Filecoin Filecoin Space Race, 1 Sep 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Filecoin Mining Workshop - Filecoin Mining: A Deep Dive

Description

Miners assemble! Join @Magik6k for the second Space Race workshop - Filecoin Mining: A Deep Dive. The most comprehensive overview of mining to date!

Join Space Race;
https://spacerace.filecoin.io/

And learn more about filecoin at:
filecoin.io

A

Hi everyone, uh a very warm welcome, uh come on in take a seat, maybe you're in a seat already, that's totally fine too. Maybe you've got a standing desk. Maybe that's your vibe. That's great! Also um we're just gonna give it a minute or so to let other people join um in the meantime, uh please let us know where you're joining from. um I am coming to you today from glasgow in scotland um and yeah. I know our team are across the world as well.

A

uh We've got someone from dc, I'm from norcal, which I think is. Northern california, I'm very jealous the weather. Here is definitely not as nice uh derek from dallas hello at benjamin from denmark, you're in my part of the world, a warm a warm welcome. We've got austin texas.

A

I love it everywhere.

A

Okay, all right! Let's, uh let's get this on our way, so a very warm welcome to the second of our foulcoin mining workshops. uh Today we are doing falcon mining and deep dive. um We it's a really exciting time uh in space race, we've got 345 miners across the world uh who have onboarded 76 pebbites the network. Just it's just incredible like we are so excited.

A

We hope you're, super excited um and- and the purpose of today is to like really get into the kind of mechanics of file coin mining um and hear from one of our very best. uh So it gives me immense pleasure to welcome falcone engineer magic um who is going to lead this workshop over to you magic.

B

Hello, everyone uh yeah. I want to really start with explaining some car concepts behind falcon and mining uh yeah we'll go through some of them. So everyone understands what I am talking about: uh yeah we start with power uh and falca and minor power, just basically like you could have winning a block is based on the amount of proven storage, and that is slightly different from other blockchains, where power is your hash rate or state fans.

B

Another thing that makes it slightly different from other blockchains is that you kind of accumulate power over time instead of using it linearly, like heart rate,.

B

And one of the coolest features of falcon is that an unfucking power is can be actually useful, so you can use the storage to store your data instead of heating up earth or not using your money uh yeah.

B

The next thing is the concept of a sector and that's the main unit of storage in falcon. It's just a fixed, fixed size, piece of data.

B

Yeah, it's each sector can contain multiple deals and a sector without pills is called committed capacity.

B

uh All proofs and other persons like that are run on sectors uh yeah moving forward.

B

We go to storage, proofs and storage groups are the magic sauce of falchine.

B

There are two core groups which is one of which is proof of replication, which proves that a unique copy of data is was created and is, and the other proof is the improved spacetime, which proves that that unique copy of data that we have just created is stored over time and the current construction called stack, drg port.

B

Which will go into more detail in a bit? uh It supports 32 and 64 gigabyte sector sizes just also relevant a bit later. Next, we go to actors which are a bit like ethereum, smart contracts,.

B

Every actor has a balanced, nonscope state, an address uh the key addresses, so the one prefixed with t1 or d3 uh are.

C

B

An actor underneath it so every key address also has a has an actor address.

B

uh Yeah the difference in mainnet is going to be that all addresses are going to start with the letter f instead of t, so that's how we will distinguish between real addresses and the less real ones or well. This doesn't interest us uh yeah. Next, we go a bit more into specific actors. First, one is the power actor which is responsible for just keeping the network power table, and that is mostly just interacting with minor actors you, as miners, usually don't interact with it directly.

B

The only case is when you are creating new minor actors which you go through, so many layers that you probably don't even notice that year that you do next is the minor actor, which is how you talk to the chain as a miner.

B

It handles all storage. Events verifies your proof structure tracks your sectors over time. uh Just ties your mining compression to the rest of the network.

B

You will be sending a lot of messages to it, so we voted up uh yeah. It can be controlled from multiple keys, but on the topic of keys, uh yeah supports two key types: uh 6p and vls, and as miners you have to use a pls key as the main or worker key that you use to create blocks. This is because bls keys have the interesting property that there is only one valid signature for one input message and we use that to derive some of the randomness into network.

