Help

Write Latency Delta chart
This chart provides the delta between Front-end latency and Back-end Latency per director. This is a key metric for Local/
Metro — the amount of overhead time metro node spends processing a write.
NOTE: The chart displays data only for the directors in the cluster to which you are currently connected. To simultaneously
view Write Latency Delta for another cluster, open a second browser session and connect to the second cluster.
Guidelines (metro node Local only)
● This is typically <1 msec
● 1-10 msec is concerning
● >10 msec should be investigated.
● Large spikes indicate metro node spent a long time adding processing time to the I/O — This is generally bad for
performance.
● During disruptive events, this value may go as high as a few seconds, which could be due to I/Os timing out in the back-end.
● Note that in a heavily loaded system, this value climb as the directors become more busy, especially if the directors approach
saturation (>95% director CPU busy).
● Large block writes (host I/O block sizes >128KB) add to extra metro node write processing time.
Guidelines (metro node Metro only)
● For distributed devices in write-through mode, this statistic includes the time required to write to the remote leg of the
distributed Raid (WAN latency plus remote storage array write latency). The write is not acknowledged to the host without
first writing to both legs of the distributed Raid, thus we only perform as fast as the slowest leg which in most cases will
be the remote leg on the other cluster. This is why WAN performance is so critical. It is good to also check WAN round-trip
latency.
● During disruptive events, this value may go as high as a few seconds, which could be due to I/Os timing out in the back-end,
or a poorly performing WAN link.
Corrective actions
● Check CPU load: If CPU load is too high, this will mean less CPU cycles available to process data.
● Check back-end errors, and back-end latency: If the metro node back-end is aborting I/Os, must re-try them and this wait
time is reflected in the metro node write delta time.
● Metro node Metro: Check WAN latency. The time spent writing to the remote leg of a distributed Raid is encapsulated in this
write delta time.
Changing the view
To view the Write Latency Delta of a single director in your metro node system, select the director name from the Director
drop-down.
Viewing the Write Latency Delta chart
1. From the GUI main menu, click Performance.
2. In the Performance Dashboard, select the tab in which you want to display the Write Latency Delta chart (or create a
custom tab).
3. Click +Add Content.
4. Click the Write Latency Delta icon.
WAN Port Performance chart
The WAN Port Performance chart on the Performance Dashboard shows the WAN port performance by IP or Fibre Channel port
for the cluster to which you are connected.
44
Monitoring the system