aboutsummaryrefslogtreecommitdiffstats
path: root/docs/troubleshooting/cpuusage.rst
diff options
context:
space:
mode:
authorJohn DeNisco <jdenisco@cisco.com>2018-07-26 12:45:10 -0400
committerDave Barach <openvpp@barachs.net>2018-07-26 18:34:47 +0000
commit06dcd45ff81e06bc8cf40ed487c0b2652d346a5a (patch)
tree71403f9d422c4e532b2871a66ab909bd6066b10b /docs/troubleshooting/cpuusage.rst
parent1d65279ffecd0f540288187b94cb1a6b84a7a0c6 (diff)
Initial commit of Sphinx docs
Change-Id: I9fca8fb98502dffc2555f9de7f507b6f006e0e77 Signed-off-by: John DeNisco <jdenisco@cisco.com>
Diffstat (limited to 'docs/troubleshooting/cpuusage.rst')
-rw-r--r--docs/troubleshooting/cpuusage.rst112
1 files changed, 112 insertions, 0 deletions
diff --git a/docs/troubleshooting/cpuusage.rst b/docs/troubleshooting/cpuusage.rst
new file mode 100644
index 00000000000..b9b8942a3dd
--- /dev/null
+++ b/docs/troubleshooting/cpuusage.rst
@@ -0,0 +1,112 @@
+.. _cpuusage:
+
+**************
+CPU Load/Usage
+**************
+
+There are various commands and tools that can help users see FD.io VPP CPU and memory usage at runtime.
+
+Linux top/htop
+==============
+
+The Linux top and htop are decent tools to look at FD.io VPP cpu and memory usage, but they will only show
+preallocated memory and total CPU usage. These commands can be useful to show which cores VPP is running on.
+
+This is an example of VPP instance that is running on cores 8 and 9. For this output type **top** and then
+type **1** when the tool starts.
+
+.. code-block:: console
+
+ $ top
+
+ top - 11:04:04 up 35 days, 3:16, 5 users, load average: 2.33, 2.23, 2.16
+ Tasks: 435 total, 2 running, 432 sleeping, 1 stopped, 0 zombie
+ %Cpu0 : 1.0 us, 0.7 sy, 0.0 ni, 98.0 id, 0.0 wa, 0.0 hi, 0.3 si, 0.0 st
+ %Cpu1 : 2.0 us, 0.3 sy, 0.0 ni, 97.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu2 : 0.7 us, 1.0 sy, 0.0 ni, 98.3 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu3 : 1.7 us, 0.7 sy, 0.0 ni, 97.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu4 : 2.0 us, 0.7 sy, 0.0 ni, 97.4 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu5 : 3.0 us, 0.3 sy, 0.0 ni, 96.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu6 : 2.3 us, 0.7 sy, 0.0 ni, 97.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu7 : 2.6 us, 0.3 sy, 0.0 ni, 97.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu8 : 96.0 us, 0.3 sy, 0.0 ni, 3.6 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu9 :100.0 us, 0.0 sy, 0.0 ni, 0.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ %Cpu10 : 1.0 us, 0.3 sy, 0.0 ni, 98.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
+ ....
+
+VPP Memory Usage
+================
+
+For details on VPP memory usage you can use the **show memory** command
+
+This is the example VPP memory usage on 2 cores.
+
+.. code-block:: console
+
+ # vppctl show memory verbose
+ Thread 0 vpp_main
+ 22043 objects, 17878k of 20826k used, 2426k free, 2396k reclaimed, 346k overhead, 1048572k capacity
+ alloc. from small object cache: 22875 hits 39973 attempts (57.23%) replacements 5143
+ alloc. from free-list: 44732 attempts, 26017 hits (58.16%), 528461 considered (per-attempt 11.81)
+ alloc. from vector-expand: 3430
+ allocs: 52324 2027.84 clocks/call
+ frees: 30280 594.38 clocks/call
+ Thread 1 vpp_wk_0
+ 22043 objects, 17878k of 20826k used, 2427k free, 2396k reclaimed, 346k overhead, 1048572k capacity
+ alloc. from small object cache: 22881 hits 39984 attempts (57.23%) replacements 5148
+ alloc. from free-list: 44736 attempts, 26021 hits (58.17%), 528465 considered (per-attempt 11.81)
+ alloc. from vector-expand: 3430
+ allocs: 52335 2027.54 clocks/call
+ frees: 30291 594.36 clocks/call
+
+VPP CPU Load
+============
+
+To find the VPP CPU load or how busy VPP is use the **show runtime** command.
+
+With at least one interface in polling mode, the VPP CPU utilization is always 100%.
+
+A good indicator of CPU load is **"average vectors/node"**. A bigger number means VPP
+is more busy but also more efficient. The Maximum value is 255 (unless you change VLIB_FRAME_SIZE in code).
+It basically means how many packets are processed in batch.
+
+If VPP is not loaded it will likely poll so fast that it will just get one or few
+packets from the rx queue. This is the case shown below on Thread 1. As load goes up vpp
+will have more work to do, so it will poll less frequently, and that will result in more
+packets waiting in rx queue. More packets will result in more efficient execution of the
+code so number of clock cycles / packet will go down. When "average vectors/node" goes up
+close to 255, you will likely start observing rx queue tail drops.
+
+.. code-block:: console
+
+ # vppctl show run
+ Thread 0 vpp_main (lcore 8)
+ Time 6152.9, average vectors/node 0.00, last 128 main loops 0.00 per node 0.00
+ vector rates in 0.0000e0, out 0.0000e0, drop 0.0000e0, punt 0.0000e0
+ Name State Calls Vectors Suspends Clocks Vectors/Call
+ acl-plugin-fa-cleaner-process event wait 0 0 1 3.66e4 0.00
+ admin-up-down-process event wait 0 0 1 2.54e3 0.00
+ ....
+ ---------------
+ Thread 1 vpp_wk_0 (lcore 9)
+ Time 6152.9, average vectors/node 1.00, last 128 main loops 0.00 per node 0.00
+ vector rates in 1.3073e2, out 1.3073e2, drop 6.5009e-4, punt 0.0000e0
+ Name State Calls Vectors Suspends Clocks Vectors/Call
+ TenGigabitEthernet86/0/0-outpu active 804395 804395 0 6.17e2 1.00
+ TenGigabitEthernet86/0/0-tx active 804395 804395 0 7.29e2 1.00
+ arp-input active 2 2 0 3.82e4 1.00
+ dpdk-input polling 24239296364 804398 0 1.59e7 0.00
+ error-drop active 4 4 0 4.65e3 1.00
+ ethernet-input active 2 2 0 1.08e4 1.00
+ interface-output active 1 1 0 3.78e3 1.00
+ ip4-glean active 1 1 0 6.98e4 1.00
+ ip4-icmp-echo-request active 804394 804394 0 5.02e2 1.00
+ ip4-icmp-input active 804394 804394 0 4.63e2 1.00
+ ip4-input-no-checksum active 804394 804394 0 8.51e2 1.00
+ ip4-load-balance active 804394 804394 0 5.46e2 1.00
+ ip4-local active 804394 804394 0 5.79e2 1.00
+ ip4-lookup active 804394 804394 0 5.71e2 1.00
+ ip4-rewrite active 804393 804393 0 5.69e2 1.00
+ ip6-input active 2 2 0 5.72e3 1.00
+ ip6-not-enabled active 2 2 0 1.56e4 1.00
+ unix-epoll-input polling 835722 0 0 3.03e-3 0.00