B

uh It also has a very nice feature that its signatures can be aggregated, so that saves a lot of space in china which we need uh next yeah mining mining is really free activities.

B

uh That's just ceiling or running starts uh just how you gain power, and then you need to prove that you store those sectors over time, which is how you keep your power, and that is the actual block mining, which is how you get the network to move forward and also get block rewards, which makes it all worth it.

B

B

Look at the like higher level architecture, uh no questions so far by the way, if anyone has any questions, feel free to ask them. uh So we go because there is a lot of context. Switching here, uh yeah, uh fundamentally, we have three note tapes uh as miners you will likely be using all of them uh on the top. You have the demon or what we call the blockchain node. This is what this talking to the network.

B

The lotus miner is what is handling all mining related tasks just connects to connects these stores to the network and that's basically everything you're interested in and then you have lotus workers which are an extension of the main lotus miner process, um just yeah. It allows you to add more resources to it. It can be run on. Multiple machines allows you to build much larger mining clusters and we'll focus on the lattice miner, mostly for this stack, because it's well where all the default.

B

Happens: yeah, it has four main components uh all perform. Some mining related activities mentioned before the block mining a window post scheduler is the thing that handles proving sectors.

B

Over time you have sector storage, which is the component which has sectors on disk and also runs most of the ceiling operations. And then you have the storage of sam, which is what.

B

Handles basically all interactions that need to happen between the blockchain and storage and market to get sectors sealed so yeah, storage of sam or finance machine.

B

uh This is what is responsible for getting data from deals, sibling sectors, seeing them getting them in chain.

B

There is a bunch of steps involved in seeing a sector. Some of them are very competitionally intensive.

B

Yeah, creating a sector.

B

Starts in concert in twice, you can either play the sector or create a committed capacity sector, which means a sector without deals, or you can start a sector.

B

By accepting some client deals and not sending them into a sector, when you accept a clan deal, you will create the sector put the deal into that sector and then keep it open for by default one hour to hopefully put more deals into that sector.

B

uh So you have to do less work to accept more deals so and hopefully get paid more for less work. Just probably what you want to do, uh yeah, also, if nothing capacity sectors can be upgraded into sectors with deals.

B

uh Upgrading a sector means that the initial pledge doesn't have to be paid again or is reduced depending on the network conditions. It usually won't have to be paid yeah. Next, from those two paths we go into the common sync path.

B

We get into the packing state where we just fill the sector up. So it's the full sector size worth of bytes yeah, it's basically just filling it with with no bites.

B

uh Then we go get to the first very fun step where we do the replication, which, with sdr, is a single threaded process. It requires the very minimum 64 gigs of ram.

B

You probably want more and then some swap memory uh using more memory just makes it a bit faster. uh The process is inherently single threaded, so you cannot accelerate it by trying markers at it or a gpu. It doesn't help at all.

C

B

The process creates 11 layer files, which reserve results and just shy of 400 gigabytes of files. In total that you need to keep on fast storage for a bit for the older steps.

B

And you are able to discard most of that after this actor is fully sealed. uh The whole process takes about four hours in a zen two and these empty based cpus and we're getting some optimizations soon. Hopefully, so it should get somewhat faster, we'll see how much uh then from pre-commit one we go into pre-committee where, where they happy and credited sectors now we have to build a few special numerical trees over the data. uh This step can be run on gps society parallel.

B

uh It writes another three times: detector size of data worth worth: data on disk, uh which is about 100 gigabytes for a 32gb sector. It doesn't care, it doesn't require too much memory and it can be run on gpu. It takes about 30 minutes on a 2080 ti, so you usually want to be a bottleneck in most cases uh after we've computed the pre-comment.

B

Of the pre-commit stuff, we get a few snippets of data that we send to the main actor where we tell where we tell our main director that we have replicated a sector and that we want to prove that it has been replicated after that message on sunshine. We.

B

We wait a bit for.

B

We write a bit for a randomness random challenge.

B

Which we then use the comment step where first in the commit one step, which is that just reading bunch of data for to answer the challenge we get from the chain we read about somewhere between.

C

B

And 20 megabytes this order of magnitude of data. That is our proof.

