Filecoin Filecoin Mining, 10 Sep 2020

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: Filecoin Miner Show & Tell

Description

Welcome to the 3rd Filecoin Miner Show & Tell! During this event, we explore hardware, architecture, and share tips and best practices. Above all, we’re hoping to spark conversations, relationships, and themes that bring us all forward together.

This week we have presentations from BenjaminH, Charles Cao of NBFS Canada, and Hlib Wondertan!

00:00 Intro - Angie Maguire
2:44 BenjaminH
18:56 NBFS Canada
38:29 Wondertan

A

uh Really excited um for today's event, this is our third minor show-and-tell, which is very exciting um yeah, so pop in the chat. Let us know where you are calling tuning in from um and we'll get underway very shortly.

A

Hi laurence from canada very warm welcome. We've got some of your uh with some of your fellow canadians, presenting today uh hi edison uh edison is in china. Very awesome, emily from washington dc, always good. uh Welcome from austin texas, um wonderful, very exciting. Okay, let's get cracking!

A

um Thank you. Everyone for joining us for the third falcon miners show and tell this week we'll be hearing from some of our top minors in north america and europe. Tonight's event is part of a series of events throughout space race, um which is our incentivized test net competition.

A

You can see what's coming up on space, race.filecoin.io and catch up on anything you've missed on our youtube channel, which is linked below we've got a fabulous lineup of speakers today. So let's get started. Our first miner to present is benjamin, who was also named a file coin community champion in last week's space. Race report welcome benjamin.

B

Thank you very much I'll just keep my video on here. So um should I just get.

A

Started yep go for it whenever you're ready.

B

Awesome all righty, so hopefully you should be able to see my screen.

B

Yep yeah, so uh so I guess I represent some of these uh smaller sized miners, so I think last time we saw some of the bigger ones. I guess what I wanted to show and tell about today is how to mine efficient on a smaller mining setup, but also the terms that you run on an unmodified bluetooth. So basically, I haven't not been doing any custom coding any tweaks or anything I basically just rely on whatever is released and hope for the best from there.

B

That also means that the setup that I'm running is is kind of treat against. You know meeting the least amounts of problems with the unmodified builders, so um just to dive right into it. uh My own setup does not look exactly like this, but this is more like a representation of what uh what it basically could look like. uh So what I've done for these uh smaller setups is basically try to keep uh each of the of the individual processes, jobs as separate as possible.

B

So, as you see here on the top, I have my miner. We don't- which holds the demon, the miner and also the storage. So that's the one goes on top, um then I have a dedicated uh worker only doing a pc one.

B

I have a dedicated worker only doing pc2, and then I have the last one doing um yeah. That's so mainly the c2 job uh all connected with the tin kick or better, uh but this is kind of the the base setup for having a an environment that mine's pretty efficient.

B

So I know that there are a lot of small miners, uh partly complaining about having a single miner setup, partly saying it works. Fine they're, also minus having multiple single minor setups. I find that this setup is probably one of the yeah least problematic setups, to have also because I have pretty good control over uh the the individual processes, so making sure my lotus, daemon and the miner is not clogging up, because it's doing other stuff that is not supposed to.

B

um Of course, it is possible to run it as one, but I just find it much easier to to try to separate things. um So at least this is a this is kind of the base for what it looks like. um I also put some prices on here uh just to give you an idea about uh what is the size of this. um So you see it's it's a little bit, I think over 16 000. There are also some parts missing.

B

What I've tried to focus on here is that I have mainly been using amd epic cpus, I like to use the fx, because you get the the ecc memory and.

A

In general, it just.

B

Feels like a more stable platform, so I would say that I have seen very few, like hardware related compute, related errors on on this setup, and I guess in the end that's also pretty nice that you don't have that additional factor of uh what, if you have like a big corruption and stuff like that on top of it. So you know trying to minimize that the other part is that it's not like amd epic cpus are really you know super expensive, so so it's still doable now.

