Chia Plotting Service in India

[booo]

• HGST's 4TB drives have a lower failure rate over 5 years vs their 8TB over 3 years & their 12TB over two years
• Toshiba's 4TB has a zero failure rate over almost 6 years vs their 14TB over 8 months
• Similarly, Seagate's 4TB, 6TB, 8TB have a lower failure rate compared to their 10TB, 12TB & 14TB (one of two models)

you are correlating different manufacturers with different processes and technologies with size of the drives. while omitting factors like vibration, temperature, power and many more things.

Now if you wanted that 16TB for video editing

NOPE… wrong use. should be nvme.

 

[dafreaking]

Well you are both sort of wrong. If you are editing video it depends a lot on the codec you use. For a typical project we have about 500gb of raw files. These are still compressed so for smooth editing and sanity I always need proxies so now I'm looking at a TB of Data. Once the proxies are in place it doesn't matter as much if the data is on a raid, flash or a single drive. Unless you run a large enough establishment the shucking vs buying regular internals doesn't matter. I usually make 3 copies of the source files and 2 copies of the proxies before the project is delivered. Once it's done it's down to 2 copies of the source data and the proxies are deleted since if they are any changes (which sadly there are) I don't really need the proxies. 2 x 8TB externals is/was cheaper than a 8tb Internal. So at the same price I have one work drive and one offline backup.

NVMEs barely help with regular video editing a regular Sata SSD is more than sufficient.

The main reason why I wanted to switch to a 8x8tb array is more for having almost all my stuff online as opposed to spread over multiple disks. Backing up various assets after a while becomes a huge pain. So the way forward is an array with 2 disk redundancy and an offline copy.

 

[rsaeon]

you are correlating different manufacturers with different processes and technologies with size of the drives. while omitting factors like vibration, temperature, power and many more things.

That's not my data, that's Backblaze's data. I'm not correlating anything. I'm pointing out what they presented. It's pretty clear, higher capacity drives are more unreliable compared to lower capacity ones — regardless of the manufacturer. Vibration, temperature, power, all are irrelevant factors because all drives are subject to the same conditions.

Have you seen how Backblaze deploy their drives? I assumed you were aware but your counterpoint tells me that it was a wrongful assumption.

NOPE… wrong use. should be nvme.

Saying 'nope' doesn't change anything and is unhelpful in carrying on this discussion. Because, whether it's a valid use case or not, it's beside the point. I was trying to hypothesize a legitimate use case for an Exos drive with no warranty for someone on a budget that wasn't media consumption or bragging rights.

If you NEED 64tb will you get 4x16tb or 16 drives of 4tb each just because that's all that's available on Amazon?

I would choose the cost effective option, but I'm also not a 'all eggs in one basket' person, so I'd choose 2x 8TB over 1x 16TB. Or 4x 4TB over a 1x 16TB. I've learned never to have a single point of failure in anything related to technology.

To answer that particular question, I see no difference in 16x 4TB's or 4x 16TB's. I might just prefer the 16x 4TB just for the insanity of it. But that's just me. And it makes sense to me. For all the reasons I explained above.

 

[booo]

Have you seen how Backblaze deploy their drives? I assumed you were aware but your counterpoint tells me that it was a wrongful assumption.

i work in storage industry. we deployed data centers bigger than backblaze.

on the NOPE part. i agree exos are not for consumer consumption. they are enterprise drives and their price includes validation etc… for enterprise use and regular consumer using them without warranty defeats that purpose.

That's not my data, that's Backblaze's data.

but the conclusion is yours.

 

[rsaeon]

i work in storage industry. we deployed data centers bigger than backblaze.

Ah, I remember your post with all of those Exos drives from a couple of weeks ago. That was pretty cool. Where in India do companies use over 150,000 mechanical drives?