B

Sending 10 to 20 megabytes of data to a blockchain doesn't work at all. So we use the magic of snarks to compress this 10 or 20-ish megabytes of data down to 192 bytes, which is a very, very large compression factor.

B

But to do that, we need to compute a very, very big snark, which requires a lot of memory.

B

Having a gpu is handy and it still takes a bunch of time, so it will use peak of around 170 gigabytes of memory, although for a very short time, so you can easily do that with 120 gigabytes of ram and then a bit of swap uh probably should use less. We are not sure why this is so much.

B

uh It may get. Optimized we'll see. uh Yeah the the whole process takes about an hour with a 2080 di for a 32 gig sector.

B

After we do that, we submit the comment message or the port proof to our my director. We wait for it to land and change.

B

The minor actor then verifies that it's correct tells if it is correct, it tells the power actor to grant us some power for the sector, and it also tells the main actor that deals in that sector are now being proven.

B

uh Which means we get to have power, maybe win black rewards, and we also get paid for the deals that we've put in that sector.

B

C

B

Have questions.

C

B

There will be more clarifications and some conflict.

C

B

We'll look at the questions later. We still have a lot to go over uh yeah the.

B

Second, probably most important component is.

C

B

Sector storage.

B

It only really has two jobs uh just to keep data around and when computation sign up. Since I get relatively simple tasks, uh so this is the architecture diagram for, for for it uh it's more tricky than it sounds so uh yeah, uh let's, let's unpack.

B

First, we first we have the storage pass, which we address, storage as pass, which is as simple as it sounds.

C

B

Path is just a directory with.

C

A bit of metadata.

B

Both the main miner, node and worker nodes can have multiple paths for workers that supported since very recently, I think only in 0.57 uh those are divided into two types.

C

B

Storage paths for long-term sector storage and ceiling, or squarespace mentioned before, seeing pass for keeping temporary data for ceiling operations.

B

You can also combine and have a path. That's both long-term storage and ceiling.

B

But usually for for seeing storage you want fast ssd drives and those are very expensive so and sitting on a hard drive is going to be much much slower.

B

So you will probably want to have separate paths for for those two to start types. uh Then we go into remove bus, so storage can be local or remote. Each node can have a different set of local paths and when I need note, I mean either a miner or a worker.

B

The lotus miner process is as a built-in as a built-in worker. So when I say note, I will usually mean a minor or a worker. Note.

B

Yeah, so going back, uh sectors need to be moved between nodes, and this is when remote paths come in handy each local path and by local I mean above that and node you can access locally. So it's just a local directory.

B

Each local path in the system is exposed by its node through http, uh and this allows nodes to transfer sectors between themselves.

B

uh It's also worth noting that us can be shared between nodes using short file systems such as nfs ceph or any other technology. That does that, and this makes it so that transfer don't have to happen over http.

B

Which can be much like this can be much faster with technologies like rdma, which can be used through with nfs decently easily, but yeah. That's that's the more exotic stuff uh yeah. Next we get the worker.

B

This is the component which gets sectors into local storage and actually runs tasks coming from higher level levels. So the storage of some.

B

Each worker will access its its local storage and it can also pull things from remote storage fast into its focal storage bus uh yeah again, the doctor's minor process embeds uh worker. So technically it's not necessary to run the lotus worker process. If you only have one machine, there are still some cases where that can be a good idea, but usually you won't have to with just one machine and now the fun part get to scheduling.

B

uh This is the logic that decides what tests run when and where it's very complicated and still didn't receive nearly enough love from us.

B

uh It's really hard because it needs to balance a lot of things we we want to support many. We have to support a diverse range of setups at many different scales, and we kind of we kind of just require everyone to build their data centers in the same way with the same parts.

B

It doesn't work so so that we have to support everyone. uh The scheduling strategy it has a few parts.

B

Initially all tasks will go into the global scheduling queue from there. The main scheduler assigns them to so-called per worker windows.

B

When, after they are assigned to a worker, the worker will will process all of those.

C

B

The hard part is nine which task to assign to which worker. And when.

