Difference between revisions of "Vestat"
(category) |
|||
| (3 intermediate revisions by 2 users not shown) | |||
| Line 2: | Line 2: | ||
The '''/proc/vz/vestat''' file contains statistics for VE CPU usage. | The '''/proc/vz/vestat''' file contains statistics for VE CPU usage. | ||
| − | The ''guessed'' content (may the Gods - blessed thy fingers - acknowledge or reject my humble info here): | + | The ''guessed'' content (may the Gods - blessed thy fingers - acknowledge or reject my humble info here) ( See http://forum.openvz.org/index.php?t=msg&goto=2687 (OpenVZ developer talks /proc/vz/vestat) for more info): |
* '''VEID''': The VE ID, obviously. | * '''VEID''': The VE ID, obviously. | ||
* '''user, nice, system, uptime''': usage in [[jiffy|jiffies]]. These are the equivalents of <code>/proc/stat</code>, but there is no ''idle'' because it cannot be measured this way. | * '''user, nice, system, uptime''': usage in [[jiffy|jiffies]]. These are the equivalents of <code>/proc/stat</code>, but there is no ''idle'' because it cannot be measured this way. | ||
| − | * '''idle, strv, uptime, used''': usage in [[cycle|cycles]]. idle and uptime are obvious, strv isn't used, and used is the used cycles by VE on all CPUs. | + | * '''idle, strv, uptime, used''': usage in [[cycle|cycles]]. ''idle'' and ''uptime'' are obvious, ''strv'' isn't used, and ''used'' is the used cycles by VE on all CPUs. |
| − | * '''maxlat, totlat, numsched''': latency statistics in '' | + | * '''maxlat, totlat, numsched''': latency statistics in cycles. ''maxlat'' is max latency in cycles meaning how long VE process has to wait before it actually got CPU time; ''totlat/numsched'' gives average scheduling latency. '''These do not seem work in OpenVZ, only in Virtuozzo.''' |
== Measure conversions == | == Measure conversions == | ||
| Line 14: | Line 14: | ||
Using cycles: | Using cycles: | ||
| − | *seconds = ''measurement'' / | + | *seconds = ''measurement'' / frequency_of_your_cpu |
Variables: | Variables: | ||
| − | * ''jiffies_per_second'' = 1000 '''(unless you have changed that in kernel config | + | * ''jiffies_per_second'' = 1000 '''(more often ''100'', unless you have changed that in kernel config, check CONFIG_HZ= in kernel .config or /proc/config.gz)''' |
* ''cycles_per_jiffy'' = ''frequency_of_your_cpu'' / ''jiffies_per_second'' | * ''cycles_per_jiffy'' = ''frequency_of_your_cpu'' / ''jiffies_per_second'' | ||
| − | * ''frequency_of_your_cpu'' (in Hz) can be read from /proc/ | + | * ''frequency_of_your_cpu'' (in Hz) can be read from /proc/cpuinfo, as: |
| − | ** <code>cpu MHz : '' | + | ** <code>cpu MHz : ''megahertz''</code>, and thus ''frequency_of_your_cpu'' = ''megahertz'' * 1048576 (for multiCPU machines this is based on the speed of one cpu only) |
Latest revision as of 12:35, 14 January 2010
The /proc/vz/vestat file contains statistics for VE CPU usage.
The guessed content (may the Gods - blessed thy fingers - acknowledge or reject my humble info here) ( See http://forum.openvz.org/index.php?t=msg&goto=2687 (OpenVZ developer talks /proc/vz/vestat) for more info):
- VEID: The VE ID, obviously.
- user, nice, system, uptime: usage in jiffies. These are the equivalents of
/proc/stat, but there is no idle because it cannot be measured this way. - idle, strv, uptime, used: usage in cycles. idle and uptime are obvious, strv isn't used, and used is the used cycles by VE on all CPUs.
- maxlat, totlat, numsched: latency statistics in cycles. maxlat is max latency in cycles meaning how long VE process has to wait before it actually got CPU time; totlat/numsched gives average scheduling latency. These do not seem work in OpenVZ, only in Virtuozzo.
Measure conversions[edit]
Using jiffies:
- seconds = measurement / jiffies_per_seconds
Using cycles:
- seconds = measurement / frequency_of_your_cpu
Variables:
- jiffies_per_second = 1000 (more often 100, unless you have changed that in kernel config, check CONFIG_HZ= in kernel .config or /proc/config.gz)
- cycles_per_jiffy = frequency_of_your_cpu / jiffies_per_second
- frequency_of_your_cpu (in Hz) can be read from /proc/cpuinfo, as:
cpu MHz : megahertz, and thus frequency_of_your_cpu = megahertz * 1048576 (for multiCPU machines this is based on the speed of one cpu only)
