SIMD-0296: Increase Transaction Size Limit

proposals/0296-increase-tx-size.md

@@ -0,0 +1,193 @@
+---
+simd: '0296'
+title: Increase Transaction Size
+authors:
+  - jacobcreech
+  - apfitzge
+category: Standard
+type: Core
+status: Review
+created: 2025-05-28
+feature: (fill in with feature key and github tracking issues once accepted)
+---
+
+## Summary
+
+Historically, messages transmitted over the network must not exceed the IPc6 MTU
+size to ensure a fast and reliable network delivery. Solana has used a 
+conservative MTU size of 1280 bytes, which after accounting for some of the
+overhead, leaves a maximum transaction size of 1232 bytes for data in the
+payload.
+
+With QUIC, we can now use a larger transaction size than the original 1280
+bytes. This proposal presents a new transaction size starting at 4096 bytes,
+and the new v1 transaction format required to support the larger size. v1 
+transactions are a direct upgrade from the legacy format, while proposing 
+a removal of Address Lookup Tables compared to v0 transactions.
+
+## Motivation
+
+The current transaction size limit of 1232 bytes is too small for many
+developer use cases. While there have been some attempts to artificially
+increase the size capacity with address lookup tables, the current cap still is
+artificially limiting the number of signatures and size of other data.
+
+Use cases that are currently limited by the transaction size include:
+
+- ZK proofs such as what is used within [Confidential
+ Balances](https://www.solana-program.com/docs/confidential-balances/zkps)
+- Untruncated Winternitz one-time signatures
+- Nested Multisig used by corporations under Squads
+- Onchain cryptography signature schemes without precompiles such as BLS
+
+and many more. A larger transaction size would also help remove the need for
+address lookup tables, which have been a source of complexity on the validator
+client as well as a frictional developer experience for many developers.
+
+A number of developers have also resorted to using Jito bundles to artificially
+increase the size of their transactions. This is a current workaround that
+developers are using, but it does not have the same guarantees as a transaction
+does on the protocol level, namely that bundles are not guaranteed to be atomic
+like singular transactions.
+
+For these reasons, increasing the current transaction size limit would enable
+developers to create new applications and use cases that were previously
+unfeasible.
+
+## New Terminology
+
+- 'v1 transaction' - A new transaction format that is designed to enable larger
+  transactions sizes while not having the address lookup table features 
+  introduced in v0 transactions.
+
+## Detailed Design
+
+Increasing the transaction size limit above the current MTU max size is
+possible with the use of QUIC. QUIC's RFC 9000 specification does not have an
+explicit maximum stream size, allowing for larger transactions to be sent.
+
+A new transaction format, `v1`, is proposed to enable larger transaction sizes.
+The `v1` transaction format would be:
+
+```
+VersionByte (u8) - >129 to distinguish from legacy/v0 formats
+LegacyHeader (u8, u8, u8) -- Required signatures from the current
+`MessageHeader` type
+Payload Length (u16) -- Total size excluding signatures
+NumInstructions (u8)
+TransactionConfigMask (u64) -- Bitmask of which config requests are present.
+ - 0: requested_compute_unit_limit
+ - 1: requested_compute_unit_price
+ - 2: requested_heap_size
+ - 3: requested_loaded_accounts_data_size_limit
+LifetimeSpecificier [u8; 32]
+NumAddresses (u8)
+Addresses [[u8; 32]] -- Number matches NumAddresses
+ConfigRequests [u64] -- Array of request values. (section size is popcount
+ TransactionConfigMask * 8). each value is a u64.
+Ixs [(u8, u8, u16, u16, u16)] -- Number matches NumInstructions. Values are 
+ (program_account_index, num_accounts, accounts_offset, num_data_bytes, 
+ bytes_offset)
+[TRAILING DATA SECTION] -- Length is such that PayloadLength is matched.
+Signatures [[u8; 64]] -- Length of `num_required_signatures` from
+ `LegacyHeader`
+```
+
+The [TRAILING DATA SECTION] contains instruction's serialized account indexes
+and instruction data bytes,
+in the order they appear in the transaction. Each instruction's account indexes
+must appear before its data bytes.
