Vienna Symphonic Library Forum
Forum Statistics

200,608 users have contributed to 43,204 threads and 259,106 posts.

In the past 24 hours, we have 4 new thread(s), 9 new post(s) and 72 new user(s).

  • RAID flavours

    Hi

    I appeal to any geeks, and in particular tecnicians at VSL to help me, please.

    I am using RAID arrays to both speed up and secure my Vienna Library. All this on a Mac. (Please no platform flaming here). I'm not planning to go SSD as yet - although my question would still apply in that case too.

    Now, for some time I have been formatting and managing my RAIDS using the highly respected SoftRAID 4.0 utility and drivers. The great thing about SR is that it has the capability to READ from a mirror, across all members of the RAID at once. Unlike the Apple software RAID, that will only read from the primary disk of a Mirror RAID. The advantage of this should be clear: you get Stripe equivalent read speeds, but Mirror security. In some situations the read will outperform a stripe. SR has many other advantages I won't go into here (email notification of drive failures is one example).

    However, in some real world tests, I have found that a SR formatted mirror RAID with 3 members is taking longer to load up a large project than a sinlge no raided disk. WTF I'm thinking.

    After speaking with SoftRaid technical support, it appears that under normal circumstances, the SR driver is multithreaded, and transparent to the OS. So a multithreaded application should in theory benefit from the multi-disk mirror read strategy of SR. But, there are situatiions were the SR driver will only use 1 thread (and heance 1 disk) and a notable example of this is disk speed test utilities, as they operate on a deeper level thatn the OS in order to control their test conditions (thereby bypassing SR in effect). No multithreaded disk speed test applications seem to exist at this time for OSX.

    My questions is, does VEPro use OS level disk reading, multithreaded, or does it use deeper methods to read, possibly inadvertently bypassing all the advantages of a multidisk mirror read? Does VEPro load "everything at once" or does it plod through instrument by instrument?

    And, does VEPRo do writing to disk as part of the load process? If so, this would possibly explain the anomaly - as obviously a mirror can only ever write to disk at the same speed as the slowest drive in the array.

    I would especially welcome some input on this form VSL technicians, although any real world experiences really count for the most useful knowledge. And does any one else out there use SoftRAID 4?

    Many thanks

    Ben


  • Hi Ben,

    I have used SR 3, when analysing the trafic with BigTop I have found curious behaviors. So I trash it !

    Can you run BigTop with SR 4 to see if it is still happening and send me the disk I/O diagram at cb dot pub at numericable dot com

    YOU MUST use last version of VI PRO and VE PRO so I can compare (VSL has divide the loading time by 40 with my raid of SSD)

    FYI : You can use a Raid 0 for your lib, and back them up on a single disk

             more you have disk in a raid 0 faster it is

    Best

    Cyril


  • Hi Cyril

    Sure I'll try Big Top - although I have never heard of it until now. I am in geeky mood so I'd like to try.

    By the way, SoftRAID is verison 4 now - which is multithreaded and has other advances included SSD support.

    I'll get back to you. What was the curious behaviour you had? Just so I know what to look for..

    Thanks


  • Can't find Big Top. Do you have a link to it?

    B


  • To have it you need to install the developer package on main OS X DVD

    Set the analysis time to what needed. to have the graphic of the disk load during loading the VE pro server 64 bit

    The behavior is difficult to explain with my knowledge of English

    Best

    Cyril


  • Ok I will do that. I'm on a dealine for some days so it may have to wait...

    In the meantime, if I observe Activity Monitor, I see that Disk Activity only exhibits a maximum of about 25MB/s. This seems really strange! I wonder if VEPro is being held up looking at the licence dongle or something.

    B


  • last edited
    last edited

    @Another User said:

    In the meantime, if I observe Activity Monitor, I see that Disk Activity only exhibits a maximum of about 25MB/s. This seems really strange! I wonder if VEPro is being held up looking at the licence dongle or something.

    Is this with the latest version of VI PRO and VE PRO server 64 bit 7846 ?


  • Yes - 7846, 64bit, OSX 10.6.6, latest eLiscence software.

    I think it's time to raise a support ticket.

    B


  • 1. VI (and VI embedded in VEPro) uses single threaded loading.

    2. VIPro uses multithreaded background loading. The number of loading threads are the same as the number of cores available in the system.

    3. We do not write anything to the disk during loading.

    Thus, if Softraid works as you say, you should experience a higher loading speed with VIPro. 25MB/sec sounds like a normal figure with a single, non-raided modern (spinning wheel) drive. The figure you want to look at regarding disk performance with our instruments, is the "64k Random read" value.

    Thanks,


  • Thanks MS. Really useful information. I am building a strategy based on your advice now.

    The key lesson here is seek times are king.

    B


  • Did you made the Diagram with BigTop


  • I will. as I say I have some deadlines.

    However, I've gone ahead with SSD disks already. Totally different speeds. The information from Martin, however, implies that it doesn't really matter what the throughput speed of a drive is - seek times are what holds up the load. Thousands upon thousands of 64k files - perfect for SSD!

    I will get around to testing though - it may be in a couple of weeks though.

    Beet

    Ben


  • You just need to load Bigtop set it to see 600 or more secs and you load your orchestral template