• Register
  • Help
Results 1 to 7 of 7

Topic: GS + RAID question

Share/Bookmark
  1. #1

    GS + RAID question

    Hi,

    I\'m in the process of buiding a giga rig for live performance and need help with a couple of questions.

    I am trying to decide whether to use either 2 x 36GB raptors in a RAID configuration OR 1 x 74GB raptor.

    Both drives would be for samples and a seperate HD for O/S

    The problem is, the 36 GB raptors have a slower seek time and are reportedly noisier, the 74GB raptor has a better seek time but I\'d have the trouble of it trying to stream a lot of data off of the one disk.

    Anyone thoughts or comments would really be appreciated.

    Ben H

  2. #2

    Re: GS + RAID question

    Thanks Lee,

    So you don\'t think RAID will make much of a difference to GS? I\'d be hoping to stream around 4-6GB in real-time.

    Do you know why the experts recommend against using RAID for GS? Truespec recommend RAID for some reason. Can you please point me to any articles where I can read up on this?


    Thanks again,

    Ben H

  3. #3

    Re: GS + RAID question

    Originally posted by Ben H:
    Thanks Lee,

    So you don\'t think RAID will make much of a difference to GS? I\'d be hoping to stream around 4-6GB in real-time.

    Do you know why the experts recommend against using RAID for GS? Truespec recommend RAID for some reason. Can you please point me to any articles where I can read up on this?
    <font size=\"2\" face=\"Verdana, Arial\">4-6GB per sec? What exactly are you streaming, the Library of Congress? Seriously, do you mean the total size of all samples is 4-6GB, or do you mean 4-6 MB per sec?

    The read throughput for a single modern ATA100/133 120MB drive is about 40MB per sec. Of course, that\'s sequential read, and a real-world scenario might bring that down to, say, 25-30MB/s on average. Lot\'s of factors there, such as size of disk, where on the disk you\'re reading from, etc.

    SATA is definately faster as a matter of shear technological fact, but unless you\'re bumping up against that 25-30MB barrier it may not be worth the extra noise and cost. Real-world tests from Giga and other streaming users will conclude that a single SATA drive setup is definately faster, I won\'t dispute that. But I think you need to consider your real-world requirements.

    Theoretically, you can stream 50MB/s by using 2 X the 25MB/s ATA drive mentioned above, AND get the benefit of pushing bits over two controller interfaces instead of one.

    This is where the RAID question comes in. A RAID 0 config gives you better aggregate performance, but you\'re forced to treat both drives as one logical unit. Contrast that with the above scenario of asking two independent drives to do two separate tasks. I don\'t know if anyone anywhere has published RAID vs. multi-drive config tests. My guess is that nobody has enough CPU to push that amount of data through a sampler to see where the top end of each config bottoms out. In the least, by keeping the drives separate you have a more flexible setup because two separate apps won\'t have to share the same logical device. (That\'s why I disbale RAID on my onboard Promise controller -- so I can stream samples/tracks from one drive while writing to another drive).

    My above argument against RAID holds equally for SATA as it does ATA.

    - Keith

  4. #4

    Re: GS + RAID question

    [craig] Ben, this is a post from Samples. don\'t want to pollute the Samples forum with HW stuff.

    >>What if I decide to dedicate one drive to GS and the other to VSTi\'s? Is RAID the better configuration option.
    [craig] Still not sure why you want RAID (I assume you are refering to RAID 0 which is all about speed). If you Streaming large quantities of contiguous data in one OS call i.e. one access, RAID 0 would be the way to go. Thing is, with GS and other disk samplers, you are only filling up relatively small pieces of RAM. Lee seems to be running fine with his RAID so maybe it won\'t hurt you (though their are disk failure risks) but many people have tried that config and haven\'t seen the payoff- so it seems. When Lee gets 3.0, maybe he can fill us in on his poly count. On the other hand I believe Dave G. said he was getting 320 poly running GS 3.0 on standard 7200 RPM drives. Not bad.

    >>Could I stream GS off of one and have the VSTis triggered from the other without the system falling over on me? Remember I\'d be hoping to play in Real-Time, not sequencing.
    [craig] Yes, put GS samples on one drive - the raptor. I would assume with the Raptor, you are going to have a lot of extra headroom available. I used to run GS96 on a P3 866MHz with one 7200 RPM drive and a sequencer and had no trouble consistently getting 96 voices. It doesn\'t take much to get good performance.


    Oh, and why do Truespec recommend RAID for their VSL GIGA systems?
    [craig] Ask Peter. He\'s always available for a chat or e-mail.

  5. #5

    Re: GS + RAID question

    Originally posted by kbaccki:
    4-6GB per sec? What exactly are you streaming, the Library of Congress? Seriously, do you mean the total size of all samples is 4-6GB, or do you mean 4-6 MB per sec?
    <font size=\"2\" face=\"Verdana, Arial\">I thought that you need to be able to stream the equivalent of RAM off of the HD when disk streaming, am I wrong?