B

The setup here is basically again taking into account that you have the the different individual workers. What you see is that actually the miner in itself is pretty uh beefed up and well. We saw on the uh the emulated attack here this morning that that is actually a good idea, because, even though that you might have a miner that can follow the chain, then it gets really stressed when trying to do this under attack and so on. So so this, at least for my setup, has been working pretty well.

B

That said, you know you could joke that, so how well has just been going on, because how much power have you left, and I think the answer is. I do not have much, but I guess I will regain that sooner or later, but I don't think that I can. You know, uh explain that to be because of uh of this setup.

B

So another thing that I wanted to point out is that the setup I chose here for the pc one, which is a lot of people, is talking about the pcr one worker, like that's the only really important thing, and I I'm pretty sure that yeah it can be difficult to make work, uh but I actually find it easy just to have like a single epic here with like half a terabyte of memory uh and then about six terabytes of the nvme disk in ray zero. So you want as much I o as possible.

B

So that's one thing, and the other thing is that um when you have the six terabyte you could ask you know. Why do you need that much? If it's only 600 gigabytes per sector, you can have 10 sectors on there. Why do you need that much you're not going to have 10 sectors in parallel here?

B

So the thing is that that it takes some time before this. This storage is released again. So you basically need to make sure that you have enough storage here, that you can keep on patching new sectors but at the same time keep that until they have been committed to the network. So that's why there is some additional on it. I do not have time to run through this in detail, but the thing is that I can share this afterwards. If somebody wants some inspiration.

B

Basically, I have 10 gigabit network on all of them, but I think that most of you sooner or later knows that it's basically between pc1 and pc2, that you have this big bottleneck. You need to move those 600 gigabytes of files yeah. So uh let's just continue here. This is actually the one that I wanted to talk most about, because I pretty sure that this looks pretty complex, but this is actually the key to efficient mining on a setup like this. Basically, you have uh every hour, so you have from zero hour to six hour.

B

Here you have is just about every hour. You do an app piece or catch a new sector, so you have a process running there. Every hour you have this kind of rolling setup where you have about six sectors, doing pc1 in parallel, so the miner is able to do about seven, but it does not really make sense in this setup. So that's why I put it to six, because it also has to balance- and that's kind of the key point here is that you need to make things balance.

B

So, while you have the six sectors here, you would see that you also have six sectors going down here. So you have six sectors working in the pc2. You have six sectors on the weight seat. Actually you can have a bit more because we don't care about them when they're there, but you can also have about six sectors competing c2 and then they go into proving.

B

So the point is that if you do this and really tweak it- and you know in reality, I might do something like five to six. But if you really tweet this, then basically you can have this system uh spitting out a new uh 32 gigabyte sector every hour. um So so that, actually is not uh that's a lot more than 500 gigabytes per day.

B

It's actually more close to 770, but the thing is that in reality you know we have all these deal errors uh syncing the chain, lotus, updates everything and, as I said here, it takes about half a day to gain momentum, because you don't want to pitch all six sectors in once. You want to have it rolling like this, and that is also to optimize how much storage you're using.

B

So uh I can definitely tell this one and I do think that it's a little bit complex to look at. But if you see the green line here, you basically see that at a point in time you have six running in parallel. Here you have a pc2 running, you have something in weight seat and you have a c2 running. So that's the idea um and just to give you an idea about what it looks like in in bluetooth.

B

You have not that many here, but you basically have your c2 jet running. You have your pc2 job running. You have a bunch of c1s running and you will see a lot of stuff over here because you have weight deals. You have something that goes packing. You have a weight seat. You have something going to piece two and so on, so you want to see that it's got like widely spread all over the place here. So that's kind of the idea to make this this work.