 

[NotMyRealName]

Maybe you two should've taken a look at the failure statistics I linked from Backblaze:

• HGST's 4TB drives have a lower failure rate over 5 years vs their 8TB over 3 years & their 12TB over two years
• Toshiba's 4TB has a zero failure rate over almost 6 years vs their 14TB over 8 months
• Similarly, Seagate's 4TB, 6TB, 8TB have a lower failure rate compared to their 10TB, 12TB & 14TB (one of two models)

The data's right there, the general trend is that lower capacity drives last longer than higher capacity drives. It's impossible to miss.

There are a few outliers, and 16TB drives have less than six months of statistics so I didn't include those in this quick comparison.



Respectfully, 2TB drives first came out 12 years ago. Perhaps the knowledge you gained from experience in the years before that is from a time that does not have any relevance now?

Drives used to have terrible reliability in the pre-1TB days, but it's obvious that lower density designs today are far more mature than higher density designs (and thereby attributing to their lower failure rate).



What good is a 16TB exos with no warranty? It's a sports car with no seatbelts.

If your main purpose for a large drive is media, then SMR/CMR has no meaningful difference since all you're doing is reads after the initial, painful copy of data.

Most FUP's these days are under 4TB, you'll spend four to six months recovering from a 16TB drive failure vs just over a month for a single 4TB failure.

Now if you wanted that 16TB for video editing, in a mirrored array, then two shucked 16TB's make for a convincing argument, on a budget. However, if you're editing videos as source of income, it's more responsible to buy internal drives with warranty instead of shucking.

Buddy, you're arguing for the sake of arguing. Doesn't make you right. Or industry experts wrong.

And this is not the first time nor the last someone on the internet will use backblaze's data to try to 'prove' their point. The data is what it is. People keep correlating it to fit their argument. I will not get into this nonsense because i've seen it way too many times before.

And my experience with hard drives didn't reset to zero when 2TB drives came out. Aren't you conradicting yourself saying pre-1TBs were unreliable but at the same time smaller HDDs are? And an old 40gb was one of the longest running drives i had, and then a 320gb seagate even longer. In fact it still runs if i give the pcb a quick rub with IPA but will stop after a reboot.

Let me worry about my drives' warranty. I'm informed enough to make the decision whether to shuck or buy retail. Ditto SMR.

Or 4x 4TB over a 1x 16TB. I've learned never to have a single point of failure in anything related to technology.

To answer that particular question, I see no difference in 16x 4TB's or 4x 16TB's. I might just prefer the 16x 4TB just for the insanity of it. But that's just me. And it makes sense to me. For all the reasons I explained above.

Says all i need to know. I won't bother explaining power consumption, cooling and space constraints, since you already know it all. I concede defeat. Now let me buy my exos in peace!

This is no longer a constructive discussion, but absolutely absurd logic to prove a point.

I keep trying to stop posting but you won't let me! Bye now...

i agree exos are not for consumer consumption. they are enterprise drives and their price includes validation etc… for enterprise use and regular consumer using them without warranty defeats that purpose.

I need a 16tb drive for my home PC. why should i not choose an exos when there are no other options around lol? FYI retail exos costs like 50k and the shuckable one 22k or something. That is also a factor when discounting warranty no?

 

[rsaeon]

I'll refrain from responding to your charged statements. I've tried my best to understand the reasoning behind your logic, countering your explanations with an alternative thought process for everything from what cryptocurrency is used for to hard drive failure reports and it appears that you're just here to vent your frustration of limited supply of high capacity external drives at someone who's trying to offer a cost effective service to the forum members that have inquired about it.

The blandest truth is that there's money to be earned with Chia (as is with mining) and we're in the worst economic contraction that India has ever known since her independence: https://www.google.com/search?q=jobs+lost+india&hl=en and people will do what they can, to earn what they can.

 

[rsaeon]

There's a healthy interest for Chia farming here by members but the atmosphere in this forum is a little unkind towards farming/mining in general so I'm flooded with messages in private instead. However, there's valuable information that should be publicly shared so I'll update this thread intermittently with new developments and discoveries about plotting/farming Chia plots without dragging other member's names who are also plotting/farming and are looking for ways to streamline/optimize their setup.

