Vienna Symphonic Library Forum
Forum Statistics

190,093 users have contributed to 42,706 threads and 256,980 posts.

In the past 24 hours, we have 4 new thread(s), 38 new post(s) and 38 new user(s).

  • RAID-0 or single drive?

    Hello everyone,

    My hunch is that this topic has already been discussed elsewhere. I couldn't find a satisfying answer yet, hence this post.

    Do the VIs benefit from a RAID-0 setup (say, eSATA), or is it better to run the instruments off a fast single drive?

    I know that streaming speed is greatly improved with a RAID-0 setup, but I'm guessing that for firing samples off a HD, faster seek times are more important, something that RAID controllers are not superior at. Moreover, various sample-library makers recommend a single drive as opposed to a RAID solution.

    Any word from the league of engineers at VSL? Any user with experience on that topic?

    Many thanks,
    Tobias

  • OOps i just posted a similar question as well

  • last edited
    last edited

     Tobias:

    The following is relevant to your question:

    @cm said:

    i can only recommend the usage of sATA disks and we already have a pile of 1 TB seagate and samsung drives in use ... the raw throughput is about 70 MB/s (sustained, not peak) ... as written in earlier posts i don't see a real reason for raiding such disks.

    christian

    ps: another advantageous sideffect: you won't spend CPU for firewire ....

    edit: a quality mark is mentioned capabilty of NCQ (native command queuing) and/or TCQ (tagged command queuing) - either for the controller and for the disks



  • last edited
    last edited

    Interesating...I would  like to find this out, too. I currently have a RAID-0. Is it worth braking it into two separate drives? It would imply some extra work in doing so, but if I got a performance improvement, that would be fine.

    I am on a Q6600 Quad-core, 8GB ram PC running Steinberg V-Stack...

      

    @tobiaswagner_2366 said:

    Hello everyone,


    I know that streaming speed is greatly improved with a RAID-0 setup, but I'm guessing that for firing samples off a HD, faster seek times are more important, something that RAID controllers are not superior at. Moreover, various sample-library makers recommend a single drive as opposed to a RAID solution.

    Tobias

     


  • Yes I would add this question - if RAID 0 is not recommended for eSata configurations using VSL SE, how should you break up the VSL SE files to reside on different disks? How do we determine what is in the SE files since they are named cryptically if say, we wanted all the strings on Drive 1 and the Brass on Drive 2 etc?

  • last edited
    last edited

    I don't think you can spread the VI SE library across multiple drives (but I might be wrong). I do have quite a few libraries (Appasionata, Brass1, Woods-1, SE plus), so I was thinking in spreading them out. So...is RAID-0 rally worse than two drives, so it is just that it is not that much better as one may think it is. In the latter case, I won't bother braking my RAID-0.

     

    @composer22 said:

    Yes I would add this question - if RAID 0 is not recommended for eSata configurations using VSL SE, how should you break up the VSL SE files to reside on different disks? How do we determine what is in the SE files since they are named cryptically if say, we wanted all the strings on Drive 1 and the Brass on Drive 2 etc?


  • last edited
    last edited

    You can, as far as I know, spread a VI library over as many disks as you'd like as long as they are all attached to one computer (assuming that 1. you don't try to split up instruments between drives and 2. your Vienna Key is plugged into a a USB port on that computer.)

    Please also note that CM stated that the speed of Seagate and Samsung 1 TB disks is sufficient, so that a RAID 0 array is not necessary, he did not recommend against using a RAID array for streaming samples. My own experience is this: I have been using a RAID 0 array composed of two Seagate 500GB drives to stream samples for over three years and have encountered NO problems. Furthermore this RAID survived the failure of the power supply of one enclosure so that when the drives were removed from that enclosure and put in a new enclosure the RAID continued to work with no problems.

