3rd Community Review of Greaterheat Allocator

[AlanGreaterheat]

Latest report:
https://compliance.allocator.tech/report/f03010601/1729817128/report.md

Since the allocator has been running, there have been a total of 5 client application requests, including 4 corporate public data applications and 1 public dataset. Overall, the SP Spark retrieval success rate has been good. However, due to the recent high gas fees and the official team promoting the DDO model, some SPs have a Spark retrieval success rate of 0, although most have provided DDO order mode vouchers.

Moving forward, we encourage more data types to join the Greaterheat allocator, and we are happy to assist and provide feedback to help the community advance the compatibility of DDO and Spark retrieval.

If you have any questions, please feel free to comment. Thank you!

[filecoin-watchdog]

@AlanGreaterheat
Allocator Application
Compliance Report
1st Review
2nd Review

1st Review score: 10PiB granted
2nd Review score: 10PiB granted

10 PiB granted to clients:

Client Name	DC
YoulePte Ltd	4PiB
taikr	2PiB
Frost& Sullivan (Singapore) Pte Ltd	2PiB
SpaceTelescope Science Institute	2PiB

Example 1 - SpaceTelescope
Client updated SP list before last allocation, adding 7 new SPs. None of them support spark.

Updated list (none of the SPs support spark):
"f03173124 Aomeige South Korea",
"f03188443 YPH SG",
"f03188440 Fei Canada",
"f03196401 XingYa Japan",
"f03190614 Laibin Dubai",
"f03190616 YM Japan",
"f03196399 Liulian Malaysia",

The client declared 5 replicas, but there are already 7 of them.

Example 2 - Frost&Sullivan
The client requested 10 PiB and declared creating 10 replicas of the 600TiB dataset. Only 6PiB would be enough, why did the client ask for 10PiB?

Most SPs match the provided SP list (only 2 were not used for deals). 5 out of 14 SPs do not support spark, and the rest have good retrieval rates.
This comment is unclear to me. On what do you base your limited trust for this client?

Example 3 - taikr
The client requested 15 PiB and declared creating 10 replicas of the 880TiB dataset. Only 9PiB would be enough, why did the client ask for 15PiB?

Those SPs don’t match the provided nor updated SPs list:
"f03144037",
"f03156617",
"f03148356",

3 SPs don’t support spark retrieval, 1 SP has low retrieval rate, and the rest(6 SPs) have good retrieval.

Example 4 - YoulePte
The client requested 10 PiB and declared creating 10 replicas of the 580TiB dataset. Only 6PiB would be enough, why did the client ask for 10PiB?

Based on a report from 4 days ago (couldn't generate a newer one), 7 out of 19 SPs don't support spark, 1 has low retrieval and the rest have good retrieval.

In general, the allocator talks to clients, does random SP checks, which is an interesting solution. Does this mean that not all SPs are checked by the allocator? I would like to hear more about this approach.

Additionally, CID reports are run and checked. The allocator's clients maintain retrieval at a fairly high level. Except for SPs that do not support spark.

[AlanGreaterheat]

Thank you very much to @filecoin-watchdog for the thorough review. The feedback provided will guide our future improvements. Regarding the issues raised, our responses are as follows:

1. Regarding some nodes not supporting Spark retrieval rates.
As Greaterheat, a SP deeply engaged in the Filecoin industry for over four years, we have been closely following the development direction of the Filecoin governance team. The official technical roadmap has been promoting the DDO model, so we also recommend our partner SPs to adopt the DDO model.
Onboarding Data Using DDO Deals
FIP-0076
Unfortunately, the Spark retrieval method is not compatible with the DDO model. For more details, please see:
Compatibility Issue

Our team has also consulted with the Spark team regarding compatibility issues, and they have responded that they will complete the compatibility between Spark and DDO by the end of November.

Meanwhile, we have requested the client to provide DDO order vouchers. I will randomly share a few screenshots:

[AlanGreaterheat]

2. Regarding the conversion of datasets and Datacap.
This topic has been thoroughly discussed by our partner, DSPA-Asia's allocator.
Discussion Link

As a leading SP within DSPA-Asia, many of our operational rules have benefited from their training and guidance. At that time, we considered the conversion ratio between real data and CAR files. Although the current method may not be the best solution, it is at least a suboptimal choice and represents a pragmatic strategy to encourage a wider variety of data types to join the Filecoin network.

Of course, for future applications, we will strictly adhere to the official 1:1 ratio.

[AlanGreaterheat]

Meanwhile, in comparison to the second review comments from @galen-mcandrew, we have made corresponding improvements:

1. The average retrieval success rate of Spark has increased from 28% in the second review to the current 55.31%. Excluding the DDO mode SPs, the retrieval success rate for the other SPs is very high.

2. Conduct random sampling of certain SPs for location proof.

3. Conduct more rigorous KYC procedures by requesting business license information to verify the authenticity of the company entity.

This also addresses the content mentioned by @filecoin-watchdog regarding Example 2 - Frost & Sullivan. Frost has provided a business license and location proofs for some nodes, so we believe continued support is warranted.

[AlanGreaterheat]

Continued Update:

We have communicated with the cooperating clients and SPs, in order to ensure the success rate of Spark retrieval, we will stop cooperating with SPs adopting the DDO order mode until the Spark team DDO mode for compatibility.

Looking forward to the subsequent SPs' Spark retrieval success rate improvement, of course, if the cooperative SPs have any technical confusion about Spark retrieval, feel free to further communicate with Greaterheat's technical team.