Is this amount of DNS requests normal?

[DennisNurtio]

Hello, I am using the SDK (1.7.1) on a fleet of devices and see a lot of data usage on them all. I ran tshark (wireshark) on the devices to check the data usage. Almost all data usage seems to come from loads of DNS communication, two queries per second. Is this normal for the IoT SDK? And if not, do you know where I need to look to change this?

Best regards,
Dennis

[jmklix]

How much is "a lot of data usage"?
It will depend on how many messages you are sending/receiving. It does seem like more than I would expect from just the keep alive pings

[jmklix]

I'm not sure of the exact reason/specifics of the DNS bursts, but the common runtime (CRT) dependency was designed with some performance improvements for things that require large file transfers. For example the s3 service benefits from this with improved performance. This is overkill for the IoT sdks as you have noticed. I don't have a workaround to reduce this, but I have opened an internal ticket to the correct team to get this improved. I'm not exactly sure what the change will look like, but I will try to keep you updated with any changes.

[DennisNurtio]

Thanks for your help so far. I hope the team can pick this up soon, so we can save like 75% on our IoT data costs.

Should an issue be created from this to track it a bit easier?

[JonathanSiegler]

I observed a similar behavior with the aws-iot-device-sdk-cpp-v2. During the first seconds (at least 30s), the device is sending two dns queries (IP4/IP6) each second containing the same address (MQTT broker URL).
This is happening every time when the device is connecting to the AWS MQTT broker.

The analysis of the data consumption shows me that approximately 1/3 of the complete data consumption is related to dns. Since the devices are using a cellular network with strict data limits, that is really costly.

The log entries of the SDK are showing that the expiration date is updated correctly and with dns caching it's not necessary to create such a lot queries. Furthermore the address is never "used" again after the mqtt client has been established the connection. If I interpret the source code of host_resolver.c correct, it's hard coded that the queries are send each second until the address is not used for 30s. Why the expiration date is not considered there?

[DennisNurtio]

@jmklix Did you already get some feedback from the internal team?

[jmklix]

Yes, check out this PR

[bretambrose]

Solution proposal can be found here: awslabs/aws-c-io#559
Additional PRs:

• Support for host resolution override per connection awslabs/aws-c-http#434 and
• Resolution override awslabs/aws-c-mqtt#269

ensure default MQTT connections do not perform any unsolicited additional resolution attempts.

[bretambrose]

v1.13.0 contains changes to DNS behavior so that all MQTT connections (MQTT5, MQTT311, websockets or direct) override the default DNS resolution frequency in a way that prevents the extra checks. This applies only to the MQTT connection itself; secondary connections made in service of establishing that connection (in particular, http-based credentials providers like cognito, sts, or imds) remain unchanged and will still ping for 30 seconds if used. If credentials provider DNS behavior is problematic for you, we can revisit the overall issue. Absent that, we encourage you to try the new version and verify that it meets your needs/expectations.

[DennisNurtio]

Our first tests with the update look good. Here you can see the DNS requests no longer having a peek at each communication (as compared to the images above).

[JonathanSiegler]

I can also confirm that the changes in the SDK solve the problem. My test showed me that only one request is sent as expected

[github-actions[bot]]

Hello! Reopening this discussion to make it searchable.