NicolaThere are limits to the speed or cost of retrieval in the symmetric PoRep.
Limits in the cost model
- A realistic cost ratio (Amax) between the honest setup and the attacker setup is ~10x.
- A realistic time of a challenge window is ~6hrs (recall that the cost model requires challenging every sector)
- Assuming that sealing is as efficient as a regeneration attack
→ Theoretical minimum: The cost of unsealing must be AT LEAST the cost of 60hrs (2 days and a half) worth of storage costs, at today’s storage costs it is $0.06 to retrieve 1 TiB (at 0.025 TiB/day)
→ Current costs For context, SDR cost per GB is (0.04 per sector / 32) is $0.00125/GiB, which would account for 10% of the S3 cost.
Ideal (0 Cost Ratio) | SDR | All other costs, bandwidth + revenue | Cost of retrieval on backblaze | AWS Cost | |
$0.00006/GB | $0.00125/GiB | $0.002/GiB | $0.02/GiB | ||
For comparison
- S3 Archival expidite access costs ~$0.01/GB, so in this setting, the unsealing cost of $0.00006/GB will not have a major impact on costs.
- From DigitalOcean current costs are at $0.0003/GiB
- Backblaze is 0.00217014$/gb
Request:
Limits in the time model
- A realistic latency ratio (Amax) between the honest setup and the attacker setup is ~5x.
- A realistic time for replying to a challenge to a reliable, secure and distributed protocol (e.g. a timestamp service) is ~200ms
- Assuming that sealing is as efficient as a regeneration attack
→ The time to unseal must be AT LEAST 1s (5x200ms)
Due to inefficiencies between sealing and regeneration attacks (today they are in the order of 20-50x!), we must multiply those numbers by another factor - optimistically another 5-10x.
cost of smart: cost + spacegap cost - storage cost
Cost of encoding a single unit of storage * space gap reduction * cost ratio