Incorrect memory usage reported

[xxxserxxx]

On FreeBSD, it shows 10.9 GB / 24 GB when the top utility shows

Mem: 3379M Active, 2029M Inact, 8809M Laundry, 7180M Wired, 1561M Buf, 2474M Free
ARC: 4001M Total, 3809M MFU, 168M MRU, 64K Anon, 12M Header, 13M Other
     3830M Compressed, 4002M Uncompressed, 1.04:1 Ratio
Swap: 4096M Total, 3073M Used, 1023M Free, 75% Inuse

Only 2.4 GB is free, but it shows that 24-10.9=13.1 GB is free.

You might also want to show all numbers, since you have space there.

• Original ticket
• Original author: @yurivict
• xxxserxxx real comments start here

[xxxserxxx]

So it's showing 10.8 GB / 24 GB but it should show like 21 GB / 24 GB? And this is just for the 'Main' memory and not Swap? And could you provide the output of uname -a? Thanks!

[xxxserxxx]

Yes, this is only for the Main memory.

$ uname -a
FreeBSD yv.noip.me 11.2-STABLE FreeBSD 11.2-STABLE #0 r338762M: Tue Sep 18 18:20:50 PDT 2018     unknown@nohost.com:/usr/obj/usr/src/sys/GENERIC  amd64

[xxxserxxx]

You might also want to show all numbers, since you have space there.

That would make more sense because only showing Free memory is not that interesting.

[xxxserxxx]

Showing all numbers would be more interesting, but the graph is only intended to show free memory and the label should probably reflect that imo.

[xxxserxxx]

I'm not going to add memory stats to the widget, because FreeBSD reports memory stats like darwin reports thermal data: there are a million to choose from. However, I welcome a separate feature request -- separated from this bug report -- with suggestions for which memory metrics to report.

As for the bug itself, top on FreeBSD doesn't appear to accurately present the data either, compared against the information from vmstat. In any case, the information gotop's reporting is clearly wrong. Closest to a clean and concise report on FreeBSD aligning more or less with vmstat is colortop, so the objective of the bug fix is to get gotop's output to match those values for now.

Edit

top on FreeBSD (running in a VM) reports even more wildly inaccurate swap numbers, when compared with the authoritative swapinfo command. This makes me extra suspicious of using top for any baseline source of truth for bug reports.

[xxxserxxx]

I'm taking this out for the moment. top's numbers are unreliable at best, and completely wrong in some cases. Swap information is the most clear type of memory information on FreeBSD, and top's numbers for Swap do not match any other system metric -- not swapinfo, not sysctl, and not any other tool. gopsutil (and, therefore, gotop) reports swap information matching system tool data closely.

I do have an open question about exactly which FreeBSD memory attribute best represents system physical system memory. gotop (through gopsutil) reports physical memory, and in testing these numbers closely match information reported by several other systems, including sysctl and freebsd-memory, and it reports numbers much closer to that reported by the kernel during boot-up (grep memory /var/run/dmesg.boot) than top's information. There are other attributes to use -- about 8, but up to a couple dozen (sysctl -s) -- and it isn't obvious that using hw.phys is incorrect.

While I won't close this, after spending a few hours investigating this, I'm not convinced this is a valid issue.