Erc20 bench #1174
Erc20 bench #1174
Conversation
tmontaigu
force-pushed
the
tm/erc20
branch
2 times, most recently
from
aa30e48
to
8119357
Compare
tmontaigu
force-pushed
the
tm/erc20
branch
5 times, most recently
from
bd28d22
to
ffa778e
Compare
This adds benchmarks of both latency and throughput of 4 variants of the erc20 transfer. It also prints the PBS count of each versions. - "whitepaper" is the variant written in the fhevm whitepaper - "no_cmux" is similar to whitepaper, but uses a "boolean multiplication" instead of a cmux - "overflow" uses an overflowing_sub to remove the need for comparison - "safe" use both overflowing_sub and overflowing_add to make sure both then sender has enough money and the that the transfer won't overflow the receiver's money "overflow" has the lowest latency, and second best throughput "no_cmux" has the second lowest latenc and the best throughput
| println!("ERC20 transfer/{fn_name}::{type_name}: {count} PBS"); | ||
| } | ||
|
|
||
| fn bench_transfer_latency<FheType, F>( |
There was a problem hiding this comment.
Are these benchmarks never meant to be stored in database ?
There was a problem hiding this comment.
hum I think we want to store them in a database but they are not integer bench so we likely need a different database/schema and we would need to store both the throughput and latency (and pbs count if posssible)
There was a problem hiding this comment.
Ok so we need to discuss how to add them.
Will add one
What do you mean by local ? You can run them on your laptop if you want but its not really going to be meaningfull especially for the thourhgput benches |
No problem, since it's just a target missing in the Makefile. I was wondering if it needed a more sophisticated infra than just one machine. |
This adds benchmarks of both latency and throughput
of 4 variants of the erc20 transfer.
It also prints the PBS count of each versions.
instead of a cmux
then sender has enough money and the that the transfer won't overflow
the receiver's money
"overflow" has the lowest latency, and second best throughput
"no_cmux" has the second lowest latenc and the best throughput