B

Is much more complicated than it sounds uh yeah, so we use the concept of resource windows mostly to prevent resource starvation. So when you have a lot of small tasks and a few big tasks, you could be scheduling this stream of small tasks and that would prevent the bigger task from ever running.

B

We define a window to be a group of tasks which worker which a worker can run all in parallel. So if you schedule a window worth of tasks, the worker won't ever run out of resources.

B

Each worker has a few windows, so windows are tied to one worker, so all windows are worker specific, and if workers have different sets of resources, their windows will have effectively different sizes.

B

uh This site is a lot, so each worker has four windows.

B

Two of those are for running and preparing for running tests.

B

The global scheduler operates in a set of open windows, and now on open window is just the task is sorry, and an open window is just a window with no tasks when at least one task is assigned to an open window.

B

That's window is no longer open and it's sent to the to its worker for processing uh worker. Then process assigned windows sequentially, which is pretty important.

B

And they do that by moving tasks from those assigned windows to the preparing window and then start preparing for an intellectual task, so preparing usually means just fetching the sector from other workers, just making sure it's locally available uh yeah when the assigned window from which we pull task into the parameter when the assigned window get this assembled. Yet we send it back to the scheduler as a new open window to get a smart task when tasks in the preparing window are ready for processing. So we fetch detector or it's already here.

B

Okay, uh the task will get moved to the running window where it is actually executed. So we actually. This is where we actually run the bigger test, like pre-commit one, two and everything else, using lots of resources for a decently long.

C

B

B

Yeah, when we assign tests to windows, we guess how much resources they use based on the static resource table, which is in the go file. The bottom of this sled can be tweaked very easy to tune.

B

Some of the scheduling logic, if you know what you are doing, if you don't know what you're doing you can end up with the scheduler using either way to view resources or just crash and call your workers with way too much tasks.

B

It doesn't care if you set it to low at all, just send everything to you and you have to somehow survive that or not uh okay. So next there will be some improvements coming soon. Ish the current project isn't perfect, we'll be working on a number of things, make it somewhat better uh yeah. We are also very, very open for any ideas, so just send her our way, if you think you know how to make this better without too much effort.

B

We want to hear about it.

B

Okay: next, we have proving storage uh the next decently big thing we have to do as miners.

B

Searches prevent through window post or we do proof of storage proof space time uh all sectors are proven every 24 hours uh unless they are fully or expired, in which case we probably can't prove them. Anyways.

B

uh Split the 24 hour, so-called proving period into 48 deadlines. Each deadline is 30 minutes that probably came from dividing 24 hours by 48.

B

Yeah, then, in each deadline you can have one well, you can have zero or more partitions worth of sectors.

B

C

B

Just an arbitrary group of sectors which you prove together in a single window, post message: unless you have more than I think 10 000 sectors in a deadline, then you specif more, but to get more than 10 000 and one that, then you have to be a pretty big miner, so guarantee it's unlikely to be an issue for anyone.

B

uh Yeah and company window post also involves computing, a snark that can also use a gpu uh to make it faster.

C

B

B

Each deadline defines a couple of important points in time. First, you have default cutoff, which is 35 minutes before the line opens, which is the latest time at which you can declare faults on recoveries.

B

Then you have the challenge bug, which is when the challenge randomness for the proof of space time becomes available. This is 10 minutes before the deadline opens.

B

Then you enter the deadline, the deadline, open ebook and from from that time, you have 30 minutes to submit the window post message to detain uh if you, after 30 minutes, miss the deadline, close ebook.

B

You can you can't submit the post message for that deadline at that point and when that happens, this is nicely defaults. uh You have two main.

B

Sector related fault types, uh you have declared faults and skipped faults, uh skipped post faults, declared faults, may pay a smaller penalty compared to skipped post falls, so you want to declare that you are not able to prove your storage in advance.

B

uh The difference is pretty big from what I remember uh sectors that are faulty uh through either. Full tape can be faulty for 14 days and after 14 days they will be permanently terminated uh unless they are recovered from the recovery mechanism.

B

B

Yeah, like faults, you have to declare recoveries uh 35 minutes or before 35 minutes before the deadline opens at the latest.

B

And when you declare recovery, you claim that a sector that you declare as recovered will be proven in the next.

