Need a meaningful memory measure out of /proc/meminfo

From: Denis Vlasenko (vda@port.imtp.ilyichevsk.odessa.ua)
Date: Mon Jul 28 2003 - 03:51:09 EST

Hi. I want to calculate and give to user a terse info
(one or just a couple of numbers) about current memory
consumption. Since today's kernels typically use almost
all RAM for cacheing it is useless to just report amount
of RAM in use, it's nearly always is $total_ram-10k.

As a first approximation I took amount of allocated RAM
which will survive oom. Look (zero swap/highmem snipped):

$ cat /proc/meminfo;./oom;cat /proc/meminfo
        total: used: free: shared: buffers: cached:
Mem: 262057984 235958272 26099712 0 3977216 126554112
Swap: 0 0 0
MemTotal: 255916 kB
MemFree: 25488 kB
MemShared: 0 kB
Buffers: 3884 kB
Cached: 123588 kB
Active: 65148 kB
Inactive: 144760 kB
LowTotal: 255916 kB
LowFree: 25488 kB
./oom: Killed
        total: used: free: shared: buffers: cached:
Mem: 262057984 108625920 153432064 0 479232 9801728
Swap: 0 0 0 ^ ^
MemTotal: 255916 kB | |
MemFree: 149836 kB | |
MemShared: 0 kB | |
Buffers: 468 kB <-------- less than 1M |
Cached: 9572 kB <-------- a 10M---------------+
Active: 24220 kB
Inactive: 68248 kB
LowTotal: 255916 kB
LowFree: 149836 kB

Ok, so I use Used-Buffers-Cached as a memory measure.
It remains almost unchanged across oom.

Although I am worried that I can lie to user because above test
shows that some 10M of buffers+cached are not really freeable.
Also I run this box swapless, what to do with swap numbers?

Please share your thoughs.

----BEGIN SHAMELESS PLUG SECTION----
Why do I need this? I found a little cute utility, nanotop,
in floppyfw project. It shows cpu load, mem load and eth load,
one line every second.

I started to play with it. I improved ability to parse command line,
specify update period, cpu bar size, which interface(s) to report,
made output even more terse. And hopefully made mem output somewhat
more meaningful.

Code is reworked to completely avoid <stdio.h> and made modular.
You can add code to monitor something else real easy now, swap
is already planned for 0.2 ;)

This toy compiles to a static binary under 4k with dietlibc.

I renamed it to nanometer, it really has not much in common with top.
Source and precompiled binary is in attached tarball.

Compare for yourself:

# ./nanotop
CPU [ ] MEM 219.6MB eth0: 35.00B/s rcv 18.00B/s snd 0 err
CPU [- ] MEM 219.6MB eth0: 6.7kB/s rcv 1.6kB/s snd 0 err
CPU [- ] MEM 219.6MB eth0: 4.8kB/s rcv 4.0kB/s snd 0 err
CPU [- ] MEM 219.6MB eth0: 5.3kB/s rcv 1.5kB/s snd 0 err
CPU [- ] MEM 219.6MB eth0: 3.6kB/s rcv 2.4kB/s snd 0 err
CPU [=- ] MEM 219.6MB eth0: 74.3kB/s rcv 62.5kB/s snd 0 err
CPU [- ] MEM 219.6MB eth0: 3.4kB/s rcv 2.4kB/s snd 0 err

# ./nanometer c m ieth0 ilo ieth1
cpu [??????????] mem 78M eth0 0 0 lo 0 0 eth1 ???? ????
cpu [..........] mem 78M eth0 4.0k 2.4k lo 0 0 eth1 ???? ????
cpu [..........] mem 78M eth0 9.6k 2.6k lo 0 0 eth1 ???? ????
cpu [UU........] mem 78M eth0 78k 64k lo 140 140 eth1 ???? ????
cpu [U.........] mem 78M eth0 7.6k 4.8k lo 0 0 eth1 ???? ????
cpu [UU........] mem 78M eth0 4.7k 9.2k lo 0 0 eth1 ???? ????
cpu [..........] mem 78M eth0 1.6k 1.3k lo 0 0 eth1 ???? ????
cpu [U.........] mem 78M eth0 10k 3.7k lo 0 0 eth1 ???? ????
cpu [..........] mem 78M eth0 152k 129k lo 140 140 eth1 ???? ????
cpu [SUUU......] mem 78M eth0 257k 171k lo 0 0 eth1 ???? ????
cpu [UUUUU.....] mem 78M eth0 352k 303k lo 0 0 eth1 ???? ????
cpu [SUUUU.....] mem 78M eth0 236k 232k lo 0 0 eth1 ???? ????
----END SHAMELESS PLUG SECTION---- 

--
vda
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/

This archive was generated by hypermail 2b29 : Thu Jul 31 2003 - 22:00:35 EST