### Simulator

TODO: create a repo with it with Jan permission, or ask him to upload it

### Results legend

- **peak allocation count**: number of live allocations made by the allocator (not deallocated at the moment of counting) during the simulation period
- **peak requested space**: space in bytes allocated by the allocator at maximum (while the memory is still in use and not deallocated)
- **peak wasted space on padding**: extra bytes required to manage the requested space
- **Final allocations**: what remains in terms of number of allocations (without a deallocation counterpart) at the end of the exercise
- **Final requested space**: what space is still occupied at the end of the exercise by previous allocations
- **Final wasted space on padding**: what space is used for padding/metadata at the end of the exercise by previous allocations
- **Extra bytes allocated**: once the simulation is over, we do some random trials of filling up the rest of the allocator heap, given the space left after all alloc/dealloc sequence is over. We can't tell exactly how much extra bytes could be allocated at the end because that space depends on the size of the allocation, which corresponds to some extra padding.
- **Physical memory used**: number of pages touched in physical memory * regular page size (4096) - which is the physical space in RAM consumed by the simulation, measured at the end of it. We compute this by looking into `/proc/self/pagemap`, which contains 64 bits for each virtual page. Bit 63 says if the virtual page is present in RAM.

### Allocation patterns on AHP

1. Used #9363 to build `polkadot-parachain` and start an AHP full node (with `-lruntime_host_allocator=debug`, which based on some later exercises in non-AHP context, might increase the allocations/deallocations sequence artificially to some extent). I used `polkadot-stable2506` branch and same commits.
2. Extracted both relaychain & parachain allocation logs.
3. I used a simulator from @koute (code should follow in a repo soon), updated to read trough a file with many host allocations for AHP runtime, from a full node which runs for an hour, synced to the tip at that time.

#### asset-hub-polkadot parachain allocation patterns

*Runtime: 6e20bc52aaaafd1de82ba7d2a3c0fa39193787e240b93608489cf72a4c46a584* (this the current on chain runtime on AHP)

Running with allocator: legacy
  Peak allocation count: 5586
  Peak requested space: 25214682
  Peak wasted space on padding: 6788230
  Final allocations (that were not yet deallocated): 4393
  Final requested space (unallocated space): 25045617
  Final wasted space on padding: 6693135
  Physical memory used: 31035392
  Bump allocator stats: AllocationStats { bytes_allocated: 31773896, bytes_allocated_peak: 32044136, bytes_allocated_sum: 571192664, address_space_used: 32248168 }
  Extra bytes allocated: 378385375

Running with allocator: new
  Peak allocation count: 5586
  Peak requested space: 25214682
  Peak wasted space on padding: 89283
  Final allocations (that were not yet deallocated): 4393
  Final requested space (unallocated space): 25045617
  Final wasted space on padding: 68911
  Physical memory used: 25538560
  Bump allocator stats: AllocationStats { bytes_allocated: 0, bytes_allocated_peak: 0, bytes_allocated_sum: 0, address_space_used: 0 }
  Extra bytes allocated: 394245905

*Runtime: e260d17fcfa34f10503c91148a7bc2fd820e356295d2e18f828b5fa4190d47f7* (I am not sure which runtime is this, maybe some older version which was used for some syncing?)

Running with allocator: legacy
  Peak allocation count: 7852
  Peak requested space: 32659186
  Peak wasted space on padding: 23185710
  Final allocations (that were not yet deallocated): 7801
  Final requested space (unallocated space): 32604198
  Final wasted space on padding: 23154330
  Physical memory used: 36519936
  Bump allocator stats: AllocationStats { bytes_allocated: 55820936, bytes_allocated_peak: 55907704, bytes_allocated_sum: 283346320, address_space_used: 55953104 }
  Extra bytes allocated: 360474882

Running with allocator: new
  Peak allocation count: 7852
  Peak requested space: 32659186
  Peak wasted space on padding: 147502
  Final allocations (that were not yet deallocated): 7801
  Final requested space (unallocated space): 32604198
  Final wasted space on padding: 146394
  Physical memory used: 39440384
  Bump allocator stats: AllocationStats { bytes_allocated: 0, bytes_allocated_peak: 0, bytes_allocated_sum: 0, address_space_used: 0 }
  Extra bytes allocated: 386472515

