One SSD to load them all - Good idea?

Discussion in 'PC/Mac Builders, Mods, Peripherals - New' started by MoeWalsaad, Dec 7, 2018 at 2:24 AM.

Tags:
  1. MoeWalsaad

    MoeWalsaad Senior Member

    50
    7
    Apr 10, 2015
    Hello,
    I want to improve my Sample library loading times and performance, I brought one new large SSD,
    So I wonder whether it's a good idea to run all my main VST libraries on it, or is it better to split between several drives (includes HDDs)?

    I know that loading too many instruments from one HDD is not a good idea, so how about SSD's reading mechanism, will it handle the load?

    Thanks in advance!
     
    Last edited: Dec 7, 2018 at 3:00 AM
  2. chocobitz825

    chocobitz825 Senior Member

    171
    165
    Aug 7, 2018
    I can only speak to my experience that a RAID SSD drive over thunderbolt 3 has generally improved my read speeds compared to HDD.
     
    MoeWalsaad likes this.
  3. JohnG

    JohnG Senior Member

    7,373
    3,468
    Nov 13, 2007
    California
    If you own it, go ahead and use it. I usually buy multiple, smaller drives, but it doesn't seem to make all that much difference with SSDs.

    The only reason to RAID an SSD is to create a larger volume so you don't have to split large libraries. Otherwise when you update it can be a pain.
     
    MoeWalsaad and MartinH. like this.
  4. MartinH.

    MartinH. Senior Member

    326
    254
    Jun 16, 2018
    I would like to do that as well, my template takes like 5 minutes or so to load. I'm 99% sure that the number of SSD drives that you have doesn't matter as long as you have SSDs instead of HDDs. What else is there to try in Kontakt settings? Or Reaper's settings?
     
    MoeWalsaad likes this.
  5. ceemusic

    ceemusic Senior Member

    386
    71
    Dec 14, 2008
    New York area
    Also I'd go with two 500GB SSDs than a single 1TB SSD.
    2 drives sustaining ~540MB/Sec instead of a single sustaining ~540MB/Sec.
     
    MoeWalsaad likes this.
  6. Damarus

    Damarus Composer n stuff

    100
    35
    Nov 15, 2017
    Actually, a larger SSD is better than multiple. I don't see any reason why you would split it up over multiple drives. Larger SSD's have more memory onboard, making them slightly faster and more efficient.

    Again, unless you're setting up a RAID config I don't see any reason to split up drives for the same purpose. There's no speed benefit.
     
    MoeWalsaad likes this.
  7. ceemusic

    ceemusic Senior Member

    386
    71
    Dec 14, 2008
    New York area
    2 because for streaming samples it yields better performance.
     
    MoeWalsaad likes this.
  8. Damarus

    Damarus Composer n stuff

    100
    35
    Nov 15, 2017
    Why would 2 separate drives yield better performance than 1?
     
  9. ceemusic

    ceemusic Senior Member

    386
    71
    Dec 14, 2008
    New York area
  10. Damarus

    Damarus Composer n stuff

    100
    35
    Nov 15, 2017
    Yeah, I read it but it doesn't necessarily work like that... One drive would be saturated over the other drive.

    Price per GB is cheaper on Larger SSD's. Larger SSD's are faster and have a longer lifespan. Fewer parts, easier organization. SSD's start to slow down the more they fill up.
     
  11. JohnG

    JohnG Senior Member

    7,373
    3,468
    Nov 13, 2007
    California
    That is wrong. You can fill SSDs almost to 100% and they don't slow down to any meaningful extent.

    If you want faster load times, consider PCIe-based storage. That can make a difference.
     
    MoeWalsaad likes this.
  12. chocobitz825

    chocobitz825 Senior Member

    171
    165
    Aug 7, 2018
    this conversation never seems to resolve well as most of it comes down to differences in opinion about what is significant enough gain to warrant the cost and risk of ssds. Obviously PCIe based storage is ideal if speed is a concern, but the cost is not great. one thunderbolt 3 ssd drive is probably enough for most people, but as an anecdotal reference, I and many others I know who have designated RAID library drives have found the speed increase significant enough to put it above HDD and single SSDs. There are a number of factors and real world results may vary, but my read speeds increased enough for SSD, and raid to be justified.
     
