Need to bounce ideas off people for my petabyte setup.



  • ![Just wired up my first half petabyte. Setting up identical unit offsite somewhere else.

    Looking for some ideas
    How i'm doing it
    A single i5 16gb ram gtx 1070 handles 250tb each.
    I'll have 4 units mining to separate wallets.

    I think that if you have one machine count half or a whole petabyte it would take too long and you risk loosing rewards from the extended count time? Is that correct?
    Any tips or pointers for just a large setup.

    Currently plotting 2 8tb drives at the same time per machine. Seems to be the best way to take advantage of the gpu and slower disk speed. Average speed is 22000 nounces each drive. using 1 gtx1070 to 2 8tb drives at the same time.
    So if my math is correct. With 4 machines i should be able to plot all drives in under 20 days. Then as i'm mining i'll optimize them to get the count times down.

    Any tips or tricks for this large scale is greatly appreciated.
    P.S. i'm not using drive letters. Also can't figure out how to upload image here. I get a serer error



  • As far as plotting goes if you want to get mining as soon as possible. You can plot in smaller 1TB increments. |That way as each plot is finished the miner auto picks up the plot and starts using it.

    Now for GPU plotting you can use direct mode which should produce optimized plots so you dont have to do anything after you finished writing.



  • @ChuckNorris Why are you even bothering with all of this?

    Chuck Norris' mining capacity can not be expressed in rational numbers.

    Chuck Norris wins every round, so he typically refrains from mining out of respect for the network.

    Chuck Norris can simply roundhouse kick the blockchain and make it rain cryptocoins.

    Chuck Norris' assets never lose value out of fear.

    Are these statements not true? Have the laws of memes changed? ;)

    You should be able to get 22000 nonces/minute plotting to both drives (11000 each) in direct mode with GPU plot generator.

    As long as you have a good enough GPU for your rig, you'll probably run out of PCI-E lanes before the GPU becomes the bottleneck for scan times. If you do all 8TB drives, then that's about 30 drives per machine so you'd need a few add-on cards and/or USB 3.0 hubs for each machine as well. I have 20 drives connected to a miner and I'm really out of practical expansion options at this point unless I were to upgrade the system significantly, at which point it would probably make more sense to build a second machine and just keep the first running.



  • @sevencardz

    Both a hilarious response and helpful. I would love to pay you or anyone some burst to help me tighten up some ideas.
    Quick history on me. have been burstmining 50tb for a year @ 30 seconds with optimized internal and some external drives. I'm now doing 500tb. 250tb per machine. It's built. Just want to get some things straight before filling it all up. Kind of want to do it right first time around.

    This is how i'm doing it so far. This is for 1 machine handling 250tb each
    -5 usb 3.0 hubs.
    -Each usb hub has it's own usb line going to it's own usb port on the back of the machine. (I believe tho that the onboard usb controller treats all usb ports as one. Because it's only one controller)
    -Each usb hub has 7 8tb external usb 3.0 drives attached.
    -I'm currently working with jminer and getting a read speed of around 200mb. When i first start it sometimes tho it's like 800mb for a while then drops.
    -Would 1 usb 3.0 onboard controller be fast enough to run jminer to count 250tb of drives in a reasonable time.

    Problems. I built these sweet mini itx towers for burstminers thinking i would just need a video card (gtx 1070) So i have only 1 PCIE slot.
    -Should i give up my mini machine dream and go with a full atx board for more pcie slots for usb multichannel usb 3.0 cards?

    -Does jminer run in parralell. So example. I'm getting 200mb with 2 drives plotted so far. As i add more drives would that number increase?
    Also does optimized drives show higher average read numbers. So like 250mb instead?

    Paying bounties on these questions.
    Answer in number format for each question will be paid 1000 burst per answer
    Example.

    1. Chuck Norris was born in the log cabin that he built
      Answer: Fact

    1.Does jminer run in parralell and the more drives you add the higher your read speed would be? so my 2 drives are 200mb. Would 10 drives be like 400+ or something? i'll find out as i plot more obviously.

