Syscall Batching

[Latrasis]

At the current moment, each syscall costs an additional DELEGATECALL. Such that if we where to write to n storage keys, we would have to make a DELEGATECALL for each one. One way to reduce these costs would be to introduce batching.

Type of Batching:

There are three types of batching we would have to choose from:

• Batch by Capability Index where:
  • You can only batch syscalls that use the same cap and thus are the same type.
  • On error, the kernel reverts. On input, the kernel returns input indexed by order.
• Batch by Capability Type:
  • You can only batch syscalls that use caps with the same type.
  • On error, the kernel reverts. On input, the kernel returns input indexed by order.
• Batch by any Capability:
  • You can batch syscalls that use any caps.
  • On error, the kernel reverts. On input, the kernel returns input indexed by order.

Issues:

• Should we allow all syscalls to be batch-able?
• How do we do error handling?
• This may introduce a potential gas-limit attack in kernel-space.
• Can we introduce such a feature with the least amount of additional complexity?