B

Deadline in which the sector has and after you declare that recovery and you do submit the window post for that sector, uh you will just recover the power for for it.

B

uh You want recover the fee. You had to pay for full think on that sector. So don't just default. It's expensive, uh yeah, the next, pretty major thing that you do as a minor. Yes, my books. uh This is where you actually get your blog rewards and get paid for everything you do.

B

C

B

Mining process is just a simple series of steps all done automatically.

B

So it starts by selecting blocks or tab sets which you'll mine. On top of, then you check if we won. If we didn't, then we go back to the first step and just wait for the next ebook.

B

If we did win, we start completing another type of proof called the winning post uh winning post. This just approves space time uh on a single random sector and.

B

The thing it does us to do is have a minor lies about some percentage of its storage so, like they have removed, say 50 of their sectors.

B

But still weren't caught by the windowed post with winning post, they will fail fifty percent of winning posts so effectively their power will be already reduced by fifty percent.

B

uh So this way you lying about your storage, the only thing it does is it slows down the network somewhat and to slow down the network. Significantly, you have to remove a huge portion of network power was a costly, very, very significant amount of funds. Just because seeing sectors is very expensive.

B

uh Yeah, after completing winning post, we select messages based on gas premiums and some other factors, such as ticket quality, in a way which, up to my discuss, reward and that's much more complicated than another blockchain, because in other blockchains you generally have one block at one height and falcon.

B

We have multiple blocks at one height and only the message so yeah everybody can include a set of messages and only debug with the lowers ticket to include a given message: we'll get its gas reward, which makes optimal message selection, a much more complicated topic yeah we are, we already have a pretty optimal algorithm implemented there, so it should be doing pretty good job.

B

uh It can't be perfect, though, because it turns out to be an np complete problem, just interesting uh yeah, after putting messages into a block.

B

We wait for the epoch timestamp and we send the block to the network and we have to wait for the april timestamp, because if we send debug too early notes which receive it too early, we'll just reject it, which yeah the mining timing.

B

uh Basically, after each ebook starts, we wait about six seconds for blocks from that e-book to our from the network.

B

uh This is the select mining base step from before. After the blocks have arrived, we check if we won.

B

If we did win, we compute the winning posts, we put a snark which has to be done quickly also for winning post. You really want a gpu because it has to be done quickly like it should be taking 10, maybe 15 seconds. If it takes longer, you need to optimize something or get better hardware yeah, so you compute this arc. Hopefully in less than 15 seconds, it can take a bit longer, but it gets pretty tight.

B

After that, you put a block together with the messages and everything, and you wait for the ebook timestamp for deeper to start, and then you release it to the network.

B

Yeah, so that's it for mining. Now we can enter the time yeah. We can either go to note setup or answer some questions, uh so maybe we'll do and maybe we'll go quickly for denoted, because it may have some interesting things: uh yeah creating a minor note.

B

uh You create a vls address, you send it some funds from a faucet or in my net through other names.

B

You then initialize lotus miner, with the lotus miner in its command specify some flags.

B

You usually want to set the sector size for my net. It will be either 32 or 64 gigabytes. uh Also something I mentioned that uh 64 gigabyte sectors. The only difference is that they use basically twice the resources for most sitting steps, but the benefit is that you send two times less messages to the chain and that can save you a lot of cost and maintenance on gas fees, and we don't know how the how they will look so, maybe maybe worth supporting 64 gig sectors.

B

Yeah, then you have the worker and owner flags, which I would specify the worker key, which is the d3a bls address. It has to be the pls address and an owner key which can be either pos or 6p. It doesn't really matter.

B

The difference between those two is that the worker is the address which.

B

Is used to create blocks the owner address.

B

In addition to not being able to make blocks, it can do all minor react, attacker actions like the worker key and it can also withdraw funds from the mine actor. The worker address cannot do that.

B

Yeah, there are also some other flags on the internet command like no local storage, which just makes the miner not have any local storage which currently, you will have to just have under storage stuff manually, because the lotus miner process has to have at least one local storage path. Unfortunately, this should be changing soon and you will be able to run and answer post challenges from workers, which will be very neat because you will will not have to have any storage in your local lotus.

