Vienna Symphonic Library Forum
Forum Statistics

196,686 users have contributed to 43,023 threads and 258,420 posts.

In the past 24 hours, we have 2 new thread(s), 4 new post(s) and 94 new user(s).

  • That does help. Thanks. Since I switch around a lot with different templates I"m going to plan on using SSDs just to experience the quick load times.

  • There is a very good new is that VSL is working on improving loading with SSD.

    The result of the actual version are disappointing

    When I add the time of the load of my 8 VE PRO individually (total of 156 000 samples) it takes 4 mn to load, but if I load them all thru a VE PRO 64 bit server, it take up to 30 mn ; it starts with very fast loading, then after a while it is going slow

    The is the diagram of the load of the previous version.

    If I reload the 8 VE PRO after 30 minutes only half of them where loaded

    Best

    Cyril


  • Here is the diagram with new version :

    Just before 17:06:40 there is an incident on the curve, I have open the last VE pro and the 1st VI PRO instrument.

    What is strange is that the I/O went down

    On the diagram you can see that there is a lot of writing to disk, why ? it is supposed to be loading samples into memory !


  • your second diagram strengthens my suspicion this is a memory management issue (of the operating system) ...

    i would try to watch changes to size of memory types and disk-I/O (read/write) during the first 5-6 minutes.

    loading of samples (into RAM) seems to collide with prefetch technology - data/code which is assumed to be used gets loaded into memory beforehand, after system realizes it doesn't need the data it gets moved to the pagefile (which is on disk) to provide more space for samples (which are marked as not to be unloaded into the pagefile, even if not currently needed).

     

    my assumption: in the beginning there is more unused memory available and the read process is fast, after a certain point prefetch data gets unloaded, then loading of sample data decreases until a second border is reached and behind that constant memory re-organization happens.

    it would be interesting to see a second diagram what happens on the system volume (where the pagefile resides) during the process and how amount and relation of different memory types change over the time-axis.


    and remember: only a CRAY can run an endless loop in just three seconds.
  • Hello Christian

    I have send the diagram to Apple support ; they say it is the fault of the programmer !!

    The swapfile is move to a special volume

    My guess : when you load the 1st VSL PRO it is started, so it is using CPU, so if you load another VE PRO it will use more CPU 

    Can you do a littel program that is loading 10 GB in memory.

    If we have the same problem we can report that to Apple because they ask me to do so. But I don't program in C !

    If you want to monitor a load you can connect to my computer with Ichat, my account is "cyrillindablanc"

    We can arrange it now or after 16 PM or tomorrow morning, 9AM ?

    Best

    Cyril


  • thanks for the reply ... meanwhile i've been also on the phone with marnix to find out there is even more in the game ...

    are you using a software raid or some hardware raid controller (if the latter, which model) and is there a special reason you formated the raid-volume intended to be read from as journaled (though this should not affect read-access at all)?

    swap-file on special volume ... does this volume reside on a seperate disk or just on a partition (eg. on the system drive)?

     

    the *CPU usage* you're referring to above relates to *threading* IMO (one process is using only one thread instead of many ... had to get some more background-info to comment on that, but we could think of it as possible reason though file access is a seperate beast).

     

    the *testing program* to load data would also have to mark the loaded data as *not to be paged out* what is an interesting idea but beyond my resources ... the *loading gap* after the first sequence is strange anyway.

     

    thanks for the offer with iChat but i'm working from home this week where i don't have a mac at my hands currently ...

     

    what made me avoid SSDs so far is the fact i noticed strange latency spikes (with read-access) on several models but i'm going to repeat my tests with an intel X25-M postville generation-2 next week ... what also looks interesting is the OWC SandForce series ... we had to find out its caching behaviour though ...


    and remember: only a CRAY can run an endless loop in just three seconds.
  • Hello Christan

    The faulty is either MAC OS X or either VSL software, it will be great if you could write this little programm that loads 10 ot 15 GB in memory so we see how MAC OS X reacts

    The raid card is a Rocketraid 640 sold by Apple in the US and my SSD are from Crucial, 

    If I use my ssd raid for other task and it is performing very fast ! not at 3.30 MB/s

    The graphic is DISK I/O not CPU usage

    My VM is on a separate disk 

    I have put you at www.cyrilblanc/vsl1.png to vsl11.png all the result of BigTop

    There is another user that have report that the new version is slower too http://community.vsl.co.at/forums/p/27428/179771.aspx#179771

    what do you think of this idea :

    Just  an idea :

    You load a VE PRO in a VE PRO server, pause this VE PRO

    Load all the VE PRO

    Then you start all the VE PRO.