    1. Will a i5 16gb ram gtx1070 be enough to count 250tb in a reasonable time on the onboard usb 3.0 contorller
    2. If i write direct the % stops at like 1% or so. Is that normal and if i wait it out until its done. It does finish? I get 22000 nounces on each drive when i plot 2 at a time with buffer. So 24 hours for for 2 8tb drives. Someone said writing direct would be about 11000 nounces doing 2 drives at once. So my direct time would be 48 hours instead? I would probably write direct if that's true to avoid optimizing later.
    3. What's the fastest way to fill up 250tb. The way i'm doing it? 2 8tb drives @ 22000 nounces each. If i try to run 3 instances it drops way down below 10000 nounces. So best so far is 2 8tb drives per day. Unoptimized tho.

    Each question is worth 1000 burst each. reply with your burst address. If the questions are even more helpful that i anticipate i will give bonus burstcoins.



  • @ChuckNorris

    1. Yes, not higher read speed but a higher combined read speed.

    2. RAM not a big factor, the better the GPU (more cores) the faster the read. It depends on the USB 3.0 whether each port is independent or share bandwidth.

    3. Yes, yes it will finish but takes a way longer than buffered, buffered -> optmizied is faster than direct. Also you can mine the drives while optmizing.

    4. Max out the RAM on the plotting PC for best performance, you could also plot to a multi drive RAID 0 SSD's, then copy/optmize the plot to the new drive.



  • @ChuckNorris I too have the mini-ITX bug, but I would say yeah, ditch the small board in favor of full ATX for expansion slots if you're building a dedicated miner. What you could do is keep one mini-ITX system and have it do nothing but plotting. Run graphics from the iGPU and use the 1070 for plotting.

    Yes, usually motherboards have only one or two on-board USB 3.0 controllers, so many of the ports will actually share the same controller on the board. Sometimes the front-panel USB 3.0 ports use a separate controller, so keep that in mind. You're right to be concerned about running into bandwidth problems with the multiple USB 3.0 hubs. My general rule of thumb is no more than four drives per USB 3.0 controller to maintain max throughput:

    A single 7200 RPM HDD averages 80 MB/s - 160 MB/s for read/write speeds.
    USB3.0 maximum read speed is 640 MB/s (5 Gbps).

    However, @Propagandalf has more experience with this, so take a look at his calculations for a better answer: https://forums.burst-team.us/topic/5409/how-many-externals-to-a-usb-hub/4

    I always write in direct mode for just that reason - no need to optimize later. Even if it takes twice as long, that's one large tedious step you don't have to bother with. Plotting two 7200RPM 8TB drives with my 1070 takes about 2.5 days. As you fill the drives up, they will start to slow down as the write heads near the edge of the disk (and have to spin more to write the same amount of data). Sometimes I get up to 25,000 nonces/minute, but that goes down to about 19,000 by the time the last plots finish.

    But let's take a step back here - what kind of 8TB drives are you plotting? If you're using those Seagate Expansion or Backup externals, then we may need an entirely different plan because creating optimized plots on those kinds of drives will take forever and put the drives through hell needlessly.



  • @ChuckNorris said in Need to bounce ideas off people for my petabyte setup.:

    1. If, for some reason, your plots were created in buffer mode (unoptimized) with a low stagger, and you are also running more than one Jminer session because you have two or more mining accounts tied to your plots, RAM can become an issue. It did with me, I ended up using insane amounts during a mining round. However, after great advice from @luxe, I installed the o & o clever cache program that did an excellent job of managing my memory cache, and so the problem was fixed.

    I still run into I/O errors occasionally on my one rig, because of two Jminer sessions I think. I need to order one more rig so that I can evenly distribute each batch of HDDs and each mining account onto one rig. I will have 3x 250 TB and 3 accounts in case you wondered, currently my biggest batch has 200 TB and reads in 90 seconds, but I know I can improve this speed.

