summaryrefslogtreecommitdiffstats
path: root/health/guides/netdev
diff options
context:
space:
mode:
Diffstat (limited to 'health/guides/netdev')
-rw-r--r--health/guides/netdev/1min_netdev_backlog_exceeded.md44
-rw-r--r--health/guides/netdev/1min_netdev_budget_ran_outs.md55
2 files changed, 99 insertions, 0 deletions
diff --git a/health/guides/netdev/1min_netdev_backlog_exceeded.md b/health/guides/netdev/1min_netdev_backlog_exceeded.md
new file mode 100644
index 00000000..dc7b6a2c
--- /dev/null
+++ b/health/guides/netdev/1min_netdev_backlog_exceeded.md
@@ -0,0 +1,44 @@
+### Understand the alert
+
+The linux kernel contains queues where packets are stored after reception from a network interface controller before being processed by the next protocol stack. There is one netdev backlog queue per CPU core. netdev_max_backlog defines the maximum number of packets that can enter the queue. Queues fill up when an interface receives packets faster than kernel can process them. The default netdev_max_backlog value should be 1000. However this may not be enough in cases such as:
+
+- Multiple interfaces operating at 1Gbps, or even a single interface at 10Gbps.
+
+- Lower powered systems process very large amounts of network traffic.
+
+Netdata monitors the average number of dropped packets in the last minute due to exceeding the netdev backlog queue.
+
+### Troubleshoot the alert
+
+- Increase the netdev_max_backlog value
+
+1. Check your current value:
+
+ ```
+ root@netdata~ # sysctl net.core.netdev_max_backlog
+ net.core.netdev_max_backlog = 1000
+ ```
+
+2. Try to increase it by a factor of 2.
+
+ ```
+ root@netdata~ # sysctl -w net.core.netdev_max_backlog=2000
+ ```
+
+3. Verify the change and test with the same workload that triggered the alarm originally.
+
+ ```
+ root@netdata~ # sysctl net.core.netdev_max_backlog
+ net.core.netdev_max_backlog = 2000
+ ```
+
+4. If this change works for your system, you could make it permanently.
+
+ Bump this `net.core.netdev_max_backlog=2000` entry under `/etc/sysctl.conf`.
+
+5. Reload the sysctl settings.
+
+ ```
+ root@netdata~ # sysctl -p
+ ```
+
diff --git a/health/guides/netdev/1min_netdev_budget_ran_outs.md b/health/guides/netdev/1min_netdev_budget_ran_outs.md
new file mode 100644
index 00000000..30539322
--- /dev/null
+++ b/health/guides/netdev/1min_netdev_budget_ran_outs.md
@@ -0,0 +1,55 @@
+### Understand the alert
+
+Your system communicates with the devices attached to it through interrupt requests. In a nutshell, when an interrupt occurs, the operating system stops what it was doing and starts addressing that interrupt.
+
+Network interfaces can receive thousands of packets per second. To avoid burying the system with thousands of interrupts, the Linux kernel uses the NAPI polling framework. In this way, we can replace hundreds of hardware interrupts with one poll by managing them with a few Soft Interrupt ReQuests (Soft IRQs). Ksoftirqd is a per-CPU kernel thread responsible for handling those unserved Soft Interrupt ReQuests (Soft IRQs). The Netdata agent inspects the average number of times Ksoftirqd ran out of netdev_budget or CPU time when there was still work to be done. This abnormality may cause packet overflow on the intermediate buffers and, as a result, drop packet on your network interfaces.
+
+The default value of the netdev_budget is 300. However, this may not be enough in some cases, such as:
+
+- Multiple interfaces operating at 1Gbps, or even a single interface at 10Gbps.
+
+- Lower powered systems processing very large amounts of network traffic.
+
+### NAPI polling mechanism.
+
+The design of NAPI allows the network driver to go into a polling mode, buffering the packets it receives into a ring-buffer, and raises a soft interrupt to start a NAPI polling cycle instead of being hard-interrupted for
+every packet. Linux kernel through NAPI will poll data from the buffer until the netdev_budget_usecs times out or the number of packets reaches the netdev_budget limit.
+
+- netdev_budget_usecs variable defines the maximum number of microseconds in one NAPI polling cycle.
+- netdev_budget variable defines the maximum number of packets taken from all interfaces in one polling cycle.
+
+### Troubleshoot the alert
+
+- Increase the netdev_budget value.
+
+1. Check your current value.
+
+ ```
+ root@netdata~ $ sysctl net.core.netdev_budget
+ net.core.netdev_budget = 300
+ ```
+
+2. Try to increase it gradually with increments of 100.
+
+ ```
+ root@netdata~ $ sysctl -w net.core.netdev_budget=400
+ ```
+
+3. Verify the change and test it with the same workload that triggered the alarm originally. If the problem still exists, try to
+ increment it again.
+
+ ```
+ root@netdata~ $ sysctl net.core.netdev_budget
+ net.core.netdev_budget = 400
+ ```
+
+4. If this change works for your system, you could make it permanently.
+
+ Bump this `net.core.netdev_budget=<desired_value>` entry under `/etc/sysctl.conf`
+
+
+5. Reload the sysctl settings.
+
+ ```
+ root@netdata~ $ sysctl -p
+ ``` \ No newline at end of file