    Last edited: Dec 7, 2018 at 9:21 AM
    MoeWalsaad likes this.
  13. JohnG

    JohnG Senior Member

    7,373
    3,468
    Nov 13, 2007
    California
    I have never seen any data that supports the claim that RAID makes a meaningful difference to music composers / those using samples. This is true for performance or load times for SSDs or for HDDs, now that we're talking about it.

    Do you have data that does support that? Not just opinion / guess / hypothesis?

    Using RAID does raise the likelihood of a failure, so for that reason, I don't do it.

    Certainly, the PCIe drives are very expensive, but who cares? They last a long time and they actually DO make a difference in performance that is noticeable, even with very demanding string libraries.

    The cost / performance tradeoff is subjective. Whether or not there is a meaningful improvement -- greater than, say, 10% -- is less so.
     
    MoeWalsaad and MartinH. like this.
  14. jbuhler

    jbuhler Senior Member

    361
    392
    Jun 19, 2016
    US
    So how many simultaneous samples can one effectively run through a 5gbps USB3 port? For most of us, that would seem to be the most relevant bottleneck for SSDs and whether you need to divide your samples across SSDs (and direct USB3 ports).
     
  15. chocobitz825

    chocobitz825 Senior Member

    171
    165
    Aug 7, 2018
    I'll just provide the results of quick test on a few drives I have. (all are in Akitio Thunderbolt enclosures)


    Thunderbolt 2 SSD RAID: 636 MB/s Write 745MB/s Read (2 drives)
    Thunderbolt 3 SSD RAID: 416 write 1294 Read (4 drives)
    Thunderbolt 3 SSD RAID (with cheaper drives): 416 write 1013 read (4 Drives and thunderbolt 3 enclosures are daisy chained)


    i don't have any more HDDs on hand to test, but in general I get sufficient read speeds from the drives. I have backups of all the libraries so loss form the RAID is not a concern, and since they're only designated for reading libraries, and not recording or anything, Ive not had a drive fail on me over the last 3~4 years. Again, just anecdotal, but in my case, they are faster than the HDD alternative, and perform faster for me on my mac than single SSDs did. no doubt i would get much more from PCIe, I just haven't seen the need. The performance difference is of course significant, but with how SSD costs went down, this was a sufficient transition for the time being.

    EDIT: just a reference Internal iMac Pro Speed: 2931 Write 2487 Read
     
  16. Damarus

    Damarus Composer n stuff

    100
    35
    Nov 15, 2017
    Fair enough, if you have the enclosure and the drives laying around.

    for OP, just get one large m.2 PCIe SSD. They are cheap right now
     
  17. chocobitz825

    chocobitz825 Senior Member

    171
    165
    Aug 7, 2018
    I am curious, what are some of the PCie prices and solutions for people who need to use them as external storage? would you connect them via USB 3.0/3.1 or thunderbolt?
     
  18. MartinH.

    MartinH. Senior Member

    326
    254
    Jun 16, 2018
    There's an option to do the sample loading in the background, I'll try that. I don't need everything to play fine as soon as I open my template, I'd be happy if I could start doing midi editing sooner and would be fine with waiting a couple minutes till all the samples are loaded for a full playback.


    Did you benchmark sequential read performance of one large data-chunk? Because that data would be meaningless for the sample library usecase. It's been too many years since I touched a proper disk benchmark tool, but iirc it's called something like "queue depth" which tests parallel reads of multiple files. On my old HDD the sequential read was around 100MB/s and the value for a lot of small reads in random locations (e.g. like playing back many different sample library samples) went down to the single digit range. Doing the same benchmark on an SSD you see barely any difference between the two values, because SSDs don't have significant seek time for drive head repositioning, because they have no mechanical drive heads...

    On HDD drives I would actually expect sample library streaming performance to go DOWN from a raid 0 setup, because it should effectively double the amount of times the drive head needs to reposition because every sample is distributed over 2 drives. You might see a minor performance increase over 1 single HDD drive, but I would estimate 2 non-raid drives to perform better if you split your libraries accross both drives. And for SSDs my prediction is that it doesn't matter because you won't be bottlenecked by the drives under normal circumstances.