B

Miner note uh yeah the next step after you run the initialize command. You will have to fetch some hundreds of gigabytes of snark graph parameter files.

B

There is fortunately no way around. Currently you just have to fetch them. Somehow, though, you are mining falcon, you probably have enough internet and other resources to do it.

B

So next, after you initialize, you probably want to look at your config.

B

It has some interesting fields to get more soon, but right now you will be most interested in the max ceiling sectors and maxing sectors for deals convex, which is the upper bound and how many sectors you can be seeing at any given time, an okay way to calculate what you want to set this to is to add up these size of your ceiling directories.

B

So you have three workers with two terabytes and vmes.

B

uh You will have six terabytes, and then you divide that number by 500 gigabytes, so you get 12.

B

With that you put that number for the maxing sector for deals, uh and then you set the max ceiling sectors for something on the tower like six or eight, because you don't like you, want to have some space for sectors with deals, always uh especially in space race, where.

B

Ceiling deals is one of the requirements uh then magnet yeah also gives you yeah like rejecting people's deals. It's not what you want to do, so you set the for deals. Number higher next. Is the weight deals delay it's the time.

B

It's how long your sectors will stay open, waiting for deals.

B

It's kind of related to the point below the expected seal duration. It's just an estimate on how long you will take to seal a sector and when you accept deals, each deal has a stark epoch.

B

When you put a sector, when you put a deal in a sector, you have to seal that sector before the start. Epoch you've always started book of any deal in that sector.

B

So this means that if you set the weight deals delay for a value, that's too high, you will have less time to actually seal the sector.

B

So yeah there is there's some balance between the weight, use, delay and expected seal duration conflicts.

B

There is the storage section uh which is just used to configure the built-in worker, um and then you have some fee config which, just let's you configure how much you want to pay for which kind of message, but it's if, if gas fees get very high, it might might not be worth putting pre-comments and comments in china so with that they will just.

C

B

Instead of getting pushed to the chain uh yeah next, we.

B

Using miners, if you want to accept deals from clients, you really want to configure public addresses.

B

There are decent docs on how to do it, so if you haven't already, you should do that. It may even increase your deal. Success rate.

B

uh Yeah, the other thing that you may want to configure additional control addresses.

B

Which is related to.

B

Message cast fees, so you, as a miner, you have to submit window posts and missing a window post can be very expensive in some cases.

B

So if the network fees get very high, you still have to post, but you don't want to you: don't necessarily want to pay high fees for pre-comment and commit messages, and the problem is that if you are sending both of those message, types from one key, the pre-comment and comment messages with low fees sets on them can block. Just because you need to have intubate messages on chain with continuous answers.

B

The pre-comment and commit messages can block your post messages or if there are too many pre-commenting comment messages you will have to set a very, very high gas fee on the post message just to push those lower fee messages group.

B

Instead of that, you can just use a separate control address for your post messages, so it's as simple as creating a new address, sending it some funds and then calling lotus miner actor control set and that address, and that will make your miner use a separate address to some posts from which may save you.

B

For a bit of funds and cassies uh next, attaching storage, bus, yeah, every miner and worker kind of multiple stretch paths they can be attached without this, without restarting anything with just the storage attach command, we also specify weight which makes it more likely to be used uh yeah. This is just pretty simple configuration. We don't have a command to detach a storage buff, yet we plan to implement it, but I don't know if that's going to happen soon or not uh yeah everything else.

B

There are a lot of topics that I didn't go much into, that are still manipulated markets and data transfers could probably get another hour of talking yeah. There are topics that have to back up and out uh no redundancy which we don't really support, yet that.

B

We should go into uh and then there's method, selection and all the gas rated fun times, which could probably also use like another hour or two of talking about uh yeah. Instead of that, I can get questions from people.

A

Awesome, thank you very, very much magic. That was incredible and I think the most comprehensive kind of deep dive that we've done into falcoid mining so far. So thank you so much for sharing your expertise. um You do have questions. I think you've answered some of these um throughout your presentation, but if you're happy to let's just run through them anyway um and let's start with meatballs question meatball is the name I know you're one of the community champions on space race uh last week. So congratulations for that.