#### asset-hub-polkadot relaychain allocation patterns

*Runtime: e260d17fcfa34f10503c91148a7bc2fd820e356295d2e18f828b5fa4190d47f7*

Running with allocator: legacy
  Peak allocation count: 27570
  Peak requested space: 159710372
  Peak wasted space on padding: 112773340
  Final allocations: 7549
  Final requested space: 159200521
  Final wasted space on padding: 112562375
  Physical memory used: 208474112
  Bump allocator stats: AllocationStats { bytes_allocated: 271823288, bytes_allocated_peak: 272569888, bytes_allocated_sum: 20547290680, address_space_used: 292902856 }
  Extra bytes allocated: 183737681

Running with allocator: new
  Peak allocation count: 27570
  Peak requested space: 159710372
  Peak wasted space on padding: 64415644
  Final allocations: 7549
  Final requested space: 159200521
  Final wasted space on padding: 64287543
  Physical memory used: 195022848
  Bump allocator stats: AllocationStats { bytes_allocated: 157230856, bytes_allocated_peak: 157558560, bytes_allocated_sum: 1364898648, address_space_used: 177034712 }
  Extra bytes allocated: 219745053

### Heavy contract execution with host allocator debug logs: https://github.com/paritytech/memory_exhaustion

1. Used #9363 + #9267 (commit b86340a) to run `staging-node-cli`.
2. Once the node is up, with `runtime_host_allocator=debug` target enabled, I deployed the contract written here: https://github.com/paritytech/memory_exhaustion

#### Extracted allocations where `staging-node-cli` is started with logging directive `runtime_host_allocator=debug`

Got around 500 millions allocation/deallocation operations in sequence, and replayed them with same simulator from Jan. The extrinsic for the contract call (which recurses for 25 times) doesn't seem to end/get finalized, but allocations/deallocations are continously logged (let it run for more than 24hours and it was still logging). If the allocator log target captures the allocations logs on `debug`, contracts execution might be slow and strange things can happen (e.g. infinite allocations logged while executing an extrinsic related to a contract call - which I saw in practice). I would consider the bellow exercise not very representative for how the allocations happen, but it is still useful to be mentioned for future reference.

*Runtime 01cefa9c6e629d4cf67dc635e800c19ad7aea335333802af4945ada00dca1f0b*

Running with allocator: legacy
  Peak allocation count: 16376
  Peak requested space: 321974359
  Peak wasted space on padding: 6350956
  Final allocations: 2855
  Final requested space: 317628029
  Final wasted space on padding: 6350955
  Physical memory used: 348807168
  Bump allocator stats: AllocationStats { bytes_allocated: 324001824, bytes_allocated_peak: 328308656, bytes_allocated_sum: 13785237856, address_space_used: 351468928 }
  Extra bytes allocated: 139785942

Running with allocator: new
  Peak allocation count: 16376
  Peak requested space: 321974359
  Peak wasted space on padding: 5285723
  Final allocations: 2855
  Final requested space: 317628029
  Final wasted space on padding: 5285443
  Physical memory used: 350203904
  Bump allocator stats: AllocationStats { bytes_allocated: 257951880, bytes_allocated_peak: 262277192, bytes_allocated_sum: 3068803176, address_space_used: 284035320 }
  Extra bytes allocated: 140989127

### Heavy contract execution w/o custom logging directives: https://github.com/paritytech/memory_exhaustion

Setup is same as for the execution with debug logging directives, only that I used for polkadot-sdk branch of #9267 at 032e005.

#### Extracted allocations where `staging-node-cli` is started w/o logging directives and host allocator logs for allocations/deallocations are logged as `warn` - so they show up by default in node logs

Looks like when logging allocator logs on `warn` the exercise finishes successfully, and the allocations are much lower (e.g. ~55MB peak requested space and ~60/56MB phyisical memory used).

*Runtime: ead5ededbf9840660462ca27ae6d8008e827f902c6974d940b2cfee5d4be84f8*

Running with allocator: legacy
  Peak allocation count: 16777
  Peak requested space: 55185677
  Peak wasted space on padding: 2580291
  Final allocations: 968
  Final requested space: 5131075
  Final wasted space on padding: 506613
  Physical memory used: 60407808
  Bump allocator stats: AllocationStats { bytes_allocated: 5645432, bytes_allocated_peak: 57772736, bytes_allocated_sum: 6860640944, address_space_used: 61215344 }
  Extra bytes allocated: 357079214

Running with allocator: new
  Peak allocation count: 16777
  Peak requested space: 55185677
  Peak wasted space on padding: 207033
  Final allocations: 968
  Final requested space: 5131075
  Final wasted space on padding: 19581
  Physical memory used: 56799232
  Bump allocator stats: AllocationStats { bytes_allocated: 0, bytes_allocated_peak: 0, bytes_allocated_sum: 0, address_space_used: 0 }
  Extra bytes allocated: 414225157