B

So uh this is just some pictures from my basement and I already showed this to a few other miners which find it's pretty fun. But but the thing is that this is my uh my basement, where I just have just done stuff thrown around. This is my miner. This is actually my pc one, which is a hpe server. Basically using a non-hp memory, non hp, nvme disks, just using it like a regular computer and it works perfectly. Then we have the pc2 and the c2 up here.

B

So over here, that's my normal networking rack where I have the firewall. I have my switch and then because this is actually a deep loading setup. It's actually have some pretty powerful switches going on here. So this is actually uh one of these um uh kind of rocky setup, where you can do a really fast uh file transfer over uh almost a lossless connection, so that also helps a lot now.

B

The other thing that I wanted to point out here is that we also have some- I I basically use liquid cooling to uh to cool the computers, so my basement is only about 25 degrees, but I basically pumped this over in my uh in my flow heating, so I remove about 200 2000 watts here at about 40 degrees, and it is pumped uh with the this is cyclical and is pumped over into the the dirty water going into my uh my heating system in the house.

B

So uh so that's also a way to uh to utilize the energy coming out of this. So that's basically what's easier heating, my house right now- and this is just a look inside one of these when you have filled it all up um so another thing that I or this is actually the last thing I just wanted to point out is: I also have this other one called the gogo intel miner and the thing is I wanted to see how crabby hardware I can actually run this on.

B

So the only thing where I check it out is on the miner itself, where you see it has a pretty uh beefy c and gold and a and a 2080 ti. So that's pretty awesome to to do that part, but the thing is that I really wanted to do some really low-key stuff for the other things. So my pc2 worker here is a super old intel xeon, which is totally beat up, but it can still do a pc2 in a little bit over an hour and then this one is, I'm really proud of.

B

That's also a totally old. You know totally underpowered uh ceon um and it does not have much, but you know still, you can do it in two and a half hours and what I did was basically just put two in to do c2. So now I have one to do a pc, uh pc2 and I have two to do the c2 and I end up having about a sector coming out a little bit over uh every hour. So this is really you know the kind of basement stuff that you know some.

B

Some miners would work and I think that it's really interesting to see how low can you actually go because it comes down to that the pc one, which is the one that you saw before? That's the one doing pc one, but that is basically the only one that really needs the amp and for the rest of it. You know you can be a lot more creative in how you set it up. So I guess that was pretty much. What.

C

B

To say um so, it might be. There are some questions here.

A

Yeah we've got, we've got one question here in the chat: benjamin um have you from edison um he's asked: have you tried using worker to do p1 plus p2.

B

P1 plus p2 yeah, okay, so pc1 pc2 yeah. So what happened- and I don't know because I haven't tried it lately, but what happened when we did it with the earlier ones in calibration, where I kind of designed this was that basically um it only does a pc one.

B

So so, when you have a new sector coming in so basically right now, if you set your uh uh your weight, uh wait for deals to like every hour which mine is set up, then basically it would just keep on pushing in new uh pc once and you know it wouldn't do a pc2 now. Obviously you are able to do this on a on a single minor setup.

B

So you can tweak this, but I find it really hard to you know to mix the things what you could do was also to have two workers on the same machine. But then again you know it gets a bit more complicated and you know it's it.

B

I I think that this, of course, you use a little bit more hardware in here, but, as the hardware itself is not really that expensive, I think that you just get more throughput from just having things isolated.

A

Yep, that makes sense. um Thank you so much benjamin, like, first of all, I'm just completely blown away by um your mining set up heating. Your house.

B

A

How did you even like go there? Were you just like? Oh maybe this will work like. I don't understand how you like come to the place where you're like? Oh, maybe this can connect to this so um yeah amazing.

B

Yeah, no, so from my work, I I work with high performance computing, but I also have this kind of fetish for water cooling and you know, that's just escalated so.

A

It seems like it seems like that's, that's so amazing and a quick question for you. um How did you kind of first come across file coin like where, like what was it that kind of motivated you to to mine on file point in the first place.