+For example, if there are two instructions:
+
+1. `Instruction { account_indexes: [0, 1], data: [2, 3, 4] }`
+2. `Instruction { account_indexes: [5, 6], data: [7, 8] }`
+
+The TRAILING DATA SECTION must look like this:
+
+```
+[0, 1, 5, 6, 2, 3, 4, 7, 8]
+```
+
+Lengths are serialized in the earlier `Ixs` section, and so are not repeated in
+the trailing data section.
+
+This new `v1` transaction format notably does not include address lookup
+tables.
+
+In consideration of what size the new transaction size limit should increase
+to, [Jito's bundle
+
+sizes](https://jitolabs.grafana.net/dashboard/snapshot/ISnwjbxw02UBLrj1xy4n4dFkAPyZ46ll?orgId=0&from=2025-05-30T18:45:00.000Z&to=2025-05-30T21:45:00.000Z&timezone=utc&var-cluster=mainnet&var-region=$__all&viewPanel=panel-40&inspect=panel-40&inspectTab=data)
+that were used to increase the overall transaction size should be considered.
+The following is the rough distribution of the transaction sizes:
+
+- 2048 bytes or lower - 50% of bundles
+- 6144 bytes or lower = 65% of all bundles
+- 9216 bytes or lower = 100% of all bundles
+
+In consideration of the above, the use cases mentioned within the motivation
+section, and the current page size for managing memory within a validator, a
+new transaction size limit of 4096 bytes is proposed. This new limit should
+cover a majority of use cases mentioned within the motivation section, as well
+as enable most of the developers using Jito bundles for larger transactions to
+migrate to the new transaction format. Similarly, since the new limit proposed
+can accomodate the max accounts used in ALTs directly in the transaction,
+developers are able to migrate from `v0` transactions to `v1` transactions.
+
+A number of changes are required to be made to the validator to support the new
+transaction size limit. Consensus would need to be updated to support the
+larger transaction size, as larger transactions included in a block would be
+marked as invalid by the cluster today. The scheduler would also need to be
+modified to  support the larger transaction sizes. While this proposal does not
+introduce a new fee structure around bytes in a transaction, the scheduler
+should prioritize larger transactions differently, requiring developers to pay
+a higher priority fee to land their transaction.
+
+Testing for this new transaction size limit should be done extensively to
+ensure that performance on the cluster is not adversely affected. This testing
+should include the different use cases mentioned within the motivation section,
+as well as the different sizes of transactions that are currently being used by
+developers on the network.
+
+## Alternatives Considered
+
+Alternative designs considered:
+
+- Having a transaction loading feature that would allow developers to load
+the transaction parts in a buffer and then be able to execute them at the
+end.
+This method is no longer considered as it requires a much higher level of
+latency on the application level and a decent amount of complexity within
+the validator
+- Bundles at the protocol level. This would not solve all problems that are
+ solved by larger transaction sizes. Bundles would still limit the ability for
+ developers to use cryptographic signature schemes that have large proof sizes.
+
+## Impact
+
+Developers would have to update with applications to use the new transaction
+format to take advantage of the larger transaction size. Those developers that
+had previously been using address lookup tables would be required to update the
+new transactions with the full address list instead of the address lookup table
+and its indices.
+
+## Security Considerations
+
+Larger transaction sizes could results in a number of bandwidth issues on the 
+cluster than need to be tested thoroughly to ensure that performance is not
+adversely affected.
+
+## Drawbacks 
+
+There is an argument to be made that an increase in the transaction size limit
+above the max MTU packet size would incur some network overhead due to
+fragmentation retransmits. For a validator, this could mean maintaining
+in-memory buffers for larger transactions compared to just receiving and
+processing a single UDP packet. These issues will need to be addressed in the
+scheduler and how priority fees for larger transactions are handled.
+
+## Backwards Compatibility
+
+As all of these changes are implemented with a new transaction format, the
+increase on the transaction size limit does not affect or break `v0` or `legacy`
+transactions.