plot_optimizer_1.6-j6jq



  • @FrilledShark I appreciate the response, I am not sure if I understand what Optimization does so I wanted to validate my assumptions.

    If the Plot is written in Direct Mode it Pads the Drive first then at the end it adds the Nonces. So this would mean that the Direct Mode Plot is Optimized as far as the File is concerned.

    If the Plot is written in Buffered Mode it requires the Plot to be Optimized for faster mining. This would be the process of Copying the data from the Buffered Plot to another File to act as it is now a Direct Plot.

    So does this Optimization also write the Nonces First and then Write the Padding? If the optimizer copies the Nonces First they will be written to the Faster part of the Hard Drive and speed up Mining on top of being Optimized.

    Or does the Optimization Program copy the Padding first and then Nonces?

    If not, could there be a way to Write Direct with Nonces placed at the start of the writing? This would give an even further edge and may be a boost for 5400 RPM drives too.

    The Direct mode GPUPlotGenerator could also send the Nonces First natively and no Optimization would be needed at all if the Optimizer does in fact write the Nonces First when it does its work.

    I am just basically trying to see if the data that is accessed during mining is on the fast part of the drive. This will become more important on bigger drives. They slow down quite a bit at the center of the drive.



  • @CryptoNick This post from an earlier Optimisation thread has a great graphic of an optimised plot. Not optimised on the left, optimised on the right.

    @FrilledShark said in Plots 101:

    Do I need to optimize my plots?
    I'm gonna start by saying, that optimizing does NOT give you better deadlines, but only makes you read your plots faster.
    While it is not required, it is recommended by almost everyone. Optimization reduces disk usage, by arranging nonces, which belongs in the same scoop, together in a single group. (See picture) This reduces seek time, which in turns reduces the total read time.

    @Blago's plot experiment
    17930413153828766298_69900000_1000000_4000 - not optimized, reading time 3818 ms
    17930413153828766298_64900000_1000000_1000000 - optimized, reading time 2143 ms

    This means you able be in on fast rounds, where you would otherwise have lost a block. Example: If you found a 30 seconds deadline after 7 seconds, you would be able to beat someone needing 35 seconds to find a 5 second deadline. (As explained by @luxe)

    0_1466538372870_1466538041209-dyhoi_82pdo.jpg Credit to @Blago for picture and technical information

    Optimization isn't needed on SSD, while it still works, the benefit is minimal.



  • @RichBC Excellent! Thanks for that info @RichBC That makes sense. So now we need to move all the Nonces Optimized to the start of the file so they are on the Fast portion of the platter. I would assume that it starts writing the Nonces first and their Scoop first.



  • @CryptoNick Should have linked to the next post in the old thread as well, explains things a bit more.

    @FrilledShark said in Plots 101:

    How does optimization work?
    Optimizing plots allows your miner to read your plots faster and with less work. Optimizing works by changing the structure of the plot, so that the plots's nonces are in the scoop they belong to. (See image above) This allows the miner to read 1 scoop at a time, without needing to do any seeking for nonces. (See image below) In a standart plot, the miner needs to find each nonce belonging to the scoop it is currently reading. This is ineffective, because the harddrive needs to do a lot more of seeking to read 1 scoop. In an optimized plot the miner doesn't need to search for the next nonce, because they are right next to each other.
    0_1466540836025_ip.bitcointalk.org.jpeg Credit to @Blago for picture and technical information



  • @CryptoNick There is 4095 scoops in a plots. There is no way, you can pick some scoops and prefer them over other scoops, as all scoops will be read every 4095 blocks.

    The difference between direct mode and buffer, is that direct prepares the plot before plotting and buffer create the plot and writes from the computer memory. The plot file will end up in the same place either way.



  • @CryptoNick Think of it this way, searching an unoptimized plot is like searching an unordered list, you have to check each and every one while searching an optimized plot is like searching a sorted list, you only have check a few as they are in sequence.



  • @iKnow0 Ok so I misunderstood the Direct VS Buffer. So 1 Scoop has to be written every 4095 blocks. The only thing optimization does is order those and then only has to read a smaller portion of the same part of the drive no matter where it may be, Inner Platter or Outer Platter. They will just be in order and must be written first and then optimized after writing. Optimizing a Direct Plot is beneficial too in this case. I get it now. Thanks everyone!



  • @CryptoNick
    No, as your direct plot was prepared in the correct order. It will also be written in the correct one.



  • @FrilledShark So Direct Plotting is already Optimized? I am creating a Direct Plot right now and I filled 15TB and it stayed at 1% Nonces for a day, then the last 500GB it started writing Nonces constantly updating the percentage 5% 10% slowly building the last of the file. This was the portion I thought would speed the read up since the other portion couldn't have Nonces in it but may be the scoops though. The Nonces are being created right now in the last process. This was the portion I had hoped to move to the Outer part of the Platter.

    So if it is now an optimized file, can we put the last part on the start of the writing, or it still doesn't help? I am going to make bigger plots too this is just a test.



  • @CryptoNick

    When using direct plotting, the plotter is preparing the place to put the plots. It is not calculation the data before it has prepared the whole plot. That is why it takes a long time, before it can begin plotting.

    You don't have to worry about fast or slow part of the harddrive, as the nonces will span across the whole plot file.

    Also, the nonces are ordered into scoops. The scoops are not additional space, but instead groups of nonces.

    If this still doesn't make sense, send me a PM and we can discuss it via chat.



  • @FrilledShark Is this why my GPU plotter has sat at 400nonce/minute, so incredibly slow, because its filling it in? I thought GPU plotting was supposed to be very quick. My ETA on 15TB is nearly 3 bloody weeks. WTF. My CPU plots at 8000nonce/min. My GPU is a HD6950 so it should be half decent right? definitely at least faster than 400nonce/min



  • @tylerk07
    It depends on which mode you are plotting in. Direct needs around 50% of the time to prepare the plot. Buffer does it without having to prepare it. Direct produces optimised plots, so it is preferred to use it over buffer.



  • @FrilledShark I guess ill just let it run but 400nonce/min just seems way too low


  • admin

    @tylerk07 it'll increase once it gets past the initial file building. Mine was at 100, is currently just passed 15,000 and will end up around 33,000.



  • @haitch Oh excellent! At least I'm not the only one!



  • @haitch Its been running now since you set it up so more than 24 Hours and its on 7.37%. At what stage does the initial file building stop?


  • admin

    @tylerk07 If it's at 7% then it's past that and is filling in the drive.



  • @haitch Oh so I should expect 400nonce/min for the remainder of the time?


  • admin

    @tylerk07 No that should keep increasing, but your drives we slow when I set it up the first time, so don't know how high it will get.



  • @haitch Thank you :)


Log in to reply
 

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