B

Yeah, so I did litecoin back in the days and you know I I think it was probably the size of the ico. Like you know, there is something to this that we need to investigate figure out. You know what what can it do and so on.

A

B

I think it's a general idea that you know if you can't do trusted storage and in the end, maybe just you know not really have cloud like we have today. You know using for for heating our houses instead of heating. You know ridiculous. Nothing, then. I think that you know that would be much better and even if you can run it on a mobile phone and use the storage there, then you know we're really getting somewhere. So.

A

Absolutely absolutely well listen! Thank you so much again um for taking the time to share your your experience and your expertise with us. Also, congratulations on being a falcon community champion. Thank you so much for helping your fellow miners um throughout space race and congratulations on being number seven on your continental table very, very exciting.

B

Yeah, let's see how it goes.

A

Awesome all right! Thank you so much benjamin um okay! um Next, our next presenter is charles carr, representing the nbfs canada team, who are currently ranked second on the north american leaderboards over to you, charles.

C

Yep just share my slides.

C

Yeah so yeah thanks everyone to give us the opportunity to present uh our staff. Yes, uh we are running uh freikon cluster mining in canada, montreal so uh so. First of all, I would like to introduce ourselves. uh Well. Everybody was curious about what we call mbfs. Actually, the nebula ai file system. uh We were doing the decentralized ai cloud computing in the past days.

C

You know our white paper in 2017 were saying that we would like to have the storage solution based on ipfs, so we can doing the decentralized storage um with the ai together. So we are waiting the day when there are a good solution can provide the storage. uh At the end, we found that the fragrance will be a very good solution potentially for us, so we jump into the mining industry to find how it works. So we have been working in the project for about three years.

C

We get a one million dollars funding from canada government. So it's quite interesting we're going here and as those are, the servers currently we're using way when directly using the servers from hp, um high performance computing servers and we are hp, official part business partners. So we with the p1 we're using the md cpu server, basically, is the hp 385 type and for p2 worker we're also using hp servers. Basically, it's a hp server with the dual core external intel server, but we added one gpu and 2080 ti into it.

C

So for the screen server, it's just a normal storage server, micro, super micro or any types it will be. Okay and vending has already introduced lots of details about how to assembling servers, how to doing either those um and storage plug plus how to scan them. So I want to repeat his work basically like we have very similar design, except that we didn't split the p2 and the c1 c2, so p2 and 72. We are in sandbox with one gpu card by the way, and then we put everything in a data center.

C

So we are using data center for mining and we have a third center in montreal, which is it can support 30, kilowatts per rack so, basically like uh with certain vehicles, we can have all the servers up to 20 in the same rack and on the rack server. We have one node server running. We have one miner and uh about three to four p1 servers with p2 servers working together.

C

So, according about that, we added the money monitoring scheduler, which can help you monitoring any service downtime. Anything like overuse used. We have an anti-spend module because early is the date. When was there the number one in north america? We get an attack from traffic. At that time we build a new anti-spam shield which can detect a malicious behavior and add to the federal policy to block them under the schedule running every day to get detect those behavior from script and logs and lots of other strings.

C

We have a semester for service which can enable our game to analyze how much production per day per hour and also put into a scheduler to monitor those service done and something not working.

C

At the end, we have authentication services built on top of the fargo infrastructure, so for people who want using a mobile to monitor each of the servers which they want to rent, they can see how it works using the username password to get a credential to access it. So this is the software in architecture reviewed and those are the dashboard we are using uh by the way. This is not the real dashboard we're currently using now. This is a conception that we use.

C

You can show the. If you have a multitude location, you can see how many servers running the traffic flow and up and down time- and this is the one screenshot is taking from our production server. So we can see that there was place, for example, this one we have a 99.99 percent is used. uh Obviously this is the p1 worker. It has the multi-threads parallel about 16 to 17 parallel processing, so it depicts about. So we need to be careful about this one.

