Skip to content

Commit 3e43b18

Browse files
will-break-itwill-break-it
authored
Costs (#300)
* docs: refactored egress cost for Praos * docs: refactored egress cost for Leios * docs: fix formatting of formulas * docs: format egress * docs: format egress formulas * docs: format egress formulas * docs: format notes * docs: fix formatting * docs: fix equation * docs: fix equation * docs: fix equation * docs: fix equation * docs: fix equation * docs: fix equation * docs: fix important note * docs: fix equation * docs: update votes * docs: update todo cost items * docs: restructured * docs: fix note * docs: remove vs praos column * docs: compute cpu cost * docs: compute cpu cost * docs: fix note * docs(cost): add storage * docs(cost): format storage * docs(cost): fix storage eb calculation * docs: fix conversion rounding * fix: docs * docs: add ram & iops * docs: update note * docs: add cost table * docs: add revenue estimate * docs: remove table * docs: fix text subscripts * docs: added note for additional compute operations (mempool & ledger) * docs: remove rb per stage * docs: refactored iops * docs: fix formula for eb and rb iops
1 parent effd76a commit 3e43b18

File tree

10 files changed

+1488
-541
lines changed

10 files changed

+1488
-541
lines changed

‎docs/cost-estimate/01-compute-cpu.md‎

Lines changed: 242 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,242 @@
1+
# Compute (CPU) cost estimation per node
2+
3+
> [!Note]
4+
> This analysis assumes fully utilized block space for conservative
5+
> upper bound estimation.
6+
7+
All CPU time values in this document are derived from the simulation
8+
configuration file (`data/simulation/config.default.yaml`), which contains
9+
benchmark measurements for protocol operations.
10+
11+
## Ouroboros Praos
12+
13+
We start with Ouroboros Praos to establish a baseline. In Praos, CPU utilization
14+
comes from block validation:
15+
16+
| Component | Validation Time | Source |
17+
| ------------------------------ | --------------- | ----------------------------------- |
18+
| Block header validation | 1ms | Same as IB header validation |
19+
| Block body validation constant | 50ms | Same as IB body validation constant |
20+
| Block body validation per byte | 0.0005ms | Same as IB body validation per byte |
21+
22+
### Block Rate Calculation
23+
24+
| Parameter | Value | Formula |
25+
| ----------------------- | -------- | ------------------ |
26+
| Slot length | 1 second | Protocol parameter |
27+
| Active slot coefficient | 0.05 | Protocol parameter |
28+
| **Blocks per second** | **0.05** | $$1 \times 0.05$$ |
29+
30+
### CPU Utilization Formula
31+
32+
$$C_{\text{praos}} = N_{\text{blocks}} \times (H_{\text{validation}} + B_{\text{validation}})$$
33+
34+
where:
35+
36+
- $N_{\text{blocks}}$ = Blocks per second
37+
- $H_{\text{validation}}$ = Header validation time in ms
38+
- $B_{\text{validation}}$ = Body validation time in ms (constant + per-byte component)
39+
40+
### Praos CPU Calculation (0.05 blocks/s)
41+
42+
$$C_{\text{praos}} = 0.05 \times (1 + (50 + 90,112 \times 0.0005))$$
43+
44+
$$C_{\text{praos}} = 0.05 \times (1 + 95.06)$$
45+
46+
$$C_{\text{praos}} = 0.05 \times 96.06 \approx 4.8\text{ ms/s}$$
47+
48+
Thus, Praos at 0.05 blocks/s consumes approximately 4.8ms of CPU time per
49+
second, which is about 0.48% of a single CPU core.
50+
51+
## Ouroboros Leios
52+
53+
In Leios, CPU utilization is distributed across multiple components: Input
54+
Blocks (IBs), Endorsement Blocks (EBs), Votes, and Certificates in Ranking
55+
Blocks (RBs).
56+
57+
### Component Processing Times
58+
59+
| Component | Size / Unit | Validation Time | Config Reference |
60+
| ----------- | ---------------- | -------------------- | -------------------------------------- |
61+
| IB Header | 304 bytes | 1ms | `ib-head-validation-cpu-time-ms` |
62+
| IB Body | 98,304 bytes | 50ms + 0.0005ms/byte | `ib-body-validation-cpu-time-ms-*` |
63+
| EB | 240 bytes + refs | 1ms | `eb-validation-cpu-time-ms` |
