private mappings. It will be accounted in the shared mappings (and removed from the private ones) only if/when another process starts using that library.
+The values don't add up exactly for all processes. Not exactly sure why... sorry.

Related Question