    The logic of RAID 0 is that its throughput is a direct multiple of the number of drives of which it is composed. Thus if the the throughput of a single drive is 70MB/s then a two drive RAID 0 array would, theoretically, have a throughput of approx. 140MB/s, a three drive array would have a throughput of 210MB/s etc. The reciprocal fact is that the likelihood of failure of a RAID 0 array is a multiple of the number of drives of which it is composed, since the failure of any one drive in the array causes the whole array to fail.AS I have noted earlier, other RAID levels, such as RAID 5 and 6 feature both speed and redundancy - thus reducing the chance of the failure of the entire array while providing faster throughput than the individual drives of which they are composed.

    Once again I strongly suggest visiting www.amug.com where you can find highly detailed tests of drives, SATA cards and external enclosures.

    I am not sure of the source of Tobias Wagner's idea that SATA controllers when used for RAID arrays have a negative effect on seek times. Logically speaking, if the same SATA controller is used for 1 disk or for multiple disks, wouldn't it have the same effect on all disks attached to it? According to the tests I've seen, one can achieve extremely high throughputs 200-1000 MB/s with most modern controllers and a sufficient number of drives. Large arrays (usually RAID 5 or 6), for example, are routinely used for film and video editing - a task which would seem significantly more demanding than streaming audio samples.

    Addendum: a test available at http://www.hyperossystems.co.uk/07042003/HD4tests.htm of WD Raptors vs. Hyperdrive 4 solid state drives (16GB capacity for $2700 without the necessary RAM chips) shows that seek time is degraded by 1/10 or 2/10 milliseconds with the use of a SATA Raid card. Thus, if a single drive had a seek time of 8.3 milleseconds, an array composed of several such drives would have a seek time of 8.4 to 8.5 milliseconds. This difference would seem insignificant in real world applications, especially as i/o throughput would be significantly increased in an array.


  • last edited
    last edited

    @stevesong said:

    You can, as far as I know, spread a VI library over as many disks as you'd like as  long as they are all attached to one computer (assuming that 1. you don't try to split up instruments between drives and 2. your Vienna Key is plugged into a a USB port on that computer.) 

    Thank you. I assume by instrument you mean a flute vs Oboe etc. But how can you tell which files go with which instruments in VSL SE?

    For example

    SpecialEdition045.dat<>
    What goes on in here?

  • Sorry, I don't have the Special Edition - - just the Cube and  the .dat files are organized into folders as you guessed. But, since the SE is only 81GB why would you need to put it on more than one drive or one drive array? 


  • Well, for one, I'd like to purchase more Vienna in the future and also utilize the drives for other legacy .gig samples I have. Also I have Ivory 1.7 on the same drive. Another - thought that I might need it distributed for those bigger tuttis. I guess its not supposed to be used in this way then...thanks

  •  Regarding small file I/O per second, the test I mentioned earlier at http://www.hyperossystems.co.uk/07042003/HD4tests.htm shows the following:

    1 WD Raptor: 1K file = 365 IOPS; 2K file = 350 IOPS; 4K file =  340 IOPS; 8K file = 300 IOPS; 16K file = 320 IOPS; 32K file = 310 IOPS

    2 WD Raptors configured as a RAID 0 array: 1K file = 600 IOPS; 2K file = 570 IOPS; 4K file =  550 IOPS; 8K file = 540 IOPS; 16K file = 530 IOPS; 32K file = 520 IOPS

    8 WD Raptors configured as a RAID 0 array: 1K file = 1890 IOPS; 2K file = 1880 IOPS; 4K file =  1870 IOPS; 8K file = 1860 IOPS; 16K file = 1850 IOPS; 32K file = 1800 IOPS

     These tests were conducted with drives attached to an Areca SATA card.


  • Steve, I appreciate your posts on this subject here and elsewhere.

    I don't own SE, but I think it has only one license. The confusion here is the VI in total can be distributed per purchase (Woodwinds I, Woodwinds, II etc.) but I assume that SE which is only one purchase has only one license, hence only one location.

    Have you ever divided your library per your two drives in a non-RAID configuration to see if performance is better?