64+
| Vote | ~150 bytes | 0.816ms | `vote-validation-cpu-time-ms` |
65+
| Certificate | 7,168 bytes | 140ms | `cert-validation-cpu-time-ms-constant` |
66+
67+
### Base Parameters
68+
69+
| Parameter | Value | Source |
70+
| --------------- | -------------- | ----------------------------- |
71+
| Stage length | 20 slots (20s) | `leios-stage-length-slots` |
72+
| EBs per stage | 1.5 | `eb-generation-probability` |
73+
| Votes per EB | 600 | `vote-generation-probability` |
74+
| Votes per stage | 900 | 600 votes/EB × 1.5 EBs/stage |
75+
76+
### Base CPU Usage Formulas
77+
78+
1. **IB Processing CPU**:
79+
80+
$$C_{\text{ib}} = N_{\text{ib}} \times (H_{\text{ib}} + B_{\text{ib}})$$
81+
where:
82+
- $N_{\text{ib}}$ = Number of IBs per second
83+
- $H_{\text{ib}}$ = IB header validation time (1ms)
84+
- $B_{\text{ib}}$ = IB body validation time (50ms + 98,304×0.0005ms = 99.15ms)
85+
86+
2. **EB Processing CPU**:
87+
88+
$$C_{\text{eb}} = \frac{N_{\text{eb}} \times T_{\text{eb}}}{S_{\text{length}}}$$
89+
where:
90+
- $N_{\text{eb}}$ = Number of EBs per stage (1.5)
91+
- $T_{\text{eb}}$ = EB validation time (1ms)
92+
- $S_{\text{length}}$ = Stage length in seconds (20s)
93+
94+
3. **Vote Processing CPU**:
95+
96+
$$C_{\text{vote}} = \frac{N_{\text{votes}} \times T_{\text{vote}}}{S_{\text{length}}}$$
97+
where:
98+
- $N_{\text{votes}}$ = Number of votes per stage (900)
99+
- $T_{\text{vote}}$ = Vote validation time (0.816ms)
100+
- $S_{\text{length}}$ = Stage length in seconds (20s)
101+
102+
4. **Certificate Processing CPU**:
103+
104+
$$C_{\text{cert}} = \frac{N_{\text{cert}} \times T_{\text{cert}}}{S_{\text{length}}}$$
105+
where:
106+
- $N_{\text{cert}}$ = Number of certificates per stage (1)
107+
- $T_{\text{cert}}$ = Certificate validation time (140ms)
108+
- $S_{\text{length}}$ = Stage length in seconds (20s)
109+
110+
5. **Total CPU Usage**:
111+
112+
$$C_{\text{total}} = C_{\text{ib}} + C_{\text{eb}} + C_{\text{vote}} + C_{\text{cert}}$$
113+
114+
### CPU Time Calculation at 0.05 IB/s (equivalent to Praos rate)
115+
116+
1. **IB Processing**:
117+
118+
$$C_{\text{ib}} = 0.05 \times (1 + 99.15) = 5.01\text{ ms/s}$$
119+
120+
2. **EB Processing**:
121+
122+
$$C_{\text{eb}} = \frac{1.5 \times 1}{20} = 0.075\text{ ms/s}$$
123+
124+
3. **Vote Processing**:
125+
126+
$$C_{\text{vote}} = \frac{900 \times 0.816}{20} = 36.72\text{ ms/s}$$
127+
128+
4. **Certificate Processing**:
129+
130+
$$C_{\text{cert}} = \frac{1 \times 140}{20} = 7.0\text{ ms/s}$$
131+
132+
5. **Total CPU Usage**:
133+
134+
$$C_{\text{total}} = 5.01 + 0.075 + 36.72 + 7.0 = 48.81\text{ ms/s}$$
135+
136+
Thus, at 0.05 IB/s, Leios consumes approximately 48.8ms of CPU time per second,
137+
which is about 4.9% of a single CPU core.
138+
139+
### CPU Utilization at Different IB Rates
140+
141+
| IB/s | IB Processing | EB Processing | Vote Processing | Certificate Processing | Total CPU Time/s | % of One Core | Min Cores Needed | Recommended Cores |
142+
| ---- | ------------- | ------------- | --------------- | ---------------------- | ---------------- | ------------- | ---------------- | ----------------- |
143+
| 0.05 | 5.01ms | 0.075ms | 36.72ms | 7.0ms | 48.81ms | 4.9% | 1 | 1 |
144+
| 1 | 100.15ms | 0.075ms | 36.72ms | 7.0ms | 143.95ms | 14.4% | 1 | 2 |
145+
| 5 | 500.75ms | 0.075ms | 36.72ms | 7.0ms | 544.55ms | 54.5% | 1 | 2 |
146+
| 10 | 1,001.5ms | 0.075ms | 36.72ms | 7.0ms | 1,045.30ms | 104.5% | 2 | 4 |
147+
| 20 | 2,003ms | 0.075ms | 36.72ms | 7.0ms | 2,046.80ms | 204.7% | 3 | 6 |
148+
| 30 | 3,004.5ms | 0.075ms | 36.72ms | 7.0ms | 3,048.30ms | 304.8% | 4 | 8 |
149+
150+
> [!Note]
151+
>
152+
> - "Min Cores Needed" is calculated by dividing total CPU time by 1000ms and
153+
> rounding up
154+
> - "Recommended Cores" adds redundancy for spikes and other node operations
155+
> - This assumes even distribution of work and perfect parallelization, which is
156+
> optimistic.
157+
158+
### CPU Component Analysis (at 0.05 IB/s)
159+
160+
| Component | CPU Time | % of Total |
