Shingled magnetic recording (SMR)


  • admin

    Watch and learn about SMR ... e.g. Seagate Archive and partly Desktop drives, and why are they quite cheap.

    Watch from min. 3 to understand why using e.g. direct mode of GPU Plotter on SMR is a bad idea.



  • Do you think that drives that are 6 TB or less are non-SMR?



  • @luxe Very interesting video. I wonder if these drives have / could have a mode to allow you to sequentially write a plot to a disc such that the overlapping part of the track was always on the currently unwritten part of the disc then Read Modify Write would not be needed and the speed of writing would be ok?

    Rich


  • admin

    @RichBC Write the plot to a non SMR drive, then copy it to the the SMR drive - it's what I'm doing on the drives I have



  • @haitch @luxe @Propagandalf @RichBC I use strictly SMR Drives and they write at 200MB/s and they read @ 300MB/s. To be more specific they are the 8TB drives that where highlighted. The only reason that those speculated readings are so low is because they are not sequential reads. For everyday use or for use in a server yes it's a bad idea but for mining burst or plotting for that matter, a pretty good cpu is the bottleneck. Just my experience.


  • admin

    @AngryChicken The issue is specific to GPU plotting in direct mode - it's close to a random writing process - the plot file is created, then the plotter jumps all around the file filling it in. In CPU mining or buffered GPU mining it's a sequential write. However for GPU plotting, direct is preferred as it creates and optimized plot.



  • I think i have 2 of these 8tb drives and I keep having to replot the plot files after a day of using them. If you see any of my post about corrupt file complaints then it these.drives. Matter fact they both were replaced 4 times this year while under warranty.0_1478664263369_IMG_20161108_230230324.jpg



  • @Burstde Are you saying that your mining on these drives was what broke them, so that you needed to get them replaced?



  • @haitch said in Shingled magnetic recording (SMR):

    @AngryChicken The issue is specific to GPU plotting in direct mode - it's close to a random writing process - the plot file is created, then the plotter jumps all around the file filling it in. In CPU mining or buffered GPU mining it's a sequential write. However for GPU plotting, direct is preferred as it creates and optimized plot.

    So if you were to create a plot on a "standard" drive and then run the optimiser writing to the SMR Drive would that also random write, or does the optimiser do things in a way which enables a sequential write?

    Rich


  • admin

    @RichBC optimizer is a single continuous write



  • @haitch Thanks, that's the way I tend to optimise anyway, so no change needed. Only problem is that the 8TB Archive drives seem to be more expensive £/TB than other conventional 5TB drives you can get in the UK.....

    Rich



  • @Propagandalf I bought these two drives only for burstcoin plots. So yes, theses drive get hot . My last pair (4th replacement) run with a usb fan pointed to them and are 2 months old which is a record. Unfortunately I bought these drives new then every replacement was a re-certified one. One replacement was dead the minute I plugged it in. I more than likely blame seagate and not burst for the drive errors. These drives were bought new in April and may. Then each replacement takes 3 to 4 weeks to receive.



  • So I'm going to necro this topic because I'm lazy.

    It seems that Seagate uses the SMR archive drives only in the 8TB Seagate Expansion model (not sure about the 'Backup' models). The 4TB and 5TB models actually have desktop-level drives in them. I am guessing that's why they jump from 5TB to 8TB with nothing in between. I had no issues writing in direct mode to four 4TB expansion drives, but everything came to a halt when I tried the same thing on the 8TB models, as you guys have noted.

    I'm starting to think that Seagate is throwing refurbished drives in these Expansion drives or something. I was able to get over 100MB/s writes on one of the 8TB models (when it was fresh), but not the other. I am guessing @AngryChicken was lucky enough to get fresh ones? I just got two 'brand new' 8TB models and I am now copying over plot files to each of them at less than 30MB/s, which makes no sense because that's just a huge sequential write operation and all I did was a quick format before that. They should be flying along at 100MB/s or better...



  • @sevencardz Well legally they would have to state that the drives are refurbished. But would you mind giving me some more info on how your rig is setup?

    I can say that if you are using the usb 3.0 that comes with the external drives, this may be why your transfer/write speeds are so slow. I personally pulled the HDD's out of the external casings that converts sata into usb 3.0, and plugged them into my motherboard directly. (This improves speed a bit)

    You should also check your motherboard manual, because the bandwidth could be shared among other ports ether usb or direct sata.

    Lastly I would check CPU usage as if your cpu is maxed out from other tasks it may, queue data until it can process it.

