Proper bare-metal instances for benchmarking workflows

[ghzlatarev]

Description

closes: #298

• Switch to c5d.metal for the benchmarking workflows, which is actually a bare-metal instance with the following specification:

Instance Name | Logical Processors | Memory  | Local Storage        | EBS-Optimized Bandwidth | Network Bandwidth
c5d.metal     | 96                 | 192 GiB | 4 x 900 GB NVMe SSD  | 14 Gbps                 | 25 Gbps

• The C5d instance runs on Intel’s Second Generation Xeon Scalable processors (code-named Cascade Lake) with sustained all-core turbo frequency of 3.6GHz and maximum single core turbo frequency of 3.9 GHz.

• Re-benchmark Calamari and Dolphin weights with the following jobs:

  • Calamari - https://github.com/Manta-Network/Manta/actions/runs/2185313764
  • Dolphin - https://github.com/Manta-Network/Manta/actions/runs/2185314353

---

Before we can merge this PR, please make sure that all the following items have been
checked off. If any of the checklist items are not applicable, please leave them but
write a little note why.

• Linked to Github issue with discussion and accepted design OR have an explanation in the PR that describes this work.
• Wrote unit tests.
• Updated relevant documentation in the code.
• Added one line describing your change in <branch>/CHANGELOG.md
• Re-reviewed Files changed in the Github PR explorer.
• If runtime changes, need to update the version numbers properly:
  • authoring_version: The version of the authorship interface. An authoring node will not attempt to author blocks unless this is equal to its native runtime.
  • spec_version: The version of the runtime specification. A full node will not attempt to use its native runtime in substitute for the on-chain Wasm runtime unless all of spec_name, spec_version, and authoring_version are the same between Wasm and native.
  • impl_version: The version of the implementation of the specification. Nodes are free to ignore this; it serves only as an indication that the code is different; as long as the other two versions are the same then while the actual code may be different, it is nonetheless required to do the same thing. Non-consensus-breaking optimizations are about the only changes that could be made which would result in only the impl_version changing.
  • transaction_version: The version of the extrinsics interface. This number must be updated in the following circumstances: extrinsic parameters (number, order, or types) have been changed; extrinsics or pallets have been removed; or the pallet order in the construct_runtime! macro or extrinsic order in a pallet has been changed. If this number is updated, then the spec_version must also be updated
• Verify benchmarks & weights have been updated for any modified runtime logics
• If import a new pallet, choose a proper module index for it, and allow it in BaseFilter. Ensure every extrinsic works from front-end. If there's corresponding tool, ensure both work for each other.
• If needed, update our Javascript/Typescript APIs. These APIs are offcially used by exchanges or community developers.
• If modifying existing runtime storage items, make sure to implement storage migrations for the runtime and test them with try-runtime. This includes migrations inhreited from upstream changes, and you can search the diffs for modifications of #[pallet::storage] items to check for any.

[ghzlatarev]

@Dengjianping @stechu @Garandor
you can find the benchmarking results from https://cockpit.expralite.pelagos.systems/system/terminal (Rob's office) in the following commit c4f9de1
The configuration is Ubuntu 20.04; Intel(R) Xeon(R) CPU E5-2609 0 @ 2.40GHz; 32GB RAM, HDD storage 15k spinning disks in hardware raid 0.

Overall the results are roughly 3 times worse than c5d.metal in the PR.
MantaPay is roughly 2x worse than c5d.metal

[Dengjianping]

LGTM!

[ghzlatarev]

You can see the results from the Hetzner machine in this commit f71d350
That configuration is exactly like Polkadot's required hardware:
AMD Ryzen 7 3700X 8-Core Processor 8 cores, 64GB ECC, NVMe SSD
And the results are actually better than the c5d.metal machine, which is expected as the RAM doesn't really matter and Polkadot's required CPU is actually better than the c5d.metal one.

[ghzlatarev]

We're going with the slightly more conservative c4d.metal weights.