Advanced Setups only 250tb+



  • Hi my name is Chuck
    I've been burst mining for over a year and am now doing a petabyte setup. Looking to establish some people here to talk about some advanced problems and setup.

    Current setup
    1 petabyte
    4 burst miners with:
    i5 kaby lake
    16gb ddr4
    gtx 1070 for plotting and jminer
    Usb pcie cards with a total of 8 dedicated usb controllers to handle 250tb per burst miner
    Each 250tb miner will have it's own wallet

    1 gtx 1070 can plot 2x 8tb seagate expansion drives per day. I run two instances of the plotter for better speeds. If i run dual plots with one instance i get lower speeds. I hope there isn't a problem doing it this way

    Device settings for gtx 1070
    0 0 8192 512 128
    (see anything wrong with these numbers)
    22000 nounces per drive with 2 drives plotting at the same time. Buffer of course.

    7 hard drives per 1 usb 3.0 hub. Each hub with it's own dedicated usb host controller
    (i know fewer are recommend per hub but its already wired)

    To make my life easier. I use starting nounces as 100000000 for drive 1. 200000000 for drive 2. Are those larger starting nounces going to be problem? and or does it matter if it's perfect. So example. For easy numbers. lets say starting is nounce 0. We write a 10 nounce chunk. Does the next start nounce have to be right in order @ 10 nounces-20 nounces. Then 20-30 etc. I did my 50tb server with just using like 2000000 nounces. Then 3000000 nounces and so on. Does that make sense?

    Any pointers are greatly appreciated. Will add more and some pictures here soon.



  • @ChuckNorris ,

    The nonces do not have to be sequential from drive to drive or file to file. My scheme for instance on my #32 drive I start at 3200000000 (drive # plus 8 zeros). I do 1TB files 3815000 nonces each. I do plot sequentially on each drive because I have a spreadsheet that makes it easy for reproduce.

    File 1 start 3300000000 -n 3815000
    File 2 start 3303815000 -n 3815000
    File 3 start 3307630000 -n 3815000
    etc.

    But this would work equally as well:

    File 1 start 3300000000 -n 3815000
    File 2 start 3304000000 -n 3815000
    File 3 start 3308000000 -n 3815000
    etc.



  • plot direct because optimized?



  • @rds Yeah. Thanks for the confirmation. I figured as much. Just making sure it. Never could get a straight answer on that one. When your setting up 100+ drives. So much easier to just say. Drive 35 starts with 3500000000 etc.

    HiDevin. I don't write direct because it seems to be 4 times slower. I've tried many different ways with many different cards. I end up with 6000 nounces per drive or so. I think my best bet is to buffer them all. And then start mining. Then optimize over the course of a month or so.



  • @ChuckNorris Plotting direct to those SMR drives would indeed be painfully slow and also put the drives through hell. To give some insight as to why, the SMR stands for shingled magnetic recording, which describes how data is 'shingled' on top of other data. The write head is much larger than the read head on the drive, which allows for greater capacity, but it also means that for any random write the drive has to first read the data it's about to overwrite, store it somewhere else, write the payload data, then retrieve and rewrite the data that it just overwrote. However, this only applies to random writes. Buffered plots seem to be written in large sequential writes so that avoids the read, write, write, read, rewrite scenario, but since the plots were not written in an optimal way for scanning, it then takes longer for the drive to scan them.



  • @sevencardz That's very good information Seven. I Knew SMR drives are slower. Obviously as they are cheaper. I never considered how they were doing it. I know these Seagate hdd's are crap. But for the price. Can't be beat. And this is all about them burstcoins. Not drive quality. It's good to know tho why a writing direct is sooo much slower. I always tried to figure that out. Thanks for the help



  • @ChuckNorris , I have 62 Seagate SMRs they are all performing well and you cannot beat the price. I have found to overcome slow direct plotting see this thread I started.

    https://forums.burst-team.us/topic/5307/how-to-attain-max-nonce-min-plotting-direct-to-smr-drives



  • @rds That's very good information. That's also very similar to the results i got when trying the same tricks. However its still faster to buffer. Here's why. my 1070 is allowing me to plot 2 drives at the same time with 24000 nounces each. I'm able to fill up 2 entire 8tb drives in less than 22 hours on average. Using this method and two machines. I will be able to fill up half a petabyte in under 17 days. Once that is finished i will then begin mining. While i'm making that bank roll i will then optimize the drives over the course of a good month or so while it's mining away. That's how i did my 50tb server a year ago. Using your method i wouldn't be able to mine for almost 2 months. We are talking about a loss of over 6k in profits from the lost time. And yeah. I know it's possible to mine and plot at the same time. However i just want to get this thing plotted right. And online.



  • It is also possible to use more than one GPU in the same machine, and they do not have to be crossfired. I can't remember my plot times, but I think I plotted 10x 8TB drives in 2 days or something like that. With many GPUs, your plotting bottleneck will probably become the host controller and bus speed.


Log in to reply
 

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