    My Drives straight out of the box write @ 220MB/s (Peak) and 98MB/s when finishing writing the final plots. The read speed of the drives differs as I don't bother optimizing my plots. (35 sec for 30TB is good enough)



  • @AngryChicken Good point, I guess these drives just suck in general when it comes to write speed and I'll just have to live with it.

    CPU/RAM: AMD FX-8350, 16GB RAM
    MOBO: ASUS SABERTOOTH 990FX R2.0
    GPU: EVGA GTX 950

    USB3.0 can theoretically handle 640MB/s and that's twenty times faster than what these drives are currently writing at in my system. Shucking them and hooking them up internally via SATA is a good idea, but it sort of defeats the purpose for which I bought them, plus I've already filled up my internal SATA ports and drive bays. I have two of these 8TB Seagate Expansion drives and each is connected to a dedicated USB3.0 port on the mobo, which is an ASUS SABERTOOTH 990FX R2.0. I also have four 4TB Seagate Expansion drives hooked up to a Rosewill RC-508 4 Port Single Bus, so that card is actually sharing its USB3.0 bandwidth between all four drives and read speeds are well over 100MB/s on each of those, so the USB3.0 bus being the issue seems very unlikely.

    Granted, the system is actively mining 50TB with jminer and I do see the write speeds drop to about 20MB/s during a mining round, which makes sense because I am copying plots from drives that are actively being mined. The write speed picks back up to 30MB/s on each drive after the round is over, but it never goes higher. Between each round, the only other software running is the local wallet, so I can't think of any reason for the slowness besides SMR.



  • You are correct with seagate using refurb drives. The two 8tb internal drives I bought new have been replaced each 4x with refurbs under warranty. this last time everything has been solid. Every replacement was refurbished with 3 going back doa or dead after format.
    I bought a seagate ext 8tb plus and that has been rock solid too. That being said they are slower than a wd 2tb internal drive which I've a found a few blocks with yet most are with seagates. All together 26tb @ 30mb/s usb connection to 4core laptop probably look at improving read times next.

    I believe the plots start to degrade on the 8tb drives after 2 months



  • @sevencardz if you get card like this you can add more direct sata ports to your computer https://www.amazon.ca/IOCrest-Controller-Profile-Bracket-Components/dp/B00ESFEI2E/ref=sr_1_5?ie=UTF8&qid=1485375953&sr=8-5&keywords=sata+controller

    Also it's not how much each usb 3.0 port can handle it's how much the bus can. For example if you have two usb ports that use the same bus then the bandwidth will be that of a single usb 3.0 instead of two, this can worse with more and more usb 3 on the same bus.

    EDIT: so i checked you motherboard model and couldn't find details on the usb bus, but if you look right next to your ram slot there is a usb 3.0 header/input slot that you can try to see if this improves your speed.
    A General rule that I follow is if the usb ports are stacked they use the same bus.



  • would someone please try Xplotter on SMR's now already!-) Or do I really have to replott one of mine? I think it could be working great! please just tell me how it behaves! ;-)



  • @AngryChicken Yes that's true, I could buy all sorts of expensive hardware but none of that would solve the problem that these drives just have terrible write speeds. Maybe the first time you write to them they're fine, or maybe some are actually refurbs, or maybe Seagate puts desktop drives in some and archive drives in others. I don't know, but what I do know is the USB3.0 ports on the back of the mobo should not be sharing a single bus on a $180 premium board and even if they were, two drives running at even 200MB/s would never saturate that 640MB/s bus. That Rosewill card I posted runs four drives from a single USB3.0 bus and none of those drives ever had these ridiculous slow downs with reads or writes. I already connected the USB3.0 header to my front panel I/O and connecting the drives there makes no difference for write speeds.

    @nixxda I've never used XPlotter. :) I tried the GPU plotter on these 8TB SMR drives and write speeds were even worse than just doing a straight copy. I'm talking single digit MB/s... no joke.



  • I'll throw some more numbers at you guys, since someone will probably find this useful at some point:

    With GPU plotter in direct mode, I can write a 500GB plot file to two 8TB Seagate NAS drives in 3.5 hours. That's somewhere around 20,000 nonces/minute. The initial plot file structure writes at nearly 200MB/s, and then that drops down to about 120MB/s while filling in the plots with nonces. So to plot all 16TB, it takes about 2.5 days. Given a generous estimate that the SMR drives can push 10MB/s writing plots in direct mode, it would take them over NINE days to write the same plots. No, I have not fully tested that, but I did start the tests and saw the write speeds in win10's resource monitor first-hand. So even if it takes 3 days to plot to the NAS drives and 3 days to write those plots back to the archive drives, that's still 3 days faster and only half of the 6 days it takes is actually spent doing intensive plotting.