Difference between revisions of "Comparing Canister Cycles vs Performance Counter"
From Internet Computer Wiki
m |
m |
||
Line 71: | Line 71: | ||
# The Canister gets a refund for the unused reply bytes. | # The Canister gets a refund for the unused reply bytes. | ||
# The Canister has already been charged in step 5 for the reply execution and 5B instructions. | # The Canister has already been charged in step 5 for the reply execution and 5B instructions. | ||
− | # The | + | # The <code>n*21</code> represents System API calls costs. |
== References == | == References == |
Latest revision as of 19:37, 16 August 2023
The example walks through a scenario of charging cycles that includes inter-canister calls and how the performance counter relates to the work done by canisters.
Setup
- Canisters A and B are on the same subnet.
- The User sends an 1KiB Ingress Message to Canister A
- The Canister A calls Canister B with an 2KiB Update Call
- The Canister B responds to Canister A with an 300-byte reply
- The Canister A responds to User with an 500-byte reply
Example Cycles Flow
Naming conventions:
1K == 1'000
1KiB == 1'024 bytes
1M == 1'000'000
1MiB == 1'048'576 bytes
1B == 1'000'000'000
1T == 1'000'000'000'000
Example Step | Canister A Cycles Balance |
Canister A Perf Counter |
Canister B Cycles Balance |
Canister B Perf Counter |
---|---|---|---|---|
1. Initial state | 1T |
N/A 1) |
1T |
N/A 1)
|
2. Canister A receives 1KiB Ingress Message | -1,2M -1KiB*2K ≈999'996M |
N/A 1) |
1T |
N/A 1)
|
3. Canister A starts execution 2) | -590K -5B/10*4 ≈997'996M |
0 (5B limit) |
1T |
N/A 1)
|
4. Canister A executes 100M WASM instructions | 997'996M |
+100M=100M (5B limit) |
1T |
N/A 1)
|
5. Canister A sends 2KiB call to Canister B 3) | -260K -2KiB*1K -2MiB*1K -590K -5B/10*4 ≈993'897M |
+3*217) +2KiB ≈100M (5B limit) |
1T |
N/A 1)
|
6. Canister A finishes execution 4) | +(5B-100M)*4/10≈995'857M |
N/A 1) |
1T |
N/A 1)
|
7. Canister B starts execution 2) | 995'857M |
N/A 1) |
-590K -5B/10*4 ≈997'999M |
0 (5B limit) |
8. Canister B executes 100M WASM instructions | 995'857M |
N/A 1) |
997'999M |
+100M=100M (5B limit) |
9. Canister B sends 300-byte reply to Canister A | 995'857M |
N/A 1) |
997'999M |
+2*217) +300 ≈100M (5B limit) |
10. Canister B finishes execution 4) | 995'857M |
N/A 1) |
+(5B-100M)*4/10≈999'959M |
N/A 1)
|
11. Canister A receives a reply from Canister B 5) | +(2MiB-300)*1K≈997'953M |
N/A 1) |
999'959M |
N/A 1)
|
12. Canister A starts reply callback execution 6) | 997'953M |
0 (5B limit) |
999'959M |
N/A 1)
|
13. Canister A executes 100M WASM instructions | 997'953M |
+100M=100M (5B limit) |
999'959M |
N/A 1)
|
14. Canister A sends 500-byte reply to User | 997'953M |
+2*21 +500 ≈100M (5B limit) |
999'959M |
N/A 1)
|
15. Canister A finishes execution 4) | +(5B-100M)*4/10≈999'913M |
N/A 1) |
999'959M |
N/A 1)
|
Total: | ~87M Cycles for ~200M WASM Instructions Executed + messaging |
~41M Cycles for ~100M WASM Instructions Executed + messaging |
Notes:
- The Performance Counter is available only during execution, so it's marked as
N/A
when the execution has not started yet. - The Canister get charged:
- 590K
- for the update message execution
- 5B/10*4
- for 5B instructions it can potentially execute
- The Canister get charged:
- 260K
- for the inter-canister call
- 2KiB * 1K
- for byte transmission
- 2MiB * 1K
- for the maximum reply size
- 590K
- for the reply update message execution
- 5B/10*4
- for 5B instructions limit it can potentially execute in the reply handler
- The Canister gets a refund for the unused Instructions.
- The Canister gets a refund for the unused reply bytes.
- The Canister has already been charged in step 5 for the reply execution and 5B instructions.
- The
n*21
represents System API calls costs.