Difference between revisions of "Status"
(→Sites network traffic)
|Line 87:||Line 87:|
== Latency monitoring ==
== Latency monitoring ==
Revision as of 17:07, 25 November 2016
Current events (maintenance, issues...)
If you experience problems, please check the grid administration schedule, where past, present and future incidents (planned or not...) are notified for all sites.
Monika displays current and scheduled OAR jobs.
You can select an individual site or cluster:
Or view the global snapshot of the grid.
Drawgantt displays past, current and future OAR jobs.
Or view the global grid Gantt diagram.
Backbone network status and load
Grid5000 Weathermap (courtesy of Renater)
Shows the actual state of the opticals links between the Grid5000 10Gb-ready sites. A link painted in black on the weathermap means that you won't be able to access this site nodes from the Grid5000 internal network.
Historical network load
Grid5000 Network monitoring User:GRID PASS:5000 (courtesy of Renater)
This page gives you some nice graphs built from the SNMP counters of Renater switches.
You can see on those graphs if one or several experiments are creating congestion on a switch interface. It is quite interesting if you experiment weird things like packet loss or anormal delay. The EoMPLS graphs are outdated since Orsay and Bordeaux have been migrated to Renater-5.
Sites network traffic
We are using the ping / ICMP probes in order to monitor the backbone's latency. Smokeping forks Fping every 300 seconds on each site in order to ping 20 times the adminfront of each site (including himself). Each site is trying to ping the others one, allowing us to get a full view of the network from each site.
Each host is pinged 20 times (similar to a ping -c 20) in order to study :
- Packet Loss (PL) occurring the link. Packet Loss can be caused by the saturation of a link. When a router or switch buffers are unable to store packets, the packets are dropped. It may also be caused by a faulty transmitting equipment somewhere, a link flapping (up/down/up/down...) caused by a faulty component.
- The variance between each ping. If the first comes back in 10ms, then the second in 50ms, then 5ms... there is something weird going on (network overload or routing issues). The graphs is plotting the median value then drawing smoke under and upper the point. If the median is 20ms, the min 10m and max 80ms, you will have a colored point at 20ms then smoke going from 10ms to 80ms.
A dashboard combining links and real-time data is also available on the Grid5000 Backbone Network Monitoring page.
Site availability over time gathers a lot of statistics about raw usage of the platform
Kaspied is a statistic tool provided to show who is using the platform.
Ganglia provides resources usage metrics (memory, cpu, jobs...) for individual sites or the whole grid.
Nagios monitors critical grid servers and services and automatically reports incidents and failures.