    Also, if you were buying new drives at the moment, do you have any preferences? I'm assuming I'm in the ballpark with Seagate 1TB's, the plain vanilla SATA's.

  •  To answer your question: I ran the Horizon series solo strings and various other sample libraries from single drives attached to a Windows machine running Gigastudio and this worked fine. Then I setup an external RAID for my G5 using a Seritek SATA card and having been using it for over 3 years and it works fine. 

    I think that the argument over seek times of single drives vs. RAID may be a bit overdone.

    For example, the published specs of the new 300GB 10,000 RPM Western Digital VelociRaptor drive claim that it has a 4.2 ms average seek (read) time and a 4.7 ms average seek (write) time. Seagate claims that the average seek times of the 1.0TB Seagate Barracuda ES-2 Enterprise-Class SATA II 7200RPM drive (reputedly a very fast drive suitable for streaming samples) are 8.5ms (read) and 9.5 (write).

    Thus the VelociRaptor is, in terms of claimed seek times, more than twice as fast as the Seagate 1.0 TB drive - - - a vastly larger difference than the .2ms difference noted in the tests I referred to between single drive and RAID array seek times. So if you think that you must absolutely have the fastest seek times, you have to go with multiple VelociRaptors (or for more money some 15,000 RPM SAS drives). But, in reality, to stream samples, multiple VelociRaptors are very likely overkill. The folks at VSL claim that a single 1.0 TB Seagate or Samsung drive can be used to stream the whole Cube.However, tests show that a RAID 0 array has faster I/O per second on file sizes from 1K to 32K - as well as on larger size files.

    In addition to the fact that the WD VelociRaptor costs $300 for 300 GB (whereas the Seagate Enterprise class 1.0 TB drive costs $250-300), it has the drawbackof  a non-standard design ((because it is based on a 2.5" drive rather than a 3.5" drive), so it won't connect to normal 3.5" backplanes like those in the MacPro - - or any 3.5" enclosures with drive trays. It will only work in situations where it connects via SATA cables (e.g. in a fixed bay within a PC, a fixed bay external enclosure or, if you are a MacPro user, some non standard, home-made, rig inside a MacPro + cables to the two unused SATA ports on the MacPro motherboard). I wrote to Western Digital asking if they planned to release a version of this drive that would mate with normal 3.5" backplanes and they said they "couldn't comment on non-existent products." Whatever that means! (Perhaps it's simply another triumph of intelligent industrial design to make a super fast drive that will not mate with the standard backplanes most users would want to use it with?)

    Several reviews of new Samsung drive 1.0 TB drive have found that it is currently the fastest 1.0 TB SATA drive. However, if you go to Newegg.com and read the customer reviews of this drive, you find many customers claiming that their drives arrived DOA or died soon after installation. So this might give one pause. I have no idea if these reviews are representative or not, nor do I have any idea of how good Samsung customer support is. In light of this, Seagate, at present, seems, to me, a possibly safer bet. However, you might try calling Other World Computing and talking to them about this as they are knowledable and they have just started selling Samsung drives. I doubt they would sell any product that gets returned frequently. 


  • This is very helpful and confirms my present views.

    "I think that the argument over seek times of single drives vs. RAID may be a bit overdone."

    I agree. If samples all play at the right time, there can't be a "more right" time. They sounded when they should with no latency. That's it. Based on your experience and Christian's observations, we've hit a feasible and relatively inexpensive price point for a whole lot of music to be made, at least as far as drives are concerned.

    "...so it won't connect to normal 3.5" backplanes like those in the MacPro...."

    This is precisely the kind of thing I strive to avoid. Thanks.

    "...many customers claiming that their drives arrived DOA or died soon after installation."

    That's where I went and what I read last night. I buy my drives at newegg. Most of the drives were averaging four stars out of five. But the Samsung user feedback spread seemed a bit aggravated (5 stars 43%, 1 star 29%). That's hardly a scientific study, but it still has psychological impact.