161+
| ---------------------- | -------- | ---------- |
162+
| IB Processing | 5.01ms | 10.3% |
163+
| EB Processing | 0.075ms | 0.2% |
164+
| Vote Processing | 36.72ms | 75.2% |
165+
| Certificate Processing | 7.0ms | 14.3% |
166+
167+
This shows that at the Praos-equivalent rate of 0.05 IB/s, vote processing
168+
dominates CPU usage (75.2%), followed by certificate processing (14.3%) and IB
169+
processing (10.3%).
170+
171+
> [!Note]
172+
> This analysis does not yet include estimates for:
173+
> - Transaction and mempool operations
174+
> - Ledger computation and state reconciliation
175+
>
176+
> These components will contribute additional CPU overhead in a complete implementation.
177+
178+
### Comparative Efficiency (Leios vs. Praos)
179+
180+
| IB/s | Leios CPU Time | Praos Equivalent CPU Time | Ratio (Leios:Praos) |
181+
| ---- | -------------- | ------------------------- | ------------------- |
182+
| 0.05 | 48.8ms | 4.8ms | 10.17:1 |
183+
| 1 | 144.0ms | 96.1ms | 1.50:1 |
184+
| 5 | 544.6ms | 480.3ms | 1.13:1 |
185+
| 10 | 1,045.3ms | 960.6ms | 1.09:1 |
186+
187+
At 0.05 IB/s, Leios requires significantly more CPU than Praos, primarily due to
188+
vote processing and certificate overhead. At higher throughput rates, the
189+
relative overhead of votes and certificates diminishes, making Leios approach
190+
the per-transaction efficiency of Praos.
191+
192+
## Monthly Cost by Cloud Provider ($)
193+
194+
Using standard compute-optimized instances:
195+
196+
| Provider | 2 Core | 4 Core | 8 Core | Notes |
197+
| ---------- | ------ | ------- | ------- | ---------------------- |
198+
| AWS c6i | $62.05 | $124.10 | $248.20 | On-demand, US East |
199+
| GCP c2/n2 | $52.34 | $152.35 | $304.78 | On-demand, US Central1 |
200+
| Azure Fsv2 | $61.76 | $123.37 | $246.74 | On-demand, East US |
201+
| DO CPU-Opt | $42.00 | $84.00 | $168.00 | Regular pricing |
202+
| Linode | $36.00 | $60.00 | $120.00 | Standard pricing |
203+
| Hetzner | $5.40 | $17.80 | $32.90 | Standard pricing |
204+
205+
> [!Note]
206+
>
207+
> - Prices are for US regions and may vary by location
208+
> - Assumes dedicated compute-optimized instances
209+
> - Does not include potential savings from reserved instances or spot pricing
210+
211+
## Recommendations
212+
213+
1. For 0.05-1 IB/s: 2 cores should be sufficient ($5-$62/month)
214+
2. For 1-10 IB/s: 4 cores recommended ($18-$152/month)
215+
3. For 10-30 IB/s: 8 cores recommended ($33-$305/month)
216+
4. Above 30 IB/s: Consider multiple nodes or higher-end instances
217+
218+
> [!Important]
219+
>
220+
> Key considerations:
221+
> Real-world performance may require more cores due to:
222+
>
223+
> - Network stack overhead
224+
> - OS operations
225+
> - Uneven workload distribution
226+
> - Memory bandwidth limitations
227+
> - Other node operations not included in calculation
228+
229+
## Compute Cost Sources
230+
231+
| Provider | Instance Type | Source | Last Updated |
232+
| ------------ | ------------- | -------------------------------------------------------------------- | ------------ |
233+
| AWS | c6i | https://aws.amazon.com/ec2/pricing/on-demand/ | Apr 2025 |
234+
| GCP | c2/n2 | https://cloud.google.com/compute/vm-instance-pricing | Apr 2025 |
235+
| Azure | Fsv2 | https://azure.microsoft.com/pricing/details/virtual-machines/series/ | Apr 2025 |
236+
| DigitalOcean | CPU-Optimized | https://www.digitalocean.com/pricing/compute | Apr 2025 |
237+
| Linode | Dedicated CPU | https://www.linode.com/pricing/ | Apr 2025 |
238+
| Hetzner | CPX | https://www.hetzner.com/cloud/pricing | Apr 2025 |
239+
240+
Note: Prices shown are for US regions and may vary by location. Many providers
241+
offer significant discounts for reserved instances or longer-term commitments.
242+
The table shows standard on-demand rates for compute-optimized instances.

‎docs/cost-estimate/01-introduction.md‎

Lines changed: 0 additions & 154 deletions
This file was deleted.

0 commit comments

Comments
 (0)