Up until now, I've been creating these 101GB K32 plot files under Windows 10 VMs across a cluster of Proxmox hypervisors. This is because my little server farm has prior commitments that cannot afford a bare metal install in any of its systems. Because of this there is some performance loss but they should be minimal and you will see better numbers if you're going to be following along with a native install.

The most powerful systems I have are with a Ryzen 5900x (purchased local), with an Aorus B550 Pro V2 motherboard (Amazon) and a 64GB (2x32GB) kit of GSkill DDR4 3600 (PrimeABGB). System drive is a pulled 512GB Samsung PM961 sourced from a refurbishing dealer, and installed on PCH lanes. Temporary storage is handled by 3x 1TB Silicon Power P34A80 (Amazon) that are all on CPU lanes, one in an M.2 slot and two in a m.2 to PCIe Adapter that MSI bundles with some of its motherboards.

This B550 has a surprisingly flexible bios that allows you to bifurcate the 16x CPU lanes in 8x + 8x or 4x + 4x + 4x + 4x configuration. If I was doing this all over again, I'd choose the Aorus B550 Master, this is the only AM4 motherboard that has both of it's m.2 slots connected to the CPU, and so the first 16x slot operates at 8x in this configuration. This allows you to install 3 NVMe drives without any expensive/rare PCIe adapters (with the third one using a simple 4x adapter). B550 motherboards also appear to be perfectly happy to have the primary GPU installed in PCH lanes.

Going back to the 1TB drives, they each have a 150GB SLC cache so in a software raid configuration, they're good for 450GB of continuous writes before speeds drop to around 750MB/s each, or 2250MB/s in Windows Storage Spaces or Linux MDADM raid. (Samsung's 970 series are best suited for this, both the 1TB Evo and the 1TB Pro. The 980 Pro is a downgrade from even the 970 Evo.) The NVMe drives are passed with their IOMMU groups directly to the virtual machine.

The other other VM on a 5900x system is OpenMediaVault with MergerFS. It's assigned 2 cores and 4GB of memory. The main plotting VM is assigned all 24 cores and 56GB, leaving 4GB for the hypervisor. Having the storage VM and plotting VM on the same system allows me to use a virtualized 10G ethernet between the two to transfer the plots, and the storage VM can later be migrated to another system for long term farming.

People aiming for over 20TB/day prefer Intel's Clear Linux but I have not been able to get that working without issues in a VM, and I lost about three precious days trying various configurations (Linux is not my forte). Yesterday, I decided to revisit installing Linux because of apparent performance improvements brought upon by an unofficial code optimization published here: https://github.com/pechy/chiapos/tree/combined

I installed Ubuntu Server 21.04, along with standard installs of chia-blockchain and Swar-Chia-Plot-Manager in the home directory. Try and resist installing them outside of the home folder (causes permission issues) or renaming these folders after you've configured them, because it will break the install and/or the python virtual environment (I also learned this the hard way). The 3x 1TB drives were configured with optimal aligned partitions using 95% of available space in Raid 0 and formatted XFS, mounted with defaults,discard,noatime in /etc/fstab. I'll later set up a cronjob for daily fstrim to keep the SSD's performant.

Assigning 6 threads and 3584MB of memory in the config file for Swar, these are the numbers I achieved for a single K32 plot with the official chiapos (Chia Proof of Space executable):

$ cat /mnt/md0/logs/official_chiapos* | grep Time Time for phase 1 = 4965.465 seconds. CPU (216.350%) Sat Jun 5 12:45:43 2021 Time for phase 2 = 3413.913 seconds. CPU (99.980%) Sat Jun 5 13:42:37 2021 Time for phase 3 = 6781.013 seconds. CPU (99.940%) Sat Jun 5 15:35:38 2021 Time for phase 4 = 363.775 seconds. CPU (99.970%) Sat Jun 5 15:41:41 2021

That's about 4 hours and 18 minutes per plot.