C

We need to distribute load to other servers and we can see that we have the partition used about 80 or 70 percent. We need to take care when it arrives. 90 percent were certain alerts to our notification and mobile. So our engineer will take a look release some space or decide if we need to add a extra hard disk for help.

C

A

C

See the throughput per hour per day, you have about 30 terabytes input and output, so sometimes it goes too high means some service inside the smart has problem. Those are the. uh If we want to check any of the services working, we can check the gpu usage. Usually when we have a gpus like this, which means it's healthy, which the pipeline is not broken. You continue doing the job. This is that big gaps between those spikes, which means that you have too much idle time.

C

You need to check your schedule to see if there are some stacks, for example, message points for or something like you have lots of: p1 jobs, idling or something will stack somewhere. You don't know you need to log in and check it and we have a dev development environment just for testing. So writing a customized scheduler to grab all the status.

C

So you can see the different color on different silver, how many jobs currently running p1 jio, p2, job and c1 other pcs, and you can see the memory usage cpus from another direction so plus this one. So you can pretty much get every information you need yeah.

C

So, but if we have a lot of issues uh during or we can see challenges during the setup due to the corporate nineteen in canada, the supply chain because of big problem, uh it already takes two to four weeks from ordering to shipping and when parts arriving, sometimes you need to customize and upgrade by yourself, because the hp servers usually have the original design.

C

But if you want to add new pieces, for example, you want more memory or you want a more cpu kit- it's not just expensive, but it takes one month or two, even two months, shipped to you, but for mining. If you want to catch up with the meetings there's, basically it won't work. So you need to manage this. However, testing is another heavy job. We have so many servers, lots of different cables and lots of components and some components which was remarkable by the community, something like a nvm or something like that.

C

You just cannot find it in canada. It doesn't exist either you order it from china or already from us, but in canada. Just don't have those equipment there, so you need to find some alternative solutions and you need to just balance the trade-off timing and the cost.

C

So operation is extremely difficult, because uh all the hardware is uh working on the high pressure almost 100 to use it. You might have broken in ssd, you might have broken your network card and you need to set up lots of monitoring and uh and you need, and then the data center is not in charge of maintenance. Those things you need to work in visit. You need to study people schedule the service you have. You need to have backup parts, so you need a management budget to say. Okay, we have 100 servers running now.

C

We need to have about 10 to backups or 20 backups. um Those things is a very complicated, unique monument and because of the the current space race, we don't expect it that we didn't expect it that the team were released on daily basis. So originally we are not a design for this, we think is that we have one thing fit up and then we can just running after 21 days. Everything is done well, okay, but actually it's not like that. We basically have every race per day, so we're struggling.

C

Do we need to follow the upgrades or just waiting a few release, then we're doing upgrade that will save lots of time and effort, but at the end I decided to follow the official releases page where to upgrade every day, even though we lost about 20 to 30 percent of the output, but I think it's worth it because from those things we found that we can hear lots of people. We found a lot of bugs in the day one, and we found a lot of things like uh people also incumbent need to help.

C

But if you see that uh no I'm sorry.

D

C

Using 0.2 5.1, so I cannot answer questions. I don't think that's helpful for the community, so we decide instead of gain high rewards. We decide to find more bugs, find them have more community members. So in that case we're doing that one please become challenges. You need to management. So there's this is some solution.

C

We've found so first of all we're building an erp system for trading the inventory and the quotations for when the conditions comes in from different, renders we immediately put in the erp system, and we manage our equipment inventory to see how much, how many ssd we remaining, how many hdd will remaining and can we need some quotations fill up fulfill our inventory. So this is not done only automatically. We have invoice system. We have quotation system, everything added together. So when there is something happens, there will be trigger alert as us to fulfill the equipment.

C

So automation pipeline is important. We have auto build system on the servers and it can delivery all the views to all the servers at the same time that other script, when workers offline, there's monitoring services, where trigger agent on each worker to wake up those services.

