filecoin-project / Allocator-Governance

7 stars 36 forks source link

3rd Community Review of 1475 EFil+ Allocator #190

Closed 1475Notary closed 4 days ago

1475Notary commented 1 month ago

First Review #57 Second Review #143

Allocator Compliance Report: https://compliance.allocator.tech/report/f03018491/1728695297/report.md

5PiBs DataCap awarded in 3rd round

example 1: 1475Notary/1475-Allocator#14 - 1.75PiB

After Q&A with the client and confirmation of their identity, we allocate datacap to the client. Before the second round of allocation, we reviewed the client's cid report and confirmed that the results were good, so we decided to allocate a new round. By the third round of allocations, we noticed that the retrieval success rate had reduced in their latest cid report. After discussing with the client we decided to allocate a new round to the client to observe their improvement. (We asked the client to add a note on their application. https://github.com/1475Notary/1475-Allocator/issues/14#issuecomment-2408454598)

example 2: 1475Notary/1475-Allocator#12 - 1.5PiB

After Q&A with the client and confirmation of their identity, we allocate datacap to the client. Before the second round of allocation, we saw their improvement in the client's cid report, so we decided to allocate a new round. By the third round of allocations, we noticed that the retrieval success rate had reduced in their latest cid report. After discussing with the client we decided to allocate a new round to the client to observe their improvement. (We asked the client to add a note on their application. https://github.com/1475Notary/1475-Allocator/issues/12#issuecomment-2408465726 The client has stopped working with sps who provided low retrieval success rate.)

example 3: 1475Notary/1475-Allocator#19 - 1.75PiB

After Q&A with the client and confirmation of their identity, we allocate datacap to the client. image Before the second round of allocation, we found that the retrieval success rate is low in cid report. Due to client just added some sps into their team, we decided to allocate new round to observe the next process. https://github.com/1475Notary/1475-Allocator/issues/19#issuecomment-2378904620 Before the third round of allocations, we saw their improvement in the client's cid report, so we decided to allocate a new round. (We asked the client to add a note on their application. https://github.com/1475Notary/1475-Allocator/issues/19#issuecomment-2408420149)

filecoin-watchdog commented 1 month ago

@1475Notary Allocator Application Compliance Report 1st Review 2nd Review

1st Review score: 2.5PiB granted 2nd Review score: 5PiB granted

3.75 PiB granted to existing clients: Existing Client Name DC
NOAA 2 PiB
ChildMind Institute 1.75 PiB
1.75 PiB granted to new clients: New Client Name DC
NationalCenter For Atmospheric Research 1.75 PiB

Example 1 1.75 PiB DC granted — This dataset was already stored several times(FIL-A-2,top value storage,filecoin-plus-large-datasets)

SPs list updated in the issue: f03100008 Guangdong f03100009 HongKong f03030649 Hebei f01660795 Guangdong f01084413 Zhejiang

SPs list used for deals: f03100008 f03100009 f03216485 f03030649 f01660795 f01084413 f03157879 f03161261 f03159626 f03224828 f03189917 f02825282

All 5 SPs updated in the issue consist of SPs list used for deals, yet 7 additional SPs were never mentioned (neither in the form nor updated later). Retrieval looks rather good. 5 SPs has low retrieval rate.

The client declared 10 replicas, but there are 12 already.

Example 2 2 PiB DC granted — This dataset was stored many times. NOAA often appears as a dataset in applications, so you need to pay special attention to applications that use these datasets. Below is a link to search just for this dataset. There are 2 search results. https://github.com/search?q=repo%3Afilecoin-project%2Ffilecoin-plus-large-datasets+noaa-ofs-pds&type=issues With allocator.tech, you can find applications open on 12 other allocators by typing in NOAA's name. Did the allocator ensure that sending more copies of the dataset would be valuable for the network?

The client declared 10 replicas, but there are 11 already.

Example 3 1.75 PiB DC granted — SPs list updated in the issue: f03100008 Shenzhen f01084413 Ningbo f03214920 Hong Kong f03161261 New York f03192503 Hong Kong f03189917 Dulles f03156722 Hong Kong

SPs list used for deals: f03100008 f03100009 f01084941 f03100002 f0114153 f01084413 f03214920 f03161261 f03192503 f03189917 f03156722

The client did not use any of the SPs listed in the form. The client's comment appeared in the thread with an update of the SP list, which included 7 SP IDs (all were used for deals). The list of SPs with which deals were made contains 11 addresses. The client did not update the additional 4 addresses used for deals. Did the allocator know about this change?

The second CID report showed a lot of irregularities, including low retrieval rate, but the allocator commented on it as "good job". Why such a conclusion after seeing a report with a retrieval rate of 8.9% on average, with SPs of unknown location and 2 out of 7 SPs that sealed over 25% of data?

Also, this dataset seems to be already stored many times on filecoin. Has the Allocator talked about this with the client?


In general, the allocator tries to keep an eye on its customers, performs KYC (however, this is a method outside of kyc.allocator.tech, gov team may ask additional questions here). The allocator should pay more attention to the application details and check if datasets are not stored on filecoin already. If this is the case, additional questions should be asked.

1475Notary commented 1 month ago

Hello @filecoin-watchdog

Example 1 We used to remind our clients to keep their sp updated. image The client has updated the new sp twice. https://github.com/1475Notary/1475-Allocator/issues/14#issuecomment-2314979014 https://github.com/1475Notary/1475-Allocator/issues/14#issuecomment-2408454598 We support our clients in distributing data to a number of sps as they described in the application. We noticed that the reason clients store in more sps than the number in their application is that they replaced sps with low retrieval rates. Next we will also push our clients to provide updated lists when necessary.

Example 2 Thanks for the helpful tips! I had a look at this dataset and got the result https://github.com/search?q=repo%3Afilecoin-project%2Ffilecoin-plus-large-datasets+s3%3A%2F%2Ffcp-indi%2F&type= issues Those past applicants that I saw didn't start storing this dataset. Nevertheless, we will follow up with improved selection of data stored by the applicants, and your method can be very helpful in doing so. On the question of replicas, the client said that they had stopped working with sps who did not give a good retrieval success rate. https://github.com/1475Notary/1475-Allocator/issues/12#issuecomment-2408465726 But the sps that have worked with them in the past still show up in the report.

Example 3

The client has disclosed full of sps to me as follows. https://github.com/1475Notary/1475-Allocator/issues/19#issuecomment-2378904620 https://github.com/1475Notary/1475-Allocator/issues/19#issuecomment-2408420149 The location of that sp I looked up via filfox, I don't know why the ip didn't show up in the report.

I saw the client's efforts to improve the retrieval success rate, which they were dealing with positively, and in short time, they increased the retrieval success rate. I hope the client continues to make progress like this.

report in 9.28 image

report in 10.4 image

In my view, it seems that most of the applicants who have been storing this dataset have not completed the whole process. We will, however, step up our review of the various aspects in the follow-up work so as to do a better job.

galen-mcandrew commented 3 weeks ago

Good evidence presented overall, but calling special attention to the following areas for the next round of allocations:

Given the KYC diligence, retrieval rate increases, and other evidence, we are requesting an additional 5PiB of DataCap.

Perhaps this team can help propose some solutions for the community to have more transparency around this issue:

In my view, it seems that most of the applicants who have been storing this dataset have not completed the whole process.