OpenVZ Forum: Support » Where can we download the "vzstat" source code?

Home » General » Support » Where can we download the "vzstat" source code?

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Where can we download the "vzstat" source code? [message #2687]

Mon, 17 April 2006 05:46

RapidVPS
Messages: 32
Registered: January 2006

Member

Hi, I downloaded "vzctl-3.0.0-4.tar.bz2" but could not find any mention of vzstat. Where is the source code to this app?

Rick

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #2688 is a reply to message #2687]

Mon, 17 April 2006 07:55

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

vzstat is not available in OVZ.
what did you want to do with it? Maybe I can suggest some workaround?

http://static.openvz.org/userbars/openvz-developer.png

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #2770 is a reply to message #2688]

Thu, 20 April 2006 03:26

RapidVPS
Messages: 32
Registered: January 2006

Member

I am building some vps realtime CPU logging/monitoring tools. How can I determine the cpu usage of a given vps? the CPU value that is in vzstat.

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #2771 is a reply to message #2770]

Thu, 20 April 2006 06:03

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

You can use /proc/vz/vestat file:

[root@tst ~]# cat /proc/vz/vestat
Version: 2.2
      VEID       user       nice     system     uptime                 idle                 strv               uptime                 used               maxlat               totlat   numsched
       100          0          0       5063   40598698       70123626936598                    0       35069066535737           8612439976                 2314            737714316      53144

in this file for each VPS you have:
1. "user nice system uptime" in jiffies. i.e. divide by jiffies_per_second to get seconds (see below).
2. "idle strv uptime used" in cycles. much more accurate. seconds = used / cycles_per_jiffy / jiffies_per_seconds.
3. values of cycles_per_jiffy and jiffies_per_seconds can be read from /proc/vz/stats

BTW, while looking into this I found a small misprint in /proc/vz/vestat output code, so thank you very much Smile

--- ./kernel/vecalls.c.usertm	2006-04-07 19:04:46.000000000 +0400
+++ ./kernel/vecalls.c	2006-04-20 10:05:24.000000000 +0400
@@ -2875,7 +2875,7 @@ static int vestat_seq_show(struct seq_fi
 	idle_time = strv_time = used = 0;
 
 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
-		user_ve += VE_CPU_STATS(ve, cpu)->nice;
+		user_ve += VE_CPU_STATS(ve, cpu)->user;
 		nice_ve += VE_CPU_STATS(ve, cpu)->nice;
 		system_ve += VE_CPU_STATS(ve, cpu)->system;
 		used += VE_CPU_STATS(ve, cpu)->used_time;

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #2790 is a reply to message #2771]

Fri, 21 April 2006 00:36

RapidVPS
Messages: 32
Registered: January 2006

Member

Hi, I could not locate /proc/vz/stats. Can you verify?

#ls -al /proc/vz/
total 0
dr-xr-xr-x 3 root root 0 Apr 20 20:35 .
dr-xr-xr-x 2756 root root 0 Feb 24 13:59 ..
-r-------- 1 root root 0 Apr 20 20:36 devperms
-r-------- 1 root root 0 Apr 20 20:35 veinfo
-r-------- 1 root root 0 Apr 20 20:36 veip
-r-------- 1 root root 0 Apr 20 20:36 vestat
dr-x------ 2 root root 0 Apr 20 20:36 vzaquota
-r-------- 1 root root 0 Apr 20 20:36 vzquota
#uname -a
Linux sodium.rapidvps.net 2.6.8-022stab070.1-enterprise #1 SMP Mon Feb 20 19:31:28 MSK 2006 i686 i686 i386 GNU/Linux

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #2795 is a reply to message #2790]

Fri, 21 April 2006 08:43

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

Oh, it's my poor memory.
cycles_per_jiffy = frequency_of_your_cpu/jiffies_per_second
jiffies_per_second = 1000

frequency_of_your_cpu can be read from /proc/meminfo, as:
cpu MHz : <frequency_of_your_cpu>

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22433 is a reply to message #2795]

Sun, 28 October 2007 19:05

KoolK
Messages: 8
Registered: September 2007

Junior Member

Hi,

I am building tool for CPU, memory usage for VPS in OpenVZ. Can anybody help me and explain what do fields like uptime, used, maxlat, totlat, numsched in file /proc/vz/vestat signify?