A

To automatically.

C

Reconnect to the pu to the miner. I think miners also aware that your workers, occasionally we are offline during whatever reason uh disk io was just ac keynote received. In any case, it will be found. We have two steps of uh monitoring as a watchdog running on the server and uh watchdog monitoring service on remote worker, which is our own work.

C

So basically, we have another distributed system working with a fraction distributed system together, it's a little bit complicated and also we've been using lots of container based technology for uh continuous delivery, so building the anti-spam system to monitor those management behavior to block those servers.

C

So after that we didn't suffer, the problem of the log, misses the wind post or lost low power due to the attack to the message attack to your miners uh and in order to find the best way to manage those hardware p1p2 how to working them together, I go to a simulator which can accelerate the process using deep learning technology, so you can simulate thousands of times how the workers, working with each other and exceptional handling, use the software instead of you defining one configuration you put it on running and you spend another 12 hours, then you know that okay, that works or not.

C

No, we don't do that, we're using uh dividing technology to involve those systems and it works very efficiently. It can self involve a self evolved with a job process.

C

Yeah and here comes our future plans, so we would like working with china suppliers for a lot of important parts and uh to scale up the money, and we want to fund the partners in china doing the marketing and the coal mining in north america, because north america for now is still low in storage compared to china and asia. So we wanted to working with them together to building another mining location, a bigger nine location in canada or even us so, and we have very good relationship with the local government.

C

We can help them to working with the local government and how to pass the compliance, follow the law and everything locally, and we would like to contribute to code for operating and testing, for example, those anti-spam software, those continuous monitoring software components. We want to contribute as soon as it is mature enough for open source, and uh then we want to working more with other management committee.

C

We get lots of help from the community, we get uh lots of interesting information and they have us a lot in the early days and we want working with them and contributes yeah. This is our team, and this is me actually. Last year I was visiting china. This is our office. Now our people install the servers and developers coding, testing everything yeah, that's pretty much uh for our.

A

C

A

Awesome, thank you so much charles. That was uh really really interesting and it's especially, I think, um valuable, to kind of hear about some of the problems your face you faced and and how you've overcome them, the solutions that you kind of implemented. So that's really really wonderful, and I know that we've seen a lot of bugs come in a bug. Reports come in from your team. That's really really helped and kind of like improve the system. So thank you so much for that.

A

We've got a couple of questions in the chat for you, stuart berman, says very professional mbfs. How much redundancy slash resilience? Does your system have.

C

Yeah, but now for our system um we about, we have about twenty percent of redundancy, uh usually like. If we have one p1 worker or two p, one worker offline, other workers, we are taking more jobs. Usually we reserve twenty percent of the capability, and uh this is from the software part and the hardware part we have components for replacement as well. Usually if the system is done, we need about six hours to get it offline on the run line online.

C

So the replacement plan is about six hours, but the service downtime is within 20 minutes.

A

Wow, okay, great, thank you so much. um Another question here from ask ender um when p2 plus c1 plus c2 will p go to another machine to continue c2. If c2 comes first, will it start c2 first.

C

Yes, that's exactly the problem we are facing now, so that's some people was asking us um to rewrite some scheduled customized scheduler. To so can better arrangement, those sequence and that's something uh the next step we want to do after the space race is completed. We would like to write some code more deeper about the scanning service.

C

I think why is working on that is that he has uh something is interesting, can reduce the p1 to 101 119 minutes, so that will be interesting, but uh I I'm thinking about that because for me I really don't want to touch this part of code. I think the team can do the best job for that, but if, when magnets are launched- and we still don't have that one ever service considerably write, it.

A

Awesome well, it's like, like we say it's like a community effort, so um yeah definitely uh like always uh welcome contributions um for everybody. Okay, one last questions for you, charles: how are you monitoring infrastructure.

