IPFS IPFS þing 2022 - Measuring IPFS, 10 Aug 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Report Back & Summary - @yiannisbot - Measuring IPFS

Description

----------------------------------------
Missed last year's IPFS þhing? Register to attend IPFS þhing 2023 in Brussels, Belgium here: https://2023.ipfs-thing.io/

A

B

C

Is the list of.

A

Topics that we have uh so far, uh one is what and how to uh to measure in ipns. um There are some magic numbers in there as well. uh What and how to measure hydro appears performance.

A

Another one is content replication in the network. So basically um we have say these. We have files. How many times does each one of them exist in the ipfs network? Is in only one? Is there any replication and so on? You know.

B

How decentralized.

A

This content itself, basically.

C

Does it leave locality.

A

Yeah, it could be content locality as well.

C

D

Yeah, I think both of those are interesting from an outsider's.

B

Perspective yeah.

A

B

A

Timeout uh yeah, I guess that means. How long do we wait for bits of discovery, yeah um yeah? Let's pick a value.

D

C

D

A value for anything, though I want.

A

D

I've heard this so many times in protocol labs like since I've been here, and I hate it. We should not be hard-coding any numbers if you're getting.

A

D

Yeah, okay, good! Stop! It do not pick any number. If you pick a number you're going to be wrong at some point, don't pick.

B

A

Create a function that picks.

D

A number yeah and then optimize the function exactly.

A

Okay, that's what we've been discussing.

C

A

A

Yeah, I think the function there is quite simple from the point of view of you know you have between it's asking for cids and depending on the number of successful answers, it sets the bits what time out so.

B

A

Nine out of ten bit swap requests are successful, then you're very much likely to find stuff through bitswap, because.

C

Your period gateway.

A

Or cluster node and so on, so you can delay starting the dht routing lookup process, because this will add extra resource uh requirements and traffic to the network and so on. If one out of one million bits of requests come comes back positive, then just set it to zero and I don't call yeah um if success heuristic.

B

Then yeah then yeah increase or lower the number as appropriate, and then your success heuristic will change exactly there. You go it's yeah evolving yeah. We don't need a full model, but.

A

B

Exactly full machinery model, we can just do it, for instance,.

A

um That would be interesting- I don't know from gabe to here from gateway operators, maybe the pinata guys or something see if they want some different setting. That would be should be taken into account.

C

B

A

Is making ipfs magic numbers dynamic? I think the first step to that is to link them. Sorry,.

B

A

To list them, um you know which ones are these that just have a value and then go and tackle them one by one. I think the k value is first good step.

D

Yeah, you have a solution for that already.

B

Like you, I think you've talked about it quite a few times right for and you have both of them. Yeah.

A

Presented, um I heard.

B

You're, like at paris p2p, but you talked about it this time, yeah.

D

B

I mean from what I've heard that sounds like it would greatly improve things.

A

Yeah um yeah another question that I posted. There is not necessarily something that requires a whole session, but how can we figure out why the dht walk cdf that I presented earlier today, you know is different between different regions.

B

A

Need to dive a little bit deeper into that. I don't know just to find an answer, because.

C

A

Whoever sees that cdf with you know three kind of distinct areas is the first question: why does it look like that.

A

Now that I think about it, it might be useful to do that before the 20th of august.

A

Because of the conference and yeah.

C

Yeah, it's gonna be yeah, yeah yeah. Definitely.

A

Yeah and then on natural punching, I also it was made up that. um What other metrics do, you think, will be needed um if any or if you think we can have a complete picture with what you have already we're.

B

Going to have a number.

A

Like the one we show like 79, 21 or whatever yeah, but you know, do we want to dive deeper, like I don't know accuracy of rtt measurements that we discussed or not type yeah yeah or someone looking at it from a different angle or split the data.

C

By some other property that those peers have yeah could be a good session. I would like to add the uh test ground discussion pad like a few minutes ago. Yeah, okay,.

A

C

I'm not sure if this actually fits into like this network measurement and yeah and like the measurements that we have so far, but this would be rather into this iqs observatory thing right: the test ground setup to be to test the new release before it is released.

A

And um I find that's quite interesting, but I'm not maybe that's up for discussion if this actually fits this. This scope we have here right now.

A

Do you mean the resource profiling.

C

C

Just having a proper testing set up, yeah releases is that so like are you talking about like testing changes.

C

C

B

What I'm, after.

C

And so that's why I'm not sure if this actually fits this thing here, but um I think setting up like a simulation of like the reputation of the actual.

C

A

We might want to wait until test ground is actually.

A

B

Okay with the same motivation here.

C

A

Nice, okay, so do we want to? I don't vote? uh I guess we're not too many. We can. I.

B

Don't know we can.

A

Split or we can be in one room and go through these- I don't know one by one and whenever we think we finish something. Oh okay, I don't know I'll get 15 minutes for the next two hours. So it's three we're going to come back at 3, 20 or something so we've got roughly say one and a half two hours do we want to spend. I don't.

B

A

10 15 minutes on each one.

C

A

Yeah there are, we can, we can go and check those as well. I think in the list of priorities they would not come at the top, um but we can scroll through as well.

C

A

uh I can double check, but I don't think so all right um yeah. What do people.

C

Think, let's check what.

A

Do people think how should we do it um be in one room all together here and allocate? I don't know, prioritize those and allocate 10 minutes or 10 minutes each or start with the most important.

A

Whatever time it takes until we maybe.

C

We can just start by hoping, and so if some of the topics, okay,.

A

Do you want to see what's left okay, so we bought by hand in the physical world.

A

Okay uh in no particular order: okay, let's start from the beginning.

B

A

uh I guess everyone would have um two three three votes.

A

Yeah: okay, um who wants to talk about the first one ipns.

A

Okay, um hydro ps performance: what and how to measure there to figure out stuff.

A

Okay, uh you write it down content replication in the network, um so that was what it would say. It is so, first of all how many copies of everyone some particular cid, exist in the network and do like uh sorry and.

B

A

At which geographic.

C

Region, I think that this one goes really in line with how many content, how many storages right.

A

With that you mean, if.

C

You have access to how much content.

A

Yeah, that could be another topic of discussion. This rfm that talks about how much content is actually stored on ipfs being actually a good topic to discuss um cool anyway content replication in the network yeah. Who would like to see talk about that.

A

B

A

I think I already done my three goals anyway, a bit of time out.

A

Okay, making ipfs magic numbers dynamic. I would vote for that as well. If I have another role.

A

A

Test ground and resource profiling, and things like that.

A

Okay, um how can we figure out why the dhd walk cdf is so different between different geographies geographic regions? I would love to yeah.

C

Anyway, maybe we get to this anyways.

A

Yeah yeah exactly- and I don't think he's yeah deserves a whole session anyway and then not whole punching. If we need anything more okay, yeah, I would do as well. Okay, so pretty clearly.

B

Magic numbers ipfs.

A

Magic numbers: okay,.

B

We'll start by listing them.

A

I guess and then saying yeah what needs to be done and then we go to bitswap timer or do we want to split or do we want to sit here all together perhaps bring a whiteboard or something I'm.

C

Able to sit here, yeah like being all together, yeah and maybe some.

A

Cool: let's do that.