Random vs Sequential | NetApp | Page 1

kindred rapids Mar 3, 2024, 11:42 AM

#

Hi,

Is there an easy way to display IO workload organized by rnd vs seq? for the whole cluster as well as for a single object (svm, volume, lun...)

swift topaz Mar 4, 2024, 11:28 AM

#

@kindred rapids The volume dashboard has these stats under the row titled Top Volume End-to-End QoS, as in the attached screenshot. Similarly, for LUN, we can retrieve statistics using the queries qos_sequential_writes{lun!=""} and qos_sequential_reads{lun!=""}. These are expressed as percentages. Therefore, subtracting the sequential percentage from 100 should give the random percentage.

kindred rapids Mar 4, 2024, 11:35 AM

#

Thanks Rahul,
So, these dashboards are only for top 5 volumes? There are no built in dashboards for all volumes or for an entire node/cluster?

swift topaz Mar 4, 2024, 11:39 AM

#

You can retrieve data for up to the top 500 volumes using a filter in this dashboard as attached. While these statistics are not available at the node/cluster level, we do have separate dashboards for Nodes and Clusters.

kindred rapids Mar 4, 2024, 11:47 AM

#

Thanks Rahul.
Ultimately, we want to know the total random and sequential workload on a cluster for sizing. I was hoping to have a statistic for it, but it seems the best scenario is to sum the top 500 objects?

swift topaz Mar 4, 2024, 11:53 AM

#

What I meant is that the default dashboard allows you to select the top 500 for these panels. All the data is available in Prometheus. You can query and extract as much data as you want from Prometheus using PromQL. The qos_sequential_writes and qos_sequential_reads are percentages, so a sum may not be what you're looking for at the cluster/node level.

opaque onyx Mar 4, 2024, 4:25 PM

#

If you're doing sizing, I'd probably just get a perfstat honestly.

#

It's not supported but it's only not supported in title only. We still ask for it sometimes on very rare occassions in perf support.

kindred rapids Mar 4, 2024, 5:41 PM

#

Thanks Paul, the issue is that perfstat only collects realtime data and does not allow for historical outlook. Still, it's a great tool and shame it's not officially supported anymore

opaque onyx Mar 4, 2024, 7:06 PM

#

Perf archives largely support 98% of former perfstat needs, but sizing is one of those that it doesn't.

#Random vs Sequential