Re: [arch-general] most efficient way to get linux kernel statistics
On Mon, 7 Mar 2016 21:20:18 +0100 Damjan Georgievski wrote: > >>> The free command gets its information from /proc/meminfo. > >>> Performance-wise, it doesn't really matter if a few additional lines > >>> need to be parsed. > >> > >> Hello, > >> Thank you Florian. Actually, I already knew it's the same, I just > >> thought that he could use the free command instead of parsing it > >> himself. You're right. Performance-wise, it's pretty much the same. > >> > > > > Sorry, I didn't express myself properly. I didn't mean to criticize > > free. What I meant to say was that the time it takes to parse > > /proc/meminfo or free is negligible. It doesn't need to be any more > > efficient. free parses /proc/meminfo. libgtop parses /proc/meminfo. > > Querying the information probably takes much more time than parsing it. > > > > Your suggestion to use free may indeed be useful if the original poster > > wants parsing to be slightly simpler and does not mind the additional > > dependency. My impression was that they considered parsing to be too > > inefficient. > > there's also `sysinfo(2)` ah, thank you! and now that i found (or rather, thought about it) the source for /proc/meminfo [0] i can see that it uses sysinfo too. but sadly the cached size is not in sysinfo. then i tried to do what /proc/stat [1] does and after googling learned some facts: - you can't use "kernel-space" functions in "user-space" apps - so i cant use all the same functions as /proc/stat uses - you can't create new "syscall" functions in a kernel module - so i can't "export/map" those "kernel-space" functions to be used in my "user-space" app so it seems the only way to get deeper than /proc/stat would be to add new "syscalls" to the linux kernel? (for functions like for_each_possible_cpu and kcpustat_cpu) i'm just a webmonkey, so ill stick to /proc for now :D (i cant notice my daemon in cpu usage yet, so not really a "problem") Cheers Andre Schmidt [0] https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/tree/fs/proc/meminfo.c [1] https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/tree/fs/proc/stat.c
Re: [arch-general] most efficient way to get linux kernel statistics
On 03/07/2016 09:20 PM, Damjan Georgievski wrote: > there's also `sysinfo(2)` > I didn't know there's a syscall for that. That seems like the easiest way to get memory data. I'm not sure if you can get the other information the original poster wanted without parsing /proc though. On 02/19/2016 04:01 AM, Jonathan Horacio Villatoro Córdoba wrote: > Don't worry. Actually, I understood your point perfectly. I suggested > free only because I originally supposed the OP wanted his program to > contain as little code as possible in his program, and didn't want to > parse the file himself. > OK :) .
Re: [arch-general] most efficient way to get linux kernel statistics
On Mon, Mar 07, 2016 at 09:15:06PM +0100, Florian Pelz wrote: > Sorry, I didn't express myself properly. I didn't mean to criticize > free. What I meant to say was that the time it takes to parse > /proc/meminfo or free is negligible. It doesn't need to be any more > efficient. free parses /proc/meminfo. libgtop parses /proc/meminfo. > Querying the information probably takes much more time than parsing it. > > Your suggestion to use free may indeed be useful if the original poster > wants parsing to be slightly simpler and does not mind the additional > dependency. My impression was that they considered parsing to be too > inefficient. Don't worry. Actually, I understood your point perfectly. I suggested free only because I originally supposed the OP wanted his program to contain as little code as possible in his program, and didn't want to parse the file himself.