C

uh So, for instance, we have uh also software working together, mostly was created by ourselves previously in the cloud computing. So we have the um software monitoring the rack, the hardware so with basically you they provide because they are servers, they have ipmi. So you can read the disk, the temperature, everything you can read it, and uh so the data matrix in collection is using graphing for sure, but for the service jobs. That is uh another service. Writing by myself. We're writing a service.

C

We create a database, so we go through all the logs with tokens using the api talking to the bluetooth api and uses a minor api. We gather the real-time information into the installation database. Then we can doing the data analyze we won't take the to. We won't describe it as a big data, but we do using lots of big data technologies to extract those informations to building the graphs.

A

Awesome very cool charles, listen! Thank you so much again for taking the time to share with us today and uh very best of luck to you and the team for the rest of the competition.

C

Thank you guys.

A

Okay, um our final presenter for today's show and tell is wondertan uh who is competing in the european continental board, with an impressive five six terabytes in storage added to the network. Wonderton. Are you with us.

D

um Yes, I am there.

A

D

A

Hi perfect, whenever you're ready just share your screen.

D

D

So um I won't um let me start, I'm cleve wonderton. I live in ukraine and I decided to go into space race for many reasons I live in ukraine, and this is this talk will be much different from what you've seen before. There won't be any kind of special uh hardware like a lot of team working on uh updating the codes.

D

uh This is kind of like just just a simple miner who wants to commit some storage for file point and maybe help uh it's to evolve right now or see any future perspective from it kind of this thing. So I would like to tell you why file point.

D

uh First of all, um I am a developer and I've been in cryptosphere for quite a long time, and I had was looking for pl uh from like the first time I've seen ipfs and um that, like their project, pilecoin made me like interested, and um I wanted to join this join it anyhow.

D

Then I uh saw a space race events and um and and and decided to join, like uh even I had a really short time frame for that, because I wasn't in uh in ukraine for for when.

B

D

To do that, and I had like uh actually two weeks before space race to prepare for them, so those like last month for me was kind of intense to manage all of this uh while file coin more. um I really think that uh file point is uh has great value for a crypto community and for uh like technical evolution of humanity at all, it's kind of different. They have really strong economic model uh having power having storage as your stake in the network, uh provable state, approvable storage.

D

This is really cool, uh so, like I'm, really, uh I'm sure that file coin is will be uh really helpful and demanding in future and be very popular crypto projects. So I decided to join it, that's kind of well, so um the team and investor. ah That's a mistake. Team.

A

D

I would like to say here so um the team is only one person me who decided to do that and found some investors for this to. I convinced them that this is a really great idea. One of them is my father and other is my like my good friend also I've put some money on it and we bought a total like seven rigs. For this, uh that's uh that's kind of a big amount of money for just uh mining in your basement. I think, but that's that's all so about hardware and architecture.

D

Getting professional server hardware in ukraine is is an issue to do this in two weeks, so uh the fast decision was to uh to buy a lot of um available hardware and available, like imd imd processors like gpus, that are popular in gaming sphere.

D

So I just bought seven.

D

Seven machines that have uh same specs, they have imd cpu um 3790x. They have geforce gtx 2008 ti three terabytes of ssds, like that. This is very popular setup that was around on slack, so I've just decided to use it and to mine about architecture. So this is architecture I want to show you. This is just a photo of custom-made shells for this, so this is uh seven um machines, seven rigs that are interconnected with the 10 gigabytes uh switch uh and uh that's all so.

D

Currently, there are just seven separate miners that are combining uh space race, but in future like before may natas, I think I will still have time to um find a like best uh way to uh use this hardware efficiently for all the uh committing steps uh not to uh have any of the hardware just sitting around without work to combine them in one minor. That would be a very interesting thing to resolve so yep.

D

Next, so um also, I would like to tell you about their experience in a space race, because when you start to dive into and try to understand, uh like how everything works uh and how to set up everything, you have only two options for space phrase: there's dogs, documentations, maybe specs and uh slack so uh docs and specs, are aren't very helpful.

