An excerpt from the man page of sar reads as follows:
----------------- cut -------------------------
" -B Report paging statistics. The following values are displayed:
pgpgin/s: Total number of kilobytes the system paged in from disk per second.
pgpgout/s: Total number of kilobytes the system paged out to disk per second.
fault/s: Number of page faults (major + minor) made by the system per second (post 2.5 kernels only). This is not a count of page faults that generate I/O, because some page faults can be resolved without I/O.
majflt/s: Number of major faults the system has made per second, those which have required loading a memory page from disk (post 2.5 kernels only)."
----------------- cut -------------------------

Does anyone know if the values pgpgin & pgpgout literally mean pages read from physical disk, or does it also include pages read in from file cache in memory?

thanks

Not exactly sure what you mean by that.

It says "from disk" and "to disk". This deals with writes to and reads from the swap device. There would be no reason to put swap device in "memory" (RAM) because if you have enough RAM you shouldn't need to swapout in the first place.

Some disk arrays (e.g. EMC Symmetrix) have a memory cache to stage writes to disks. If that was where your swap was then the "disk" would be the array's memory cache rather than the actual disk because the OS would consider it written to disk once the array had reported it as such even though behind the scenes the array is doing the write to physical device afterwards. There is no way you could use sar to determine that was happening - it would simply look like a very fast write.

You don't really have to worry much about pagein/pageout unless you see an excessively high number of the former compared to latter as it would indicate you are memory constrained.

thanks for your reply. I was under the impression that `sar -B` reported total pages read/written from the disk during normal application activity (i.e. not swap related). In fact I tested this on an Oracle DB server that is very active, but not using any swap space (naturally) and it did show high pgpgin/pgpgout values which I took to mean normal read/writes from the DB and application files.

As Linux caches much of it's disk reads to page cache for faster access if needed later, my question is then, will the pgpgin value increment if files are read from file cache or will it only increment if it needs to go to disk to access that information? In other words, does sar consider reads from the file cache "disk" access?

(or have I completely missed the purpose of `sar -B`) - thanks

It will only include physical I/O to the "device". Hits on page-cache are effectively transparent - minor page faults give some indication, but not much.
There is an unfortunate mix of terminology here - in Linux the page cache is (also) used for mmap'd file I/O. This explains the numbers the OP sees. Swap(-cache) is a different in that it is only used for (dirty) anonymous pages. It's effectively a subset of pagecache.