Thanks in advance

[Updated on: Sun, 28 October 2007 19:06]

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22448 is a reply to message #22433]

Mon, 29 October 2007 08:31

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

it's quite easy:

[root@dhcp0-95 ~]# cat /proc/vz/vestat
Version: 2.2
                VEID                 user                 nice               system               uptime                 idle                 strv               uptime                 used               maxlat               totlat             numsched
                 100                 1634                    0                 1712            343011435     1277332114762816                    0      638679179771038          16971458021                 4921           5633592426               145722
                   1               157635                    0                29103            488618806     1818929923633445                    0      909796661898187         512196984005                 4858          50362877628              2524559
                 101                  759                    0                 1844            488627361     1819537209544684                    0      909812591751777          33663331702                 4529          10257230389               713559
                 103                  765                    0                 1468            488628230     1819533866071061                    0      909814208459884          28408677122                 5117          11686660006               712016
                 105                 4522                    0                 5088            488628270     1819572969105220                    0      909814283772065          22621645151                 4949           7557632985               125350

- column 1 "VEID": VE id Smile

- column 2 "user", 3 "nice", 4 "system", 5 "uptime" - corresponding std user/nice/system/uptime values in jiffies to standalone linux /proc/stat. Note, there is no "idle" time here, since it can't be calculated this way.

the next group comes in cycles units:

- column 6 "idle" - idle time, 7 "strv" - not used, 8 "uptime" - uptime in cycles, 9 "used" - used time by VE across all CPUs in cycles

the next group is scheduling latency statistics in cycles:
- column 10 "maxlat" - max latency in cycles meaning how long VE process has to wait before it actually got CPU time.
- column 11/12 "totlat/numsched", i.e. divide 11 on 12 to get average scheduling latency.

you can also read:

[root@dhcp0-95 ~]# cat /proc/vz/stats
Version: 2.6
cycles_per_jiffy: 1860831
jiffies_per_second: 1000

to understand how many cycles and jiffies are per second.

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22456 is a reply to message #22448]

Mon, 29 October 2007 11:07

Valmont
Messages: 225
Registered: September 2005

Senior Member

Impresses. It is easy to trace cpu using Smile

[Updated on: Mon, 29 October 2007 11:07]

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22766 is a reply to message #22448]

Sat, 03 November 2007 06:28

KoolK
Messages: 8
Registered: September 2007

Junior Member

Hi Dev,

Thank you very much for help...I have few doubts... Confused

(1) Do the two uptime fields in /proc/vz/vestat denotes same (time since last boot) or different values( if so which values) because I tried to convert each field in file to sec by formulas give in post above..

first uptime in file:
uptime in sec= uptime_in_jiffies/jiffies_per_sec;

second uptime in file:
uptime in sec= uptime_in_cycle/cpu_freq

(cpu_freq is in MHZ..so no need to divide by jiffies_per_sec*jiffies_per_sec...I tried with this also still both values do not match)

the values are not same.

(2) I ran several infinite processes in two different VEs each...still the values of Maxlat, Totlat, Numsched remain 0. Am I doing any mistake?

thanks in advance...

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22767 is a reply to message #22766]

Sat, 03 November 2007 09:45

rickb
Messages: 368
Registered: October 2006

Senior Member

Hi, I thought these fields were only avaialble in the virtuozzo kernel.

virtuozzo 3.x 2.6.9

Quote:

[root@yttrium ~]# cat /proc/vz/vestat
Version: 2.2
VEID user nice system uptime idle strv uptime used maxlat totlat numsched
1030 34518869 562190 10999333 783842709 9045444692949996 0 2402735226105746 141451539991400 31786364 249070126464914 69045720
2086 29900400 0 4506271 1636856393 19450221162219816 0 5017502202734454 108719002917852 2942466202 345587087247930 57824594
1024 21191653 0 23079329 2060298740 24614175256611440 0 6315492003670710 141218633354376 55571736 187231807734994 75413693
1033 22317199 0 72997502 2744987204 12939562299845728 0 8414286920957352 1083430833802996 162172532 22732088731566506 3584853667
9889220 52074253 4596608 18967106 3807321208 34535853929424632 0 11670689390962514 289382121038592 5121172724 19637723631512872 191177819
[root@yttrium ~]# uname -a
Linux yttrium 2.6.9-023stab033.7-enterprise #1 SMP Wed Nov 15 14:58:00 MSK 2006 i686 i686 i386 GNU/Linux
[root@yttrium ~]#

openvz 2.6.18 el5 kernel:

Quote:

[root@beast ~]# more /proc/vz/vestat
Version: 2.2
VEID user nice system uptime idle strv uptime used maxlat totlat numsched
9889998 133305 119 32588 18161899 337438033162426 0 42371308510342 397498699339 0 0 0
9890597 3578 59165 216691 20955483 385816188773967 0 48888681988797 675360951251 0 0 0
9890697 1082186 151 167382 21137597 384817445703089 0 49313548979135 2953753939612 0 0 0
9891687 98727 395 100033 59602854 1088224694685604 0 139052138699873 525936556194 0 0 0
9890885 879421 144 32052757 70982874 577295962439301 0 165601472128687 80575127114055 0 0 0
9890873 15481 338 10649 76233353 1390464816241648 0 177850721787158 74312085540 0 0 0
9890825 325128 330 192029 91927040 1699110331139417 0 214463745852951 1260942674159 0 0 0
9891890 67485 625 42338 121264030 2162101569813360 0 282906292312196 264202327921 0 0 0
9890978 278071 538 298084 146878626 2668734525714551 0 342666792897541 1481079142459 0 0 0
9890715 311897 635 187855 147589429 2735871461013208 0 344325080204707 1291547097603 0 0 0

-------------
Common Terms I post with: http://wiki.openvz.org/Category:Definitions

UBC. Learn it, love it, live it: http://wiki.openvz.org/Proc/user_beancounters

[Updated on: Sat, 03 November 2007 09:45]

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22830 is a reply to message #22766]

Mon, 05 November 2007 18:24

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

they should be the same, but the uptime in cycles should give more precise values.
P.S. However, there are rare cases when it doen't work in case your SMP system have non-syncronized cycles accross cpus.

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22951 is a reply to message #22830]

Wed, 07 November 2007 23:30

KoolK
Messages: 8
Registered: September 2007

Junior Member

Hi,

My second doubt is still unsolved.I created two VEs and then executed infinite process with sleep of 5 seconds. In VE I could run 40 instances of infinite processes in background. After that system failed to run the process as there was no memory to allocate.

Still fields in /proc/ve/vestat maxlat, totlat, numsched remains 0. According to my understanding the latency should increase as system load increases. I want to use this to show system performance for eg to show values in red if average latency in last scanning period has increased. My plan is to show it in graphs.

So please anybody can tell why maxlat, totlat and numsched remain 0?

Thanks in advance

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22952 is a reply to message #22951]

Thu, 08 November 2007 07:36

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

what do you mean by "executed infinite process with sleep of 5 seconds"?

if process sleeps - it doesn't take CPU and doesn't affect latency.
as you can see in my example output of /proc/vz/vestat latencies were not 0. Do you have zeros there in your case?

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #22953 is a reply to message #22952]

Thu, 08 November 2007 08:59

KoolK
Messages: 8
Registered: September 2007

Junior Member

Hi,

by infinite process I mean following file

#include<stdio.h>

int main()
{

for(;;);
return 0;
}

so I run 40 processes of such kind with some having nice value 19, after that system says memory insufficient to spawn new process.

Still in this case /proc/vz/vestat shows 0 for maxlat, totallat n numsched. even in rickB's example in previous post, the file has 0 values.

this is my top command output in VE 1001 after running 37 instances of infinite process