    I use a 2-core 3.3 ghz pentium on the rig that has a single account and 250 TB (mostly unoptimized), and it barely gets the job done. I would have preferred to have a better CPU for the task, but perhaps optimized plots go easier on system resources.

    I considered mini-ITX as well, but chose not to because of only one PCIe slot. Instead I went for one that had space for one GPU and two multi controller cards. You need at least PCIe version 2.0 x4 in order to run one of those cards, and many mobos have PCIe version 3.0 x1, which is too slow. The mobo I settled for had x8 lane speed across three out of four PCIe slots, which was perfect. The only drawback is that the GPU might not be able to pump out enough juice when it is not allocated x16, but I don't know enough about PCIe distribution to know if this is the case or not. Ideally I wanted to find a microATX with four PCIe slots with sufficient lane speed, because I want to use three multi controller cards, but two will do.

    My Burst dedicated mining rig, with comments:
    ASRock Z270M Extreme4 (microATX is the most smug form factor for this purpose, I reckon)
    16 GB RAM (I could use less I suppose)
    120 GB SSD disk (I will only use SSD for OP, HDDs are too slow)
    2-core 3.3 ghz pentium (I should go for something more powerful)
    EVGA Supernova 650 G2 PSU (I think I could go lower, I measured 220 W from the wall during a mining round)
    USB Wi-Fi dongle
    HDMI dummy (for remote access comfort using teamviewer)
    Sapphire rx 470 nitro+ 4 GB GPU
    o & o clever cache program
    win 10



  • All great answers and very helpful. The first person that answered the questions i sent 3500.
    Even tho he was first. Other posts were more detailed and even more helpful. I sent 2000 burst to each person. Thanks guys..

    I have 68 total 8tb seagate expansion drives. ( i know they are crap. but cheap! )
    1 machine will handle 34 drives each.

    Follow up questions with more burst up for grabs. 250 burst each

    1. Best device info config for gtx 1070? I'm using 8192 512 128. Only gets me 8000 nounces writing direct dual or single. 22k in buffer dual plotting 2 drives at once.
    2. I already have the two burst miners built. I have one pcie slot. I just got into jminer. Can jminer also use cpu? So can i when i'm done remove gtx 1070 and install dedicated usb 3.0 host card with independant channels and use jminer cpu to count everything. It's an kabylake i5. Is gpu jminer that much faster? I'm ok with a 90-120 second 250tb plot read if i can use i5 with usb expansion card.
    3. Whats the read time you really have to stay under for good results? block time is like 4 min. Is there a rule of thumb for like 90 seconds being the max or something for good results?
    4. I bought a gtx 1070 for each miner thinking i would do jminer gpu or zcash mining at the end. Should i send it back and just use one of my rx470 nitro+ 8gb for plotting? Is the rx470 just as fast as 1070 for plotting or close?

    Please don't make me disassemble and send a lot of parts back. lol. I have 4 burstminers built already. Would really just like to plot with gpu. take it out at the end. put in usb expansion host card. Use i5 to jminer or other fast cpu miner to count plots. As always. Thanks for the help guys. Looking to have knowledgeable person on retainer that i'm willing to pay in any crypto for advice.. Already have a petabyte of those crappy seagate drives. And time is money with these things of course. So paying someone a little means nothing to me. And i love helping the community. Been mining for 4 years. Have a eth hash rate of 1000mh at this time. Just really love burstcoin. Been mining it for over a year. but 50tb server is no problem. half a peta... well. Should have looked into this stuff first.



  • @ChuckNorris looks like you are creating a really big farm! lol
    Below are some advices although i never really tried to plot such a big amount neither I have a good GPU to properly test the GPU plotter, so I hope my answers are helpfull for you... xP
    1 - Jminer run in parallel yes... Although I am not quite sure about this: "and the more drives you add the higher your read speed would be? so my 2 drives are 200mb. Would 10 drives be like 400+ or something?", because the way Jminer reads the plots is the order you have in the properties file, so let me try to explain it better:

    • Properties file:
      Path: D:/plots, E:/plots,......