Re: [arch-general] most efficient way to get linux kernel statistics
>>> The free command gets its information from /proc/meminfo. >>> Performance-wise, it doesn't really matter if a few additional lines >>> need to be parsed. >> >> Hello, >> Thank you Florian. Actually, I already knew it's the same, I just >> thought that he could use the free command instead of parsing it >> himself. You're right. Performance-wise, it's pretty much the same. >> > > Sorry, I didn't express myself properly. I didn't mean to criticize > free. What I meant to say was that the time it takes to parse > /proc/meminfo or free is negligible. It doesn't need to be any more > efficient. free parses /proc/meminfo. libgtop parses /proc/meminfo. > Querying the information probably takes much more time than parsing it. > > Your suggestion to use free may indeed be useful if the original poster > wants parsing to be slightly simpler and does not mind the additional > dependency. My impression was that they considered parsing to be too > inefficient. there's also `sysinfo(2)` -- damjan
Re: [arch-general] most efficient way to get linux kernel statistics
On 03/07/2016 04:36 PM, Jonathan Horacio Villatoro Córdoba wrote: > On Sun, Mar 06, 2016 at 05:13:37PM +0100, Florian Pelz wrote: >> On 03/06/2016 04:47 PM, Jonathan Horacio Villatoro Córdoba wrote: >>> On Sun, Mar 06, 2016 at 03:23:33PM +0100, Andre Schmidt wrote: hello archers, thought i ask here first, before i try some kernel mailing list. i'm writing (for fun) a tiny daemon that sends linux usage (cpu, mem, net, etc.) statistics as efficiently as possible to another machine, repeatedly, to get "live" data. at the moment i'm simply sending /proc files, but they sometimes have too much data. so was wondering if there is a more efficient way to get (only parts of) the data thats available in /proc? for example, /proc/meminfo has all this info: […] if /proc is the only way to get this info, i wonder if creating a kernel module for this would be more efficient, or even possible? >>> >>> Hello Andre, >>> >>> According to what you need, you may want to use the 'free' command >>> instead, as its output is much simpler than the output of the >>> /proc/meminfo file >>> >>> […] >>> >> >> Hello, >> >> The free command gets its information from /proc/meminfo. >> Performance-wise, it doesn't really matter if a few additional lines >> need to be parsed. > > Hello, > Thank you Florian. Actually, I already knew it's the same, I just > thought that he could use the free command instead of parsing it > himself. You're right. Performance-wise, it's pretty much the same. > Sorry, I didn't express myself properly. I didn't mean to criticize free. What I meant to say was that the time it takes to parse /proc/meminfo or free is negligible. It doesn't need to be any more efficient. free parses /proc/meminfo. libgtop parses /proc/meminfo. Querying the information probably takes much more time than parsing it. Your suggestion to use free may indeed be useful if the original poster wants parsing to be slightly simpler and does not mind the additional dependency. My impression was that they considered parsing to be too inefficient.
Re: [arch-general] most efficient way to get linux kernel statistics
On Sun, Mar 06, 2016 at 05:13:37PM +0100, Florian Pelz wrote: > On 03/06/2016 04:47 PM, Jonathan Horacio Villatoro Córdoba wrote: > > On Sun, Mar 06, 2016 at 03:23:33PM +0100, Andre Schmidt wrote: > >> hello archers, > >> > >> thought i ask here first, before i try some kernel mailing list. > >> > >> i'm writing (for fun) a tiny daemon that sends linux usage (cpu, mem, net, > >> etc.) statistics as efficiently as possible to another machine, > >> repeatedly, to get "live" data. > >> > >> at the moment i'm simply sending /proc files, but they sometimes have too > >> much data. so was wondering if there is a more efficient way to get (only > >> parts of) the data thats available in /proc? > >> > >> for example, /proc/meminfo has all this info: > >> > >> […] > >> > >> if /proc is the only way to get this info, i wonder if creating a kernel > >> module for this would be more efficient, or even possible? > > > > Hello Andre, > > > > According to what you need, you may want to use the 'free' command > > instead, as its output is much simpler than the output of the > > /proc/meminfo file > > > > […] > > > > Hello, > > The free command gets its information from /proc/meminfo. > Performance-wise, it doesn't really matter if a few additional lines > need to be parsed. Hello, Thank you Florian. Actually, I already knew it's the same, I just thought that he could use the free command instead of parsing it himself. You're right. Performance-wise, it's pretty much the same.
Re: [arch-general] most efficient way to get linux kernel statistics
On 03/06/2016 04:47 PM, Jonathan Horacio Villatoro Córdoba wrote: > On Sun, Mar 06, 2016 at 03:23:33PM +0100, Andre Schmidt wrote: >> hello archers, >> >> thought i ask here first, before i try some kernel mailing list. >> >> i'm writing (for fun) a tiny daemon that sends linux usage (cpu, mem, net, >> etc.) statistics as efficiently as possible to another machine, repeatedly, >> to get "live" data. >> >> at the moment i'm simply sending /proc files, but they sometimes have too >> much data. so was wondering if there is a more efficient way to get (only >> parts of) the data thats available in /proc? >> >> for example, /proc/meminfo has all this info: >> >> […] >> >> if /proc is the only way to get this info, i wonder if creating a kernel >> module for this would be more efficient, or even possible? > > Hello Andre, > > According to what you need, you may want to use the 'free' command > instead, as its output is much simpler than the output of the > /proc/meminfo file > > […] > Hello, The free command gets its information from /proc/meminfo. Performance-wise, it doesn't really matter if a few additional lines need to be parsed.