A

Meatball's question is related to the lotus miner set up. Can we get more clarification on what the expected sale, duration, max weight deal, sectors max, sealing sectors, max sealing sectors for deals and weight deals? Delay are in the lotus miner config.tomol file and what would be some suggested settings for them in single minor and multi-worker setups.

C

B

uh Angie, if you were thinking anything, I probably didn't hear you, because my idea was broken.

A

No worries um okay, so we had a question in the chat from meatball, um which was this. Is this is a hard one to say out loud? um Can we get more clarification of what the expected sale, duration, max weight deal, sectors max sealing sectors, max sailing sectors for deals and weight deals? Delay are in the lotus miner config.tommel file and what would be some suggested settings for them in single miner and multi-worker setups.

B

Yeah, so that's basically what I've gone into here, mmd config uh yeah, just very quickly again! That's how much! How long do you expect to see incorporation to take? You should set it to something 24 hours for a small miner, uh mike's radio sectors, maxine sectors just divide your squad, storage by about 500 gigabytes, each.

C

A

Cool. Thank you very much. Magic and next question comes from danny wu, any way to reload the configuration file of lotus miner without restarting and losing deals.

B

uh Some parts of the config can be changed dynamically for the api. Some cannot, uh we will be very likely, will be swapping the whole config sub system for something that's property based, and that allows reloading every configuration like element without having to restart your notes, because it's very bad. We also try to implement something which allows you to restart or upgrade your notes without losing progress on.

B

Ongoing operations.

B

I don't really know when that's going to happen. Hopefully before I mean that we'll see.

A

Cool awesome thanks magic. Next question comes from igor s: does a minor itself have an internal worker in the process.

A

There you go super easy, it does okay, another question from danny woo: does the deal bot look at expected seal duration at all, or is it client only.

B

I am not sure if it's exposed in the storage desk my experience from experimenting with it and finding a lot of deals is that it does not uh do from what I know the like: dual style. Duration was recently increased, so the deals you're getting from the bot, not sure. What's the current number yes ian might know.

B

I think it's 48.

C

B

Now um so you're definitely safe setting it to 24 hours. You are maybe safe setting it to 48 hours, I'm not sure, probably.

A

Yeah, I think ian actually popped a comment in there below that says. The deal bot expects deals to be done within 48 hours from proposal you can think of it as a testing for a certain quality of service, so that um that adds some additional context to the answer. Okay, uh magic. Another question here from benjamin is using one shared nfs documented anywhere.

C

B

No, we should do that, but basically it is as simple as setting up a nfs mount on two boxes and then just adding the storage path on both ends. uh You will have to do. Storage attach dash dash unit on the first box, and then you will drop the init flag from the second one and it should just work.

A

Awesome, thank you magic. um Let me see a question here from youtube uh from august. How will miners at first obtain the initial file coin necessary for mining magic? I know the answer to this one. If you're happy for me to take it, um we have a falcon faucet, um which we can link to you.

A

um In fact, we can link here below um which will give you uh the kind of initial um pledge collateral that you need for experimenting on the test nets, um and one of the team will link it here below for you august, okay, I think we have time for maybe one more question: okay, uh this question's from hyun. What are some conditions that cause deadlocking in the scheduler, sometimes running ceiling, jobs, command, results in halfing process and breaks the scheduler.

B

There shouldn't be any uh if that happens, to you, uh paying us, send us a back report on an issue. The ceiling issue template at github.

B

Tells pretty clearly what debug data we want, but we mostly care about getting a curtain down from your miner. So if that happens to you, please report it because it really shouldn't happen.

A

Awesome, thank you. Well, I'm afraid. That's all we've got time for today. Folks, if you have any additional questions, please head over to the falcon slack, which we'll link here in the chat, a huge thank you to magic for taking the time to share his expertise with us and to all of you for joining us coming up tomorrow, we've got a live ama with the falcon team and a crypto week. One workshop with juan, where he'll be talking about the recently published engineering falcons economy report.

A

All of the info and registration details can be found on space, race, dot, valcoin dot, io enjoy the rest of your morning afternoon or evening and happy mining thanks. So much.

A