    Jminer will dedicate one thread to read the first plot existing in D:/Plots, then dedicate another thread to read the second plot of D:/plots (if exists... if not goes to the first plot of E:/plots), and so on...

    And since each optimized plot will only need 1/4096 nonces of it to be readed we can say that each drive of 8 Tb will only need 2 Gb of data to be readed from each drive (8 Tb / 4096 = 0.001953125 Tb * 1024 = 2 Gb)...

    So to answer the question I would say that your reading speed in mb/s will just increase wile you have less plots than GPU threads, once you have more plots the reading speeds will start to decrease...

    2 - I would say that it can be enough taking into account more controllers, with just one USB controller I don't think it's feasible to attach 250Tb tbh... I would advise another PC, with the same GPU but with a i7, and a board where you can attach a lot of USB controllers and don't have to share USB bandwidth that you need to be used to read the plots the fast as possible...

    3 - I believe its normal and the reason for this i think it has to do with the GPU plotter doesn't refresh the cmd window very often... Regarding the 48 hours I believe it is yes. I would advise the contrary if you have such a big amount to plot... I believe you would be maxing you hardware more if you plotted in buffer mode, then optimized it with your CPU... Since you can mine the plots before they are plotted, you will be mining more capacity even of lower, and with time your plots will be all optimized... Just remember to save the space to optimize all the plots ;D

    4 - I believe its the way you are doing right now but with the optimizer running with the CPU... Although I would recommend a new ATX mobo that allows more controllers ( I would not recommend to connect more than 3-4 external hdds to each controller 3.0 and 8 hdds to a controller 3.1), also would recommend an i7 or somethingwith the most threads and Frequency as possible so you can optimize the faster as possible too and have your farm the most optimized as possible in the shortest time...

    Good luck with the farm and let me know if I can help you with something ;D



  • Killer response gpedro. Gotta at least tip you 1k burst.

    1. I think you hit a point i was going to ask. Can i plot using my gpu and also move/optimize a different plot with my cpu at the same time?! so gpu plot to drive 1. drive 2 is already done plotting. move plot from drive 2 to drive 3 with optimizer using cpu while drive 1 is getting plotted?!

    2. is gpu jminer really needed for 250tb per rig. Or would an i7 count fast enough with optimized plots?
      if that's the case. What's the best/fastest cpu counter? sorry. i call it drive counting as to not confuse people.

    3. In xplotter and jminer. my cpu doesn't shop up as a device. if i use 1 to show devices under setup. It shows just my gpu. why doesn't the onboard cpu show up. On my other rig it does. Is it because of kaby lake or z200 series chipset? or is bc the motherboard over rides and gives priority to pcie lane with gpu and shuts off cpu graphics? change onboard bios settings to multimonitor or something? or install intel hd graphics driver? even in my device manager no other display adapters listed other than gtx 1070. Just noticed all of this.



  • @ChuckNorris Thanks... my address is on the signature ;P
    1 - Right on the spot
    2 - I believe you will really need jminer to read 250Tb, but the 1070 you have I think it should be enough but provably another GPU even if lower grade would be good for you to be able to mine a little better while you are still plotting... Regarding the CPU I just suggested for you to optimize it faster, you will not be using much from your CPU, unless you also make your CPU plot with the XPlotter wich already give you optimized plots... hehehe
    3 - I cannot help much with that since I don't have much of experience with the GPU plotter but I would say that is a drivers issue, so I would reinstall the drivers again



  • Recaps and a new direction thanks to you guys. let me know if this all sounds right. Send all burst as tips already.

    Step 1. Probably take back 1070's and just use the rx470's i already have and have plotted with before.
    Step 2. Dual plot buffer to 2 8tb drives. 24 hours for both drives.
    Step 3. Change over gpu plotter to next drive and start plotting same way
    Step 4. Setup and use optimizer to move and optimize plot while it's gpu plotting ( can i optimize two drives at once? since i plot 2 drives per 24 hours. Can i also optimize 2 drives in 24 hours or so.