After compiling and substituting the chiapos from pechy's combined branch, the improvements were substantial:

$ cat /mnt/md0/logs/pechy_combined* | grep Time Time for phase 1 = 2970.301 seconds. CPU (286.930%) Sun Jun 6 01:38:11 2021 Time for phase 2 = 3069.988 seconds. CPU (101.820%) Sun Jun 6 02:29:21 2021 Time for phase 3 = 2625.610 seconds. CPU (145.270%) Sun Jun 6 03:13:07 2021 Time for phase 4 = 247.394 seconds. CPU (103.440%) Sun Jun 6 03:17:14 2021

That's 2 hours and 28 minutes! Overall a sustantial improvment at minimal increased CPU usage. Memory is usage is apparently much higher from other people's reports though I have not been able to log this through the CLI.

That's it for now. More after I figure out Input/Output errors for CIFS shares under Linux.

 

[roofrider]

Do keep updating the thread with info. I won't get into Chia but would like to know more about crypto and stuff.
Others mining Chia shouldn't hesitate to post too. This should be about sharing information and learning.
A separate thread can be created to debate if needed.

 

[tech.monk]

Speaking so much about HDD. Anyone know anyplace to buy like cheap large capacity drives? Buying a new one is really expensive these days and finding second hand at cheap prices is difficult. Any suggestions?

We do have good news about it; Seagate introduced new tech - you may call it innovation or market opportunity considering the right time.

New Seagate disk tech could drive down the cost of high-capacity HDDs

Seagate reportedly secures a dozen customers for new drives

www.techradar.com

 

[Mann]

There is even a plotripper SSD in the market. The manufacturers are serious about it. With the price of GPU going to the moon, I guess this is not that bad for some who has most of the hardware with them. Worth a try. One big difference between this and any other type of mining/plotting is that for 99% of the other coins, you mine and you move ahead. Here you have to keep holding on to the plot if you need it to win a Chia. Some thing like a drive failure could be disastrous to your potential for a win.
Even if you are pooling???

 

[NotMyRealName]

I'll refrain from responding to your charged statements. I've tried my best to understand the reasoning behind your logic, countering your explanations with an alternative thought process for everything from what cryptocurrency is used for to hard drive failure reports and it appears that you're just here to vent your frustration of limited supply of high capacity external drives at someone who's trying to offer a cost effective service to the forum members that have inquired about it.

The blandest truth is that there's money to be earned with Chia (as is with mining) and we're in the worst economic contraction that India has ever known since her independence: https://www.google.com/search?q=jobs+lost+india&hl=en and people will do what they can, to earn what they can.

I'm only replying again because I noticed you edited your post and now accusing me of 'charged statements'. As I and other have pointed out, it's your comments that are bordering on fantasy and delusion, and not at all reflective of fact. Just because you define your perspective as 'alternative' doesn't make any of it true. Instead of sticking to the topic of crypto mining, you are the one going on and on defending chia miners saying that the hard drive shortage isn't real. The facts are out there plain as day to see.

And buddy, if you want to promote your business, offering a 'service' to others in public, be prepared to face the other side as well. Because those of us who think this shit is bogus WILL call it out in this very thread itself.

Yeah, yeah, hard times...need to make money. Why not start a pyramid/ponzi scheme as well? Fits the logic.

 

[rsaeon]

You accused me of 'absurdly insane' logic when I said I preferred multiple 4TB drives over a single 16TB because I can't see myself redownloading 16TB of data over the course of six months in case that 16TB drive fails. Sure, let's call being pragmatic as 'absurdly insane'. There are harsh real-world realities of dealing with a large drive failure that diminish any bragging rights that comes with a Exos as D:\

You really want to spend four to six months redownloading 16TB of media? How is that not a fantasy? We could all be dead from this pandemic and your sitcom would still be waiting for someone to seed the season finale. For me, life and time is way too precious to redownload 16TB of anything. So I'll stick with multiple 8TB or 4TB drives until I'm able afford to at least two 16TB drives for a mirrored array. But yes, this logic is Absurdly Insane To Prove A Point™.

Also, your statements have become increasingly combative where you haven't produced a single reference so that others may trust that your opinion has some basis in the real world. Instead you counter with isolated anecdotal experiences and dismissive denial, generalizing with no focus.