    Typically, each of my instruments are 4L chromatically sampled and about 7ish sec per note. That means that the instruments work out to be about 154 MB each.

    If I have a stack of 6 x 4L instruments eg for a brass stack then theoretically that is 924 MB. Plus I would also hope to have other instruments loaded so I can switch between them.

    Am I way off?

    Ben H

  6. #6

    Re: GS + RAID question

    Originally posted by Ben H:
    </font><blockquote><font size=\"1\" face=\"Verdana, Arial\">quote:</font><hr /><font size=\"2\" face=\"Verdana, Arial\">Originally posted by kbaccki:
    4-6GB per sec? What exactly are you streaming, the Library of Congress? Seriously, do you mean the total size of all samples is 4-6GB, or do you mean 4-6 MB per sec?
    <font size=\"2\" face=\"Verdana, Arial\">I thought that you need to be able to stream the equivalent of RAM off of the HD when disk streaming, am I wrong?

    Typically, each of my instruments are 4L chromatically sampled and about 7ish sec per note. That means that the instruments work out to be about 154 MB each.

    If I have a stack of 6 x 4L instruments eg for a brass stack then theoretically that is 924 MB. Plus I would also hope to have other instruments loaded so I can switch between them.

    Am I way off?

    Ben H
    </font><hr /></blockquote><font size=\"2\" face=\"Verdana, Arial\">You\'re talking about the aggregate size of the instrument (total of all sample data) -- that total size is not what you need to consider in determining your streaming throughput requirements.

    What you need to consider is: over all instruments loaded and playing at the same time, how many sampler voices will you be using simultaneously?

    What\'s a sampler voice?

    A sampler voice is a single virtual audio channel that processes audio when a note is pressed (or released, for that matter). That processing is in the form of playing back a sample from disk, synthesizing sound, applying DSP, etc., or any combination of those. A single note-on may invoke several voices -- e.g., layered sounds require multiple simultaneous voices per note. At any given time a synth/sampler will mix the output of all active voices and write that mixed output to, say, a single stereo audio device pair (real or virtual device output).

    As an example, let\'s say I have a sampled piano instrument that uses 1 velocity zone per note for the string sound and 1 velocity zone per note for a hammer-mechanism sound. If I play 1 note on the piano I hear the string being struck and the hammer clunking away at the same time -- that\'s 2 voices per note. (For these examples let\'s assume I\'m using mono samples.)

    As another example, let\'s say I have a sampled trumpet that uses 2 velocity zones per note for the horn timbre (a soft timbre for low velocities, and a brighter timbre for higher velocities). The 2 velocity zones are exclusive -- I\'m either playing the high zone or the low zone. In addition I have a 1 velocity zone per note that gives me the sound of the valve being depressed, and that zone covers the full range of velocities. I know this is contrived, stay with me... If I play 1 low velocity note I will use 2 voices (soft timbre+valve), and if I play a high velocity note I will also use 2 voices (bright timbre+valve).

    Of course, voices come and go -- in the piano example if I hold the note down I might hear the string sound for a full 10 seconds, but the hammer sound is finished after 1 second. So I\'ll being using 2 voices for only 1 second, then 1 voice thereafter.

    =====

    Now, when you\'re thinking about your particular instruments, the aggregate size of the instrument doesn\'t matter. What does matter is how many simultaneous voices you will be using. Here\'s why: a single mono 48KHz 24-bit streaming sample translates to:

    48000 sample/sec x 3 byte/sample = 144,000 bytes/second

    It doesn\'t matter whether that sample is being played from a 100MB instrument or a 1GB instrument -- a 48K 24-bit sample requires the same throughput regardless of how long the sample is. If a sampler voice is streaming that sample off of disk, then 144K/sec is what you need to be able to pull off of the disk. For stereo samples you need to double that throughput.

    Finally, let\'s say you break down your instrument requirements, look at voice-per-note counts for all instruments, and you figure that at peak times you will need 200 simultaneous voices (on average), and you know that all voices are stereo and all samples are 48KHz @ 24-bit. At that point figuring the required throughput is simple multiplication:

    200 voice x 2 channel/voice x 48000 sample/sec x 3 byte/sample = 57,600,000 byte/sec = 55MB/sec

    That would be worst-case, and is a simplification of the real-world. Keep in mind that samplers will avoid pure disk I/O on playback by buffering the first N seconds of all samples. I forget what N is for giga. I know you can set N in HALion (e.g., I use .75 sec buffers).

    With that in mind you can ask those that really know your specific libraries what that 55MB/sec really translates to in real-world usage. Even if you required 55MB/sec, I think you could get that with 2 x ATA and most definately with 2 x SATA, 2 x SCSI, SATA RAID 0, and SCSI RAID 0. Whether or not you could get that with a single SATA or SCSI drive, I\'d say probably. The experts should have those numbers

    - Keith

  7. #7

    Re: GS + RAID question

    Thanks Keith, that spells out what I wanted to know.

    And thanks everyone else for also being so helpful.

    Ben H

Go Back to forum

Tags for this Thread

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •