Vienna Symphonic Library Forum
Forum Statistics

181,658 users have contributed to 42,178 threads and 254,544 posts.

In the past 24 hours, we have 1 new thread(s), 2 new post(s) and 49 new user(s).

  • MIR: Loading samples too slow?

    Hi,

    is it possible, that MIR has a problem with data-loading? Until yesterday I thought this is because I had a Software-RAID, which had bad performance at all, but today I don't have RAID anymore and the problems persist:

    1. Loading Big Projects after Startup
    Windows Performance Monitor shows, that MIR has a loading-data-rate of 4MB/s whereby the HD is capable of 80-100MB/s. That means: MIR slowes down the HD by factor 20-25.

    Result: Loading of a complete Orchestra takes up to 15 Minutes, which is quite long :-/

    2. Disk-Streaming
    It seems that also when streaming samples from disk at playback the datarate is also limited 4MB/s at max. Problem with that: I get audio-glitches when playing a song for the first time and a new Instrument-Section enters the Tutti. When playing same song a second time, it gets better and third time playing its OK.

    My Conclusion: Reading-Datarate of MIR is too slow for Sample-Streaming. The "Disk"-Streaming only works if the stuff which has to be streamed is in Windows-HD-Cache and has not to be read from Disk.

    Is it possible, that the Disk-Storage-Containers have a performance-problem? Does the fileformat need to many seeks on disk?

    Is it possible, to optimize loading-performance?

    Thanks and Best Regards,
                      Tobias

    System: Win7 64 Prof, Core i7, 24GB RAM, dedicated 1TB HD for VSL-Samples


  • Hi Tobias,

    thanks for your detailed report. Our developers have already looked into it, but they couldn't reproduce the problem. They were able to load with about 160MB/sec from an SSD disk.

    Could you please verify your loading times from another disk - just for testing? - Our developers will look into it closer in any case.

    Kind regards,


    /Dietz - Vienna Symphonic Library
  • Hi Dietz,

    thank for your fast reply. I did some testing, also with changing disk-drives (that took me some time...). I made some logs. My Configuration is now:

    Drive C: dedicated 1TB HD
    Drive D: dedicated 1TB HD
    Drive E: dedicated 1TB HD

    1. Copying a File from D to E

    see screenshot of Win-Ressource-Monitor in http://dl.dropbox.com/u/3600096/picture_read_big_file.png

    Data-Rate is expected at 80MB.

    2. MIR when loading Konzertsaal-Impulse-Responses

    http://dl.dropbox.com/u/3600096/MIR-Loading%20Impulseresponse.png

    3. MIR a moment later, Impulse-Responses are ready, Samples are still be loaded in Background

    http://dl.dropbox.com/u/3600096/MIR%20loading%20samples.png

    What I see in case 3 is (please correct me if I am wrong):

    Sample loading is at all with 10MB Datarate (instead of 80MB). But: There are many files accessed in parallel which means lots of seeking for my harddrive. That could be an explanation why its working with SSD but not with mechanical drives...

    Am I wrong?

    Best Regards,
                Tobias


  • Thanks again for taking the time to test this. As I said before, our developers are looking into it already.

    Kind regards,


    /Dietz - Vienna Symphonic Library
  • Hi,

    I really don't want to push anyone - but I am interested... have you also seen the problem in your labs when loading libraries from "traditional" harddrives?

    Best Regards,

                    Tobias


  • There has been quite some effort form our developers to make loading as fast as possible recently:

    -> [URL]http://community.vsl.co.at/forums/t/28664.aspx[/URL]

    HTH,


    /Dietz - Vienna Symphonic Library
  • last edited
    last edited

    @topse said:

    What I see in case 3 is (please correct me if I am wrong):

    Sample loading is at all with 10MB Datarate (instead of 80MB). But: There are many files accessed in parallel which means lots of seeking for my harddrive. That could be an explanation why its working with SSD but not with mechanical drives...

    Am I wrong?

    You are right, the prehistoric spinning drives with mechanical heads are limited by physics. The constant jumoping of the head while switching between samples to read, causes the read throughput to be substantially lower than during a sequential read operation. A figure of 10MB/sec is not uncommon.

    10ms average seek time means 1s of pure seeking for 100 samples. Each sample uses 64kb precache. Reading 100 samples every second gives an average rate of 6.4MB/sec. Modern drives have some caching which allows the throughput to be slightly higher, but it is still hard to come above 15MB/sec with a "Spinning Jenny". SSD drives have no problem reaching 200MB/sec or above.


  • last edited
    last edited

    @MS said:

    You are right, the prehistoric spinning drives with mechanical heads are limited by physics. The constant jumoping of the head while switching between samples to read, causes the read throughput to be substantially lower than during a sequential read operation. A figure of 10MB/sec is not uncommon.

    10ms average seek time means 1s of pure seeking for 100 samples. Each sample uses 64kb precache. Reading 100 samples every second gives an average rate of 6.4MB/sec. Modern drives have some caching which allows the throughput to be slightly higher, but it is still hard to come above 15MB/sec with a "Spinning Jenny". SSD drives have no problem reaching 200MB/sec or above.

    Which leads to the innevitable question of VSL software soon allowing users to change the precache in anticipation of SSD-based workstations, making it possible to run the orchestra without having to spend silly money on truckloads of RAM? Seek times are easilly 70-100 times faster on SSDs than HDs, IOPS is higher by ridiculous amounts and sustained transfer rates are now exceeding 500Mbps on SATA3, so my logic tells me it should be possible to do away with the requirements of all this expensive RAM we need to run proper sessions and templates. When MIR entered the scene it didn't really become better, since that (fantastic) monster gobbles GBs of RAM before you even load your first sample.

    THAT would be good news to go along with the MIR Pro release...    along with an autosave function. [:D]


  • I already stated in another thread, that the next release of ViPro contains a setting for precache size, all the way down to 4kb/sample. My tests show it can play well even with that size, and a proper SSD setup.


  • last edited
    last edited

    @Vagn Luv said:

    Which leads to the innevitable question of VSL software soon allowing users to change the precache in anticipation of SSD-based workstations, making it possible to run the orchestra without having to spend silly money on truckloads of RAM? Seek times are easilly 70-100 times faster on SSDs than HDs

     

    Unfortunately you'll find that the money is even sillier for buying SSD, when compared with the "prehistoric" ones. [;)]

    DG


  • Its a bit pitty that the requirements for the software (MIR) don't tell, that SSD is in fact essential... Will there also be a switch to increase cache per sample above 64kb?


  • SSD is not essential for MIR. AFAIK samples for MIR itself are loaded into RAM, so all an SSD would do is make loading faster. It's only the instrument samples that are streamed.

    DG