    One second of 16 bit 48khz stereo audio uncompressed should be a little less than 200kb, on any regular SSD you should easily get 300 MB/s, that's roughly 1500 parallel sample voices being played. Probably more in reality because they use lossless compression, shifting i/o bandwidth load to CPU decompression load. Sure, with multiple mic positions and lots of backed in reverb you might get there, but I really don't think you need a raid 0 setup for SSDs. I'll happily stand corrected if you can provide some hard data that proves me wrong.

    Side note: if anyone is using SSDs in external enclosures I think it's worth doing a disk benchmark that tests the parallel random reads because with USB 2 iirc there was a considerable bandwidth loss for certain types of operations that came from the "overhead" of shoving the instructions over USB 2. Wouldn't be entirely surprised if there's some performance degradation on USB 3, even if the theretical maximum bandwidth couldn't get saturated by the drives max read/write bandwidth. Just a hunch though, might be fine after all. I only use internal SATA connectors.
     
    MoeWalsaad likes this.
  19. chocobitz825

    chocobitz825 Senior Member

    171
    165
    Aug 7, 2018
    Sure thing. pulled out the various library, backup, archive and data drives I could find.

    Samsung T5 1TB via USB-C:
    Seq Read 40MB/s Write 21.11MB/s
    4KQD32 Read 11.25 MB/s Write 9.37 MB/s

    WDC HDD RAID 1 via Thunderbolt 2:
    Seq Read 140MB/s Write 97.75MB/s
    4KQD32 Read 0.811 MB/s Write 1.862 MB/s

    2TB SSD 2 Drive RAID via Thunderbolt 2:
    Seq Read 682 MB/s Write 446MB/s
    4KQD32 Read 92.74 MB/s Write 30 MB/s

    8TB SSD (4 Drives) RAID #1 via Thunderbolt 3:
    Seq Read 1443 MB/s Write 343MB/s
    4KQD32 Read 96 MB/s Write 76 MB/s

    8TB SSD RAID #2 via Thunderbolt 3:
    Seq Read 1470 MB/s Write 355 MB/s
    4KQD32 Read 94.13 MB/s Write 68 MB/s

    Apple 1TB PCie SSD (2 drive RAID by default if i recall)
    Seq Read 2725 MB/s Write 2765 MB/s
    4KQD32 Read 237 MB/s Write 160 MB/s
     
    MartinH. likes this.
  20. MartinH.

    MartinH. Senior Member

    326
    254
    Jun 16, 2018

    Awesome! Thanks a lot for testing, that's some data that we can use! That 4KQD32 line is the one I'd focus on, because it should be closer to how a drive behaves when streaming lots of samples from it, than durint the sequential read benchmark. Sequential read/write would be interesting for backups or first time install speed of libraries.


    The first drive is only getting USB 2.0 speed, might want to look into that if it's supposed to be 3.0. Some mainboards have both USB 2 and 3 connectors on board (mine does). Might be plugged into the wrong port maybe? It's also a good example of what I said about the USB 2 overhead having big impact on bandwidth, because there's no way you're bottlenecked by the SSD itself down to 10 MB/s.

    HDD raid gives unusable performance (less than 1MB/s read), just as I expected. I wouldn't be surprised if the Raid is actually detrimental here. On my smallest HDD I get 1.79 MB/s read an 1.97 MB/s write.

    All the thunderbolt raids give usable speeds that should cause you no problems, even though they are roundabout 10 times slower than sequential read/write. I don't know anything about thunderbolt, so I can't say what's causing this. I would have expected higher speeds on regular internal SATA connected non-raid SSD drives.


    Here are speeds for my SSD drives:

    Samsung 850 Evo 500 GB:
    4KQD32 Read 245 MB/s Write 214 MB/s

    M4-CT128M4SSD2 128 GB:
    4KQD32 Read 190 MB/s Write 149 MB/s

    I'm not sure both are connected to SATA II ports, the Samsung one is, but the other one might not be. But it's a very old drive, I doubt it would make a big difference.

    Personally the conclusion that I would draw from all these benchmarks is, that everything other than a normal internal SSD connected via SATA is either detrimental or doesn't see the real-world performance increase that a sequential read/write benchmark (like they use in all the advertisement material) would promise.


    P.S.: In my old rig I used to run a "SATA port multiplier" that split one SATA port to several SATA ports that shared the bandwidth of one. I haven't followed the developments of those, but if someone is keen on getting more SSDs than available SATA connectors into one case, it might be worth looking into.
     

Share This Page