Add READMEs and Makefile to build and integrate LIKWID

This commit is contained in:
Thomas Roehl 2021-03-27 13:39:43 +01:00
parent 5c5e8b8f2d
commit bc5d4b288e
4 changed files with 94 additions and 25 deletions

View File

@ -9,33 +9,31 @@ Open questions:
# Configuration
Configuration is implemented using a single json document that is distributed over network and may be persisted as file.
Granularity can be either `node`, or `core`. Frequency can be set on a per measurement basis.
Supported metrics are documented [here](https://github.com/ClusterCockpit/cc-specifications/blob/master/metrics/lineprotocol.md).
``` json
{
"sink": "db.monitoring.center.de",
"report": {
levels: ["core","node"],
interval: 120
"sink": {
"user": "admin",
"password": "12345",
"host": "localhost",
"port": "8080",
"database": "testdb",
"type": "stdout"
},
"schedule": {
"core": {
"frequency": 30,
"duration": 10},
"node":{
"frequency": 60,
"duration": 20}
},
"metrics": [
"ipc",
"flops_any",
"clock",
"load",
"mem_bw",
"mem_used",
"net_bw",
"file_bw"
"interval" : 3,
"duration" : 1,
"collectors": [
"memstat",
"likwid",
"loadavg",
"netstat",
"ibstat",
"lustrestat"
]
}
```
All available collectors are listed in the above JSON. There are currently two sinks supported `influxdb` and `stdout`. The `interval` defines how often the metrics should be read and send to the sink. The `duration` tells collectors how long one measurement has to take. An example for this is the `likwid` collector which starts the hardware performance counter, waits for `duration` seconds and stops the counters again. For most systems, the `likwid` collector has to do two measurements, thus `interval` must be larger than two times `duration`.

View File

@ -1,7 +1,13 @@
# LIKWID version
LIKWID_VERSION = 5.1.0
# Target user for LIKWID's accessdaemon
DAEMON_USER=root
# Target group for LIKWID's accessdaemon
DAEMON_GROUP=root
#################################################
# No need to change anything below this line
#################################################
INSTALL_FOLDER = ./likwid
BUILD_FOLDER = ./likwid/build

53
collectors/README.md Normal file
View File

@ -0,0 +1,53 @@
This folder contains the collectors for the cc-metric-collector.
# `metricCollector.go`
The base class/configuration is located in `metricCollector.go`.
# Collectors
* `memstatMetric.go`: Reads `/proc/meminfo` to calculate the **node** metric `mem_used`
* `loadavgMetric.go`: Reads `/proc/loadavg` and submits the **node** metrics:
* `load_one`
* `load_five`
* `load_fifteen`
* `netstatMetric.go`: Reads `/proc/net/dev` and submits for all network devices (except loopback `lo`) the **node** metrics:
* `<dev>_bytes_in`
* `<dev>_bytes_out`
* `<dev>_pkts_in`
* `<dev>_pkts_out`
* `lustreMetric.go`: Reads Lustre's stats file `/proc/fs/lustre/llite/lnec-XXXXXX/stats` and submits the **node** metrics:
* `read_bytes`
* `read_requests`
* `write_bytes`
* `write_bytes`
* `open`
* `close`
* `getattr`
* `setattr`
* `statfs`
* `inode_permission`
* `infinibandMetric.go`: Reads InfiniBand LID from `/sys/class/infiniband/mlx4_0/ports/1/lid` and uses the `perfquery` command to read the **node** metrics:
* `ib_recv`
* `ib_xmit`
* `likwidMetric.go`: Reads hardware performance events using LIKWID. It submits **socket** and **cpu** metrics:
* `mem_bw` (socket)
* `power` (socket, Sum of RAPL domains PKG and DRAM)
* `flops_dp` (cpu)
* `flops_sp` (cpu)
* `flops_any` (cpu, `2*flops_dp + flops_sp`)
* `cpi` (cpu)
* `clock` (cpu)
# Installation
Only the `likwidMetric.go` requires preparation steps. For this, the `Makefile` can be used. The LIKWID build needs to be configured:
* Version of LIKWID in `LIKWID_VERSION`
* Target user for LIKWID's accessdaemon in `DAEMON_USER`. The user has to have enough permissions to read the `msr` and `pci` device files
* Target group for LIKWID's accessdaemon in `DAEMON_GROUP`
It performs the following steps:
* Download LIKWID tarball
* Unpacking
* Adjusting configuration for LIKWID build
* Build it
* Copy all required files into `collectors/likwid`
* The accessdaemon is installed with the suid bit set using `sudo` also into `collectors/likwid`

12
sinks/README.md Normal file
View File

@ -0,0 +1,12 @@
This folder contains the sinks for the cc-metric-collector.
# `sink.go`
The base class/configuration is located in `sink.go`.
# Sinks
There are currently two sinks shipped with the cc-metric-collector:
* `stdoutSink.go`: Writes all metrics to `stdout` in InfluxDB line protocol. The sink does not use https://github.com/influxdata/line-protocol to reduce the executed code for debugging
* `influxSink.go`: Writes all metrics to an InfluxDB database instance using a blocking writer. It uses https://github.com/influxdata/influxdb-client-go . Configuration for the server, port, user, password and database name are in the global configuration file
# Installation
Nothing to do, all sinks are pure Go code