Module polkadot_sdk_docs::guides::enable_elastic_scaling_mvp

How to enable elastic scaling MVP on a parachain.

Enable elastic scaling MVP for a parachain

This guide assumes full familiarity with Asynchronous Backing and its terminology, as defined in the Polkadot Wiki. Furthermore, the parachain should have already been upgraded according to the guide.

Quick introduction to elastic scaling

Elastic scaling is a feature that will enable parachains to seamlessly scale up/down the number of used cores. This can be desirable in order to increase the compute or storage throughput of a parachain or to lower the latency between a transaction being submitted and it getting built in a parachain block.

At present, with Asynchronous Backing enabled, a parachain can only include a block on the relay chain every 6 seconds, irregardless of how many cores the parachain acquires. Elastic scaling builds further on the 10x throughput increase of Async Backing, enabling collators to submit up to 3 parachain blocks per relay chain block, resulting in a further 3x throughput increase.

Current limitations of the MVP

The full implementation of elastic scaling spans across the entire relay/parachain stack and is still work in progress. The MVP is still considered experimental software, so stability is not guaranteed. If you encounter any problems, please open an issue. Below are described the current limitations of the MVP:

1. Limited core count. Parachain block authoring is sequential, so the second block will start being built only after the previous block is imported. The current block production is capped at 2 seconds of execution. Therefore, assuming the full 2 seconds are used, a parachain can only utilise at most 3 cores in a relay chain slot of 6 seconds. If the full execution time is not being used, higher core counts can be achieved.
2. Single collator requirement for consistently scaling beyond a core at full authorship duration of 2 seconds per block. Using the current implementation with multiple collators adds additional latency to the block production pipeline. Assuming block execution takes about the same as authorship, the additional overhead is equal the duration of the authorship plus the block announcement. Each collator must first import the previous block before authoring a new one, so it is clear that the highest throughput can be achieved using a single collator. Experiments show that the peak performance using more than one collator (measured up to 10 collators) is utilising 2 cores with authorship time of 1.3 seconds per block, which leaves 400ms for networking overhead. This would allow for 2.6 seconds of execution, compared to the 2 seconds async backing enabled. More experiments are being conducted in this space.
3. Trusted collator set. The collator set needs to be trusted until there’s a mitigation that would prevent or deter multiple collators from submitting the same collation to multiple backing groups. A solution is being discussed here.
4. Fixed scaling. For true elasticity, the parachain must be able to seamlessly acquire or sell coretime as the user demand grows and shrinks over time, in an automated manner. This is currently lacking - a parachain can only scale up or down by “manually” acquiring coretime. This is not in the scope of the relay chain functionality. Parachains can already start implementing such autoscaling, but we aim to provide a framework/examples for developing autoscaling strategies.

Another hard limitation that is not envisioned to ever be lifted is that parachains which create forks will generally not be able to utilise the full number of cores they acquire.

Using elastic scaling MVP

Prerequisites

• Ensure Asynchronous Backing is enabled on the network and you have enabled it on the parachain using crate::guides::async_backing_guide.
• Ensure the AsyncBackingParams.max_candidate_depth value is configured to a value that is at least double the maximum targeted parachain velocity. For example, if the parachain will build at most 3 candidates per relay chain block, the max_candidate_depth should be at least 6.
• Use a trusted single collator for maximum throughput.
• Ensure enough coretime is assigned to the parachain. For maximum throughput the upper bound is 3 cores.

Phase 1 is NOT needed if using the polkadot-parachain or polkadot-omni-node binary, or polkadot-omni-node-lib built from the latest polkadot-sdk release! Simply pass the --experimental-use-slot-based ([`polkadot_omni_node_lib::cli::Cli::experimental_use_slot_based`]) parameter to the command line and jump to Phase 2.

The following steps assume using the cumulus parachain template.

Phase 1 - (For custom parachain node) Update Parachain Node

This assumes you are using the latest parachain template.

This phase consists of plugging in the new slot-based collator.

1. In node/src/service.rs import the slot based collator instead of the lookahead collator.

use cumulus_client_consensus_aura::collators::slot_based::{
	self as slot_based, Params as SlotBasedParams,
};

2. In start_consensus()
   • Remove the overseer_handle param (also remove the OverseerHandle type import if it’s not used elsewhere).
   • Rename AuraParams to SlotBasedParams, remove the overseer_handle field and add a slot_drift field with a value of Duration::from_secs(1).
   • Replace the single future returned by aura::run with the two futures returned by it and spawn them as separate tasks:

slot_based::run::<Block, <AuraId as AppCrypto>::Pair, _, _, _, _, _, _, _, _, _>(params);

3. In start_parachain_node() remove the overseer_handle param passed to start_consensus.

Phase 2 - Activate fixed factor scaling in the runtime

This phase consists of a couple of changes needed to be made to the parachain’s runtime in order to utilise fixed factor scaling.

First of all, you need to decide the upper limit to how many parachain blocks you need to produce per relay chain block (in direct correlation with the number of acquired cores). This should be either 1 (no scaling), 2 or 3. This is called the parachain velocity.

If you configure a velocity which is different from the number of assigned cores, the measured velocity in practice will be the minimum of these two.

The chosen velocity will also be used to compute:

• The slot duration, by dividing the 6000 ms duration of the relay chain slot duration by the velocity.
• The unincluded segment capacity, by multiplying the velocity with 2 and adding 1 to it.

Let’s assume a desired maximum velocity of 3 parachain blocks per relay chain block. The needed changes would all be done in runtime/src/lib.rs:

1. Rename BLOCK_PROCESSING_VELOCITY to MAX_BLOCK_PROCESSING_VELOCITY and increase it to the desired value. In this example, 3.

   const MAX_BLOCK_PROCESSING_VELOCITY: u32 = 3;

2. Set the MILLISECS_PER_BLOCK to the desired value.

   const MILLISECS_PER_BLOCK: u32 =
       RELAY_CHAIN_SLOT_DURATION_MILLIS / MAX_BLOCK_PROCESSING_VELOCITY;

   Note: for a parachain which measures time in terms of its own block number, changing block time may cause complications, requiring additional changes. See here more information: crate::guides::async_backing_guide.

3. Increase the UNINCLUDED_SEGMENT_CAPACITY to the desired value.

   const UNINCLUDED_SEGMENT_CAPACITY: u32 = 2 * MAX_BLOCK_PROCESSING_VELOCITY + 1;