Re: [arch-general] most efficient way to get linux kernel statistics
On Sun, Mar 06, 2016 at 03:23:33PM +0100, Andre Schmidt wrote: > hello archers, > > thought i ask here first, before i try some kernel mailing list. > > i'm writing (for fun) a tiny daemon that sends linux usage (cpu, mem, net, > etc.) statistics as efficiently as possible to another machine, repeatedly, > to get "live" data. > > at the moment i'm simply sending /proc files, but they sometimes have too > much data. so was wondering if there is a more efficient way to get (only > parts of) the data thats available in /proc? > > for example, /proc/meminfo has all this info: > > MemTotal:8051660 kB > MemFree: 6046020 kB > MemAvailable:6843080 kB > Buffers: 109116 kB > Cached: 708336 kB > SwapCached:0 kB > Active: 1332416 kB > Inactive: 455428 kB > Active(anon): 971040 kB > Inactive(anon):22340 kB > Active(file): 361376 kB > Inactive(file): 433088 kB > Unevictable: 16 kB > Mlocked: 16 kB > SwapTotal: 1048572 kB > SwapFree:1048572 kB > Dirty:16 kB > Writeback: 0 kB > AnonPages:970412 kB > Mapped: 323384 kB > Shmem: 23008 kB > Slab: 71760 kB > SReclaimable: 44860 kB > SUnreclaim:26900 kB > KernelStack:6544 kB > PageTables:22924 kB > NFS_Unstable: 0 kB > Bounce:0 kB > WritebackTmp: 0 kB > CommitLimit: 5074400 kB > Committed_AS:3487604 kB > VmallocTotal: 34359738367 kB > VmallocUsed: 0 kB > VmallocChunk: 0 kB > HardwareCorrupted: 0 kB > AnonHugePages:413696 kB > HugePages_Total: 0 > HugePages_Free:0 > HugePages_Rsvd:0 > HugePages_Surp:0 > Hugepagesize: 2048 kB > DirectMap4k: 168600 kB > DirectMap2M: 8091648 kB > > but i only need: > > MemTotal:8051660 kB > MemAvailable:6843080 kB > Cached: 708336 kB > > event better would be if kernel would only tell me: > > 8051660 6843080 708336 > > if /proc is the only way to get this info, i wonder if creating a kernel > module for this would be more efficient, or even possible? Hello Andre, According to what you need, you may want to use the 'free' command instead, as its output is much simpler than the output of the /proc/meminfo file E.g. >$ free > totalused freeshared buff/cache available >Mem: 5951844 486324 4969788 75812 495732 5339492 >Swap: 00 0 or, if you prefer it be displayed in mebibytes, >$ free -m > total usedfreeshared buff/cache available >Mem: 5812496 483273 484 5193 >Swap: 0 0 0 you can also check the man page for more display options Hope this helps, Jonathan
[arch-general] most efficient way to get linux kernel statistics
hello archers, thought i ask here first, before i try some kernel mailing list. i'm writing (for fun) a tiny daemon that sends linux usage (cpu, mem, net, etc.) statistics as efficiently as possible to another machine, repeatedly, to get "live" data. at the moment i'm simply sending /proc files, but they sometimes have too much data. so was wondering if there is a more efficient way to get (only parts of) the data thats available in /proc? for example, /proc/meminfo has all this info: MemTotal:8051660 kB MemFree: 6046020 kB MemAvailable:6843080 kB Buffers: 109116 kB Cached: 708336 kB SwapCached:0 kB Active: 1332416 kB Inactive: 455428 kB Active(anon): 971040 kB Inactive(anon):22340 kB Active(file): 361376 kB Inactive(file): 433088 kB Unevictable: 16 kB Mlocked: 16 kB SwapTotal: 1048572 kB SwapFree:1048572 kB Dirty:16 kB Writeback: 0 kB AnonPages:970412 kB Mapped: 323384 kB Shmem: 23008 kB Slab: 71760 kB SReclaimable: 44860 kB SUnreclaim:26900 kB KernelStack:6544 kB PageTables:22924 kB NFS_Unstable: 0 kB Bounce:0 kB WritebackTmp: 0 kB CommitLimit: 5074400 kB Committed_AS:3487604 kB VmallocTotal: 34359738367 kB VmallocUsed: 0 kB VmallocChunk: 0 kB HardwareCorrupted: 0 kB AnonHugePages:413696 kB HugePages_Total: 0 HugePages_Free:0 HugePages_Rsvd:0 HugePages_Surp:0 Hugepagesize: 2048 kB DirectMap4k: 168600 kB DirectMap2M: 8091648 kB but i only need: MemTotal:8051660 kB MemAvailable:6843080 kB Cached: 708336 kB event better would be if kernel would only tell me: 8051660 6843080 708336 if /proc is the only way to get this info, i wonder if creating a kernel module for this would be more efficient, or even possible? Cheers Andre Schmidt