You're way too upset at a single 16TB being out of stock, maybe you need a reality check?

And buddy, if you want to promote your business, offering a 'service' to others in public, be prepared to face the other side as well.

How are you desi and not a fan of samosas? Also Rs 75/plot makes no sense from a business perspective when the mean price is quite a few times more. I'd have to make 2700 plots at this price just go to break even on one plotting machine. There is no profit here. And if this is really a profitable business, why I am sharing all my findings here and in messages? I don't gain anything by offering advice to others to make their own plots, or is there some kind of invisible money that I'm not aware of? Ha, crypto joke.

The truth, as I've mentioned before, is that I missed out on BTC and I do not wish to make the same mistake, when I feel like there's something promising about Chia. Do you not care about other likeminded people that they may also benefit with some good news that you are optimisitic about? Or are you the kind of person that keeps a good deal to themselves?

Because those of us who think this shit is bogus WILL call it out in this very thread itself.

Have you heard of 'speak good or remain silent'? You haven't anything constructive or encouraging to add and you have absolutely no interest in Chia farming, I would think you'd have better things to do with your time, even during a curfew and lockdown.

Yeah, yeah, hard times...need to make money. Why not start a pyramid/ponzi scheme as well? Fits the logic.

This is the classic example of someone criticizing something they don't understand, so I'd suggest you have a samosa and cool off somewhere.

<Mod edit> Please avoid personal attacks

I wish you well.

 

[NotMyRealName]

Right, so now personal attacks because the truth is not to your convenience, and I'm the one who's combative? Very mature.

You're the one who brought out increasingly absurd topics like number of drives etc. All I said is that chia mining has severely affect storage worldwide and in India. You spent multiple posts arguing that India was not affected. When that was proved wrong, you went about drive sizes. Look back through the thread. It's all there (unless you've edited that too).

So in spite of all the evidence out there, I still have to now provide a goddamn bibliography so that you can then call it cherry picking? Everything's clearly out there for anyone interested in the truth.

Others have spoken about how chia is wasteful and a bubble just waiting to burst. I wasn't interested in debating that till you made it an issue. All i said is this latest fad is causing storage shortage just like the gpu one. Sadly, people like you will go to any lengths to get their way. And prove their point.

Have you heard of good men keeping silent and the result of that? It's now our moral duty to speak out against yet another wasteful and destructive fad like chia, which hopefully can be stopped unlike BTC which is too late. This is the most constuctive thing I or anyone else can do; warn people about the impending (and current) disaster that is chia.

 

[tech.monk]

not targeting anyone but the entire discussion is rant.

Either we should arrange the popcorns so others can enjoy or @Mods @puns @Crazy_Eddy @vyral_143 help here.

I also understand that the situation is becoming worst for end users/gamers and those who have money (be it scalper/miner/businesses) are making money still in this situation.

 

[rsaeon]

@NotMyRealName I'll disagree. But let's go have some samosas, I feel like we've bonded somehow through this. This isn't sarcasm. Having a nemesis is healthy, we could take turns being the bad guy. I can't find the thums up emoji, I'm tired so just imagine I put it here at the end of the sentence: X

not targeting anyone but the entire discussion is rant.

It's alright, there's nothing left to discuss.

 

[NotMyRealName]

I already said nothing personal. I don't even know you. No hard feelings. I just feel as strongly about chia affecting storage as others do about gpus that they started a whole thread about it.

And nemesis? please. I couldn't nemesis a kitten!

wish you well too !

 

[Stronk]

I couldn't nemesis a kitten!

But trust me, kittens can nemesis you quite quickly

 

[NotMyRealName]

But trust me, kittens can nemesis you quite quickly

They do secretly control teh interwebs. And most of the world...

 

[booo]

Lets do a thought experiment...
suppose, you have few plots. (in normal terms... files)
instead of storing them on the hdds locally, put them into cloud (say aws) as objects instead. pay the cloud provider money for the storage used....
now, when you get your money back from chia, how much money one can make after taking out the expenses paid for aws, and local cpu usage (power+cooling) plus the hardware(the beefy mining computer) ????