    Minor questions.

    1. If i'm optimizing a drive that's on the same usb hub and or same usb controller. Does that kill my nounces for plotting?
    2. i know it's better to have 3-4 drives per hub. I def don't want to change that. it's already perma wired. Will i still get acceptable results with 7 drives per hub. with a total of 5 hubs each on their own usb 3.0 channel.
    3. Would a rx470 8gb have similar results plotting as 1070. i think the 20k nounces per drive is the bottlekneck of the drive? And or how about jminer. Does jminer gpu like 1070's more than rx470's? i know stream processors and cuda cores are diff. Which does it prefer.
    4. Is there any way an i5 or i7 using jminer cpu only would be fast enough to count 250tb?
      Heres the card i'll probably get.

    https://www.sonnetstore.com/products/allegro-pro-usb3



  • @ChuckNorris

    Step 1: I think AMD GPUs are the ones that work best with the GPU plotter, but I cannot confirm this. With MSI rx 470 8GB Gaming X I was able to get a consistent 40-45k nonces per minute in buffer mode. You can plot multiple drives at once, but then you must divide your total stagger by the amount of drives you plot.

    Step 2: I just saw from what you wrote that you already knew about dual plotting. ;)

    Step 3: You could experiment with plotting more than two drives at once. Sooner or later you will find the sweet spot where your GPU(s) cannot handle more nonces per minute, or that they will produce less nonces per minute because you are bottlenecked by your USB 3 host controller(s). I tried using multiple GPUs and multiple multi-controller cards, but this can lead to I/O errors.

    Step 4: The optimization process uses a little bit CPU, but mostly it uses the read/write of your HDDs I think (and it puts load on your host controller(s) again). I don't know how long it takes to optimize.

    Q1: Yes it can, either due to GPU resources or host controller resources.

    Q2: Yes you will likely get speeds you are satisfied with. You should look at additional price costs of building additional rigs for additional host controllers versus how much you think you are losing out by not having superfast read speeds.

    Q3: See answer to step 1. I believe I used the settings 0 1 64 64 8192 for optimal speed and stability, but it's long ago so I can't remember.

    Q4: I really doubt the openCL GPU that is built in to those cards will be able to handle 250 TB, but you could always try! Please let me know what you find out.



  • @ChuckNorris Steps 1 - 3 sound good. I'm not sure on step 4 though - I wouldn't start optimizing plots until you have the first two drives completely done in buffer mode. Then, I would take those two drives and optimize their plots onto another pair of drives and start writing buffered plots to two other drives. Ideally, use one machine to optimize and one machine to plot to keep all the tasks completely separate. To speed it up, try writing buffered plots to four drives at once rather than just two. GPU Plot Generator can use more than one GPU at once, so you could run both rx470s for example. I've never actually tried that though. :) I'm also not sure on the optimizing time or if it can be run in parallel. I assume with separate instances it would be.

    I'd also recommend doing a full format on the drives before you plot them and use a 64K allocation unit size. Helps to weed out any drives that might be lemons so you can send them back within the return period.

    Q1. Optimizing a drive while plotting a drive on the same controller - this might work but I wouldn't try it. As stated above, I'd recommend separate machines. If the plotter goes down for some reason, it'd be ideal for that to not negatively impact optimizing.

    Q2. Your read times will likely start to suffer and so you wouldn't really be maximizing your investment. The difference between 60 seconds and 30 seconds doesn't sound like a lot, but it adds up to a lot over years and years of mining rounds.

    Another card to consider:

    https://www.newegg.com/Product/Product.aspx?Item=N82E16812400546



  • I forgot to mention something: Consider plotting each batch of 250 TB to a different mining account if you are going to use one rig per batch. I have done this, and that enables you to mine on several pools so that you become more flexible, you take less risk as regards a pool going down, you support decentralization and you can more easily compare pool payouts over time.


Log in to reply
 

Looks like your connection to Burst - Efficient HDD Mining was lost, please wait while we try to reconnect.