D

They just tell you how to do some basic thing on how to install dependencies, maybe and initialize your miner, but nothing more special, because when you dive deeper into um sectors uh how she doing works, what are the steps?

D

uh How this all works? What time it takes for every step to complete how to improve work distributed in scheduler? What workers to do how to set up workers, and do you need them? There is a lot of things you need to understand before actually starting to mind, so I've got a lot of nights just uh reading the and finding the answer for my questions since like if somebody already asked them to understand what and how everything works like take everything by pieces and to somehow like create a whole picture.

D

In your mind, uh also as a go developer, I had got some experience and, like just uh I just.

A

D

Code, because it was this simply then understanding it, some concepts from slack, and even uh there is a lot of misinformation there.

D

There are three like four um parameters in um configuration for ceiling like the max deal sectors, all this stuff and uh they're really not intuitive from like from the first side, you see them and how this work, what what they are doing and a lot of people telling different stuff on slack that are not meshing, and I just decided, like okay, I'll, read the code and understand what it does, and that took me much less time to understand that trying to to get this from slack um like- and I I got my hardware like two days before the start, so uh there was a huge rush and I got many mistakes to actually stop all of this.

D

First of all, like I understood that I can do one minor in my um in my setup and just to run uh workers on every machine and the scheduler will be able to like do the whole work uh itself, but then I realized that that isn't working and I haven't no other options just to uh set up like a minor on every worker.

D

um So I've done a lot of mistakes and lost a lot of time to actually gain some, uh some more terabytes and and actually like. My current position in europe would have been higher uh for the reason I said, like I've lost some time on understanding things and right now, like I've, lost a lot of power through uh post submission failures. uh For various reasons.

D

I think you know them on slack. uh So I'm like my current, uh like power, is about like 12 13, not, I think 12 terabytes is uh visits. So yes, and thanks like that's all, I think I would like to tell you. This is kind of I think, not uh usual uh minor, show uh presentation, but I think that you you liked it. So thanks.

A

Yeah, that was that was awesome um and, I think, actually um ben our first speaker had a kind of similar approach as a kind of independent minor.

A

So it's just great to see like a real variety of approaches to falcoin mining um and also uh great, to see you moving back up the table again, um like you said, you lost power there for a while, but um it seems like you're and there's stop, there's still quite a few days to go in the space race, so I'll be excited to kind of watch you, as you continue to progress in the competition.

A

um There's a question here from you from ben. Actually, um if you run the servers as seven individual miners, uh then one, why did you set up a 10 gigabit network.

D

A

D

A good question I just bought it uh for like originally, I thought that they should uh exchange some data, uh but that setup just to be honest, that switch just like, sits there without work, and this is for future braid, where I'll like combine this all into one minor and the 10 gigabit switch there is, is a must.

A

That makes sense awesome. Well, listen! Thank you so much again uh leave. I know that.

D

You've done yeah, and I want to tell thanks the benjamin who helped me a lot to find out and understand some things so.

A

I think I think benjamin's gonna have to win the community champion award again this week. He's uh he's really been awesome, and I know that he's super active in the slack channels, and I know that you've been helping people to leave. So thank you so much um okay.

A

Well, unfortunately, uh that's the end of uh the minor show and tell today I'd like to close the session by saying a huge thank you to our wonderful presenters, ben charles and leave for sharing their falcoid mining expertise and experience with us and to all of you for tuning in, if you'd like to present at the next show and tell them we'll be posting a sign up form in the file coin slack shortly.

A

You can also catch up on all of the events that have been taking place throughout space race, on our youtube channel, which we'll post in the chat below and we'll also link to in the far coin, slack we'll be announcing the next minor show and tell very soon in the meantime, enjoy the rest of your week and your weekend and happy mining. Thank you. So much take care.