Argument-0005: Retention at Rank 1

Member details

• Matrix username: @serban.iorga:parity.io
• Polkadot address: 14oHMAJ5btnDCusHrTWraw1wTsLJwZeqPDLxusm1R1Zh3Vxa
• Current rank: 1
• Date of initial induction: 2024/04/22
• Date of last report: 2025/07/07
• Area(s) of Expertise/Interest:
  • Bridges
  • BEEFY
  • MMR
  • Omni-Node
  • XCMP/HRMP/DMP

Reporting period

• Start date: 2025/07/08
• End date: 2025/09/24

Argument

During this reporting period, I have been focusing mainly on reliability and efficiency improvements for the message passing mechanisms.

On the efficiency side I added support for receiving double encoded XCMs through XCMP. For context, at the moment parachains pass XCM messages between them through XCMP pages that use the XcmpMessageFormat::ConcatenatedVersionedXcm format. These pages contain concatenated encoded VersionedXcms and on the receiving side, in order to split the page into individual messages, we need to first decode them and then re-encode and forward them to the pallet-messages-queue. This adds extra overhead (about 2.5 microseconds + some cost per byte). My PR added a new (XcmpMessageFormat::ConcatenatedOpaqueVersionedXcm) format that will be used for pages with double-encoded XCMs. This makes the decoding much easier and almost free, improving the XCMP bandwidth. This new format is not used yet, and we'll have to come out with a roll-out plan next.

On the reliability side, I worked on some small fixes and improvements. Mentioning just a few:

• some small improvements related to the PoV accounting for the XCMP enqueueing logic: 1, 2.
• prepared a referenda for reducing the number of pending messages per HRMP channel. This would prevent overloading the receiving parachains with too many messages at once.
• limited the number of signals per XCMP page to 3: https://github.com/paritytech/polkadot-sdk/pull/9781.

I also continued the work related to limiting and optimizing the heap memory usage in Polkadot. Here, there already was a pending PR, but as part of the code review a completely different approach was suggested. In order to address this, I opened a new PR.

Voting record

Misc

• Question(s):

• Concern(s):

• Comment(s):