🚀 Introducing Throttle Statistics

With IBM FlashSystem 9.1.3, we introduce Throttle Statistics — a capability that brings visibility to I/O throttling behavior.

Throttle Statistics allows administrators to observe and quantify the impact of configured throttles through low-resolution historical data. These statistics are collected in XML format and appear in existing host and vdisk XML files alongside standard I/O statistics, allowing administrators to compare I/O performance and throttling impact in a single view.

What This Enables

With enhanced visibility into throttling behavior, administrators can now:

• Clearly understand the impact of throttling on I/O performance
• Distinguish throttle-induced latency from underlying storage latency
• Troubleshoot performance issues more efficiently
• Tune throttle limits with greater precision
• Plan capacity based on actual workload behavior

Answering Key Questions

• Which component in the I/O path is responsible for the throttling delay?
• How many I/Os are being queued by throttling?
• Is application latency influenced by throttling?
• What is the worst‑case delay experienced by an I/O during a given interval?
• Does I/O latency increase when throttling limits are reached?

Throttle Metrics Overview

Throttle Statistics provides detailed metrics that capture the volume, impact, and delay characteristics of throttled I/O operations, including:

• Number of I/Os evaluated and affected by throttling
• Amount of data transfer delayed due to throttling
• Number of queued or delayed I/Os
• I/O failures due to excessive delay
• Latency introduced by throttling (average and peak)

🛠 How Throttle Statistics Are Collected and Accessed

Once throttling is configured on any supported object (such as a host, host cluster, vdisk, mdisk group, or offload), statistics are automatically collected as part of the existing I/O statistics framework and updated for both the host and vdisk, since both endpoints experience the resulting delay.

Workflow

Step 1 — Configure Throttling

Administrators define throttling policies based on IOPS and bandwidth limits using existing CLI commands (mkthrottle).

Step 2 — Monitor Activity

As I/O flows through the system, throttled operations are tracked at the node level and reflected on both:

• The host issuing the I/O
• The vDisk serving the I/O

Step 3 — Automatic Collection

Throttle statistics are collected periodically (default: every 5 minutes) using the XML stats framework.

The interval can be adjusted with the startstats CLI.

Step 4 — Accessing Throttle Statistics

XML Files →

Collected data appears in the existing host and vDisk XML files, where throttle-related attributes are identified with the t prefix.

Below is an example of how throttle statistics appear within the XML file:

<host idx="0"

id="AB"

ro="1127" wo="43535172" wou="0" rb="46042" wb="348441120"

rl="74" wl="106365865" rlw="0" wlw="0" xl="1337651"

wxl="1337612" rxl="38"

uo="0" ub="0" uou="0" ul="0" ulw="0"

tiocnt="43536299" tbw="43560953" tqueuediocnt="43424052"

tfailediocnt="0" tmaxdelay="0" tcumdelay="99569892">

</host>

REST APIs → Enable programmatic access for external integrations.

Cloud Call Home (CCH) → Throttle statistics are part of the inventory data shared with

IBM Storage Insights.

Snap/DD → Throttle statistics are captured and available within Snap data.

🔍 Want to Explore More?

📄 Metric Definitions → Detailed descriptions of throttle statistics and additional metrics.

https://ibmdocs-test.dcs.ibm.com/docs/en/flashsystem_reviews_913_test?topic=monitoring-statistics-collection

🔗 Accessing Throttle Statistics via REST API → Learn how to retrieve throttle statisticsprogrammatically (example based on host XML fields)

https://ibmdocs-test.dcs.ibm.com/docs/en/flashsystem_reviews_911_test?topic=statistics-getting-performance
https://ibmdocs-test.dcs.ibm.com/docs/en/flashsystem_reviews_911_test/storageflashsystem911/gettingstarted/rest_api_hoststats.html

🤝 Collaborators

Kaushal Khachane (Kaushal.Khachane1@ibm.com)

Nisarga Lolage (NISARGA.LOLAGE1@ibm.com)