top - 11:58:15 up 2:22, 0 users, load average: 33.96, 26.71, 13.17
Tasks: 51 total, 36 running, 15 sleeping, 0 stopped, 0 zombie
Cpu(s): 99.3% us, 0.3% sy, 0.3% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.0% si
Mem: 505200k total, 495700k used, 9500k free, 39208k buffers
Swap: 1084348k total, 0k used, 1084348k free, 187308k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
22084 root 25 0 1384 232 184 R 3.7 0.0 0:12.62 infinite
22036 root 25 0 1384 232 184 R 3.3 0.0 0:13.89 infinite
22037 root 25 0 1384 232 184 R 3.3 0.0 0:13.19 infinite
22038 root 25 0 1384 232 184 R 3.3 0.0 0:13.14 infinite
22039 root 25 0 1388 236 184 R 3.3 0.0 0:13.12 infinite
22040 root 25 0 1384 232 184 R 3.3 0.0 0:13.10 infinite
22043 root 25 0 1384 232 184 R 3.3 0.0 0:13.09 infinite
22044 root 25 0 1384 232 184 R 3.3 0.0 0:12.89 infinite
22046 root 25 0 1384 232 184 R 3.3 0.0 0:12.92 infinite
22049 root 25 0 1388 236 184 R 3.3 0.0 0:12.90 infinite
22050 root 25 0 1388 236 184 R 3.3 0.0 0:12.89 infinite
22051 root 25 0 1384 232 184 R 3.3 0.0 0:13.59 infinite
22052 root 25 0 1388 236 184 R 3.3 0.0 0:12.89 infinite
22065 root 25 0 1384 232 184 R 3.3 0.0 0:11.80 infinite
22066 root 25 0 1384 232 184 R 3.3 0.0 0:12.59 infinite
22067 root 25 0 1384 232 184 R 3.3 0.0 0:11.89 infinite
22068 root 25 0 1384 232 184 R 3.3 0.0 0:11.84 infinite

But still output of /proc/vz/vestat is -
[root@khyati-laptop home]# cat /proc/vz/vestat
Version: 2.2
VEID user nice system uptime idle strv uptime used maxlat totlat numsched
1001 990058 931 1922 8628377 2320326584544 0 3988923534912 1559444504772 0 0 0

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #32962 is a reply to message #2687]

Thu, 11 September 2008 01:55

gurtaj
Messages: 12
Registered: September 2008
Location: London, Ontario, Canada

Junior Member

Hello. I've found this message while looking for information about how to calculate the CPU usage percentage of a container (yeah... one more trying to do the same).

The fact is that I'm having some troubles, so I'd appreciate any help.

I started a container (# 109) and captured the following values:

VEID  user  nice  system  uptime  idle           strv  uptime         used       maxlat  totlat  numsched
109   9     0     30      225316  4494871181910  0     4494982391820  793267650  0       0       0

So, let's start with the calculations:

- if jiffies_per_second = 1000 (as indicated by dev), the first uptime value equals 225.316 seconds.

Anyway, I was trying to match both uptime values, the one in jiffies and the one in cycles.

As far as I understand, the second value should be divided by cycles_per_jiffy:

CPU: Intel Core Duo T2450 @ 2.00 GHz
CPU MHz: 1995.115

cycles_per_jiffy = 1995115 / 1000 = 1995.115
(which seems pretty small compared with the one given by dev as example of his processor, which I guess it was a 1.86 GHz)

Now, if I devide the second uptime value by the previously obtained cycles_per_jiffy:

seconds = uptime / cycles_per_jiffy = 4494982391820 / 1995.115 = 22527270349023.399653654

... OK, the value starts similar, but they are quite different.

So... what I'm understanding wrong?

Thanks for the help!

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #32984 is a reply to message #2687]

Sun, 14 September 2008 19:22

gurtaj
Messages: 12
Registered: September 2008
Location: London, Ontario, Canada

Junior Member

Anybody around?

I kept playing with the containers in this Ubuntu system.

I started a container, obtained the first uptime value, waited 60 seconds and obtained the uptime value again.

First measure: uptime = 2033
60 seconds later: uptime = 8030

The difference is 6000 jiffies (in 60 seconds), so there are 100 jiffies per second.

Report message to a moderator

Re: Where can we download the "vzstat" source code? [message #32985 is a reply to message #2687]

Sun, 14 September 2008 19:43

gurtaj
Messages: 12
Registered: September 2008
Location: London, Ontario, Canada

Junior Member

Following on the previous message...

The second uptime value (on cycles) in those two measures was:

First measure: uptime = 40566822150
60 seconds later: uptime = 160205154630

Dividing by the CPU Hz (should get seconds) obtained from /proc/cpuinfo, 1995067:

First measure: uptime = 20333.563810138
60 seconds later: uptime = 80300.638840701

As it can be seen, both uptime values from the first measure are similar, but differ on orders of magnitude.

uptime in jiffies = 2033 ==> 20.33 seconds
uptime in cycles = 40566822150
uptime in cycles/frequency ==> 20333.563810138 seconds

So, where am I missing the point?

Report message to a moderator