    Best

    Cyril

    P.S. How do I return to previous version ?


  • Test done with a raid 0 of 6 raptor  and normal disks show exactly the same, fast load to start then it slow down

    Best

    Cyril


  • This is a very interesting thread for computer nerds like myself :).  I can't wait to see what ultimately is the problem.

    Maestro2be


  • last edited
    last edited

    @commalot said:

    I'm going to build a new computer around an I7 950 with a RME 9632. I've been reading about SSD drives but I'm not sure I would get any appreciable performance increase from them. I will be using VSL SE +, Spectrasonics, Symphonic Choirs and selected samples in Kontakt 4. I can afford them but do I need them?

    According to tests done on other forums, there is an appreciable difference in performance using SSD when using Windows as your OS. Not only is the loading time faster, but the pre-load buffer for Kontakt can be reduced, which  not only reduces the RAM footprint, but also reduces loading time still further. You should also be able to get increased polyphony from the drives.

    However, it all depends on your workflow whether or not this has any real point. For me the faster loading times would make very little difference, as I only load the computer once a day. However, if you like to swap templates often, then the faster loading time will certainly be noticed.

    DG


  • Just recently upgraded the boot and program partitions to SSD.

    Even without Win7 & Advanced Host Controller Mode the difference is huge.

    Vienna Instruments & Cubase just "pop" out instead of the HD cantiliver scratching around, killing time.

    Same of course is true for general system performance ( bootup time and "feeling under the mouse button"). That alone is worth the price.

    Sample libraries on SSDs maybe a different story. I think they are extremely strong in random read. For pure data transfer ( and that´s what happens when you have to load >1GB of samples in one row) I´m not sure if they are superior to conventional HDS.

    Against regular discs I think they still opearte faster.

    But regarding some of the system specs of Vienna users here ( i7 extreme...6 Velociraptors in Raid 0 [:)] I don´t want to bet on it


  • I just had contact with vsl support concerning my problem, that during tutti passages in my 8 Minutes romantic orchestral piece, the VI-64bit plugin of my Cubase 5 64bit host needs more speed to grab the data from 5400 rpm Harddisk. There ist two parameters, I guess, which have to be considered important. One: the transfer rate speed and two: the access speed (which is important for grabbing small chunks of data spread over the storage media. I think that the access speed (german: Zugriffszeit) is the factor to keep in consideration. But I'm not a specialist in such things. cheers,

  •  5400 is just too slow for sample streaming. Use 7200 and all will be well.

    DG


  • average seektime is often the problem (the time the heads need to get positioned over the right sector of the disk).

    usual values for 7.200 rpm disks are around 9 ms, WD raptors ca. 4.5 ms, velocity raptor 4.2 ms, SSDs < 0.1 ms, 5400 rpm disks around 15 ms


    and remember: only a CRAY can run an endless loop in just three seconds.
  • Well I don't think so. I just put in a 7200 rpm 2,5 HDD and it didn't work. Same problem. Are there other solutions? Selling my notebook and by a Desktop, just because the VSL philosophy depends on smallest access times which can only be achieved with 3,5 7200 rpm drives or SSD Drives? Wow that's hard stuff. My notebook is just 4 months old. Nobody not even the VSL support told me that it is impossible to operate a large orchestra with a notebook, which is brand new and has a powerfull provessor. It's access time stupid. That's what I'm going to tell my image in the mirror before getting mad.

  • hmmm, i can't believe a somehow modern notebook with a 7.200 drive cannot handle SE and even a little bit more ... is there anything else eventually preventing throughput (you don't record = write simultaneously to the disk, do you?), what about your soundcard (ASIO settings)?

    paul often does clinics with full orchestra arrangements on his macBook PRO (which was *only* some core 2 duo model ...)


    and remember: only a CRAY can run an endless loop in just three seconds.
  • I think it is dependent on what you call "large" orchestra.

    Did you connect the new 7200rpm harddics internally or do you still use an external USB2.0 case ?

    I think USB 2.0 will be more a bottleneck than the 7200rpm, although as mentioned before, access time is also an important factor.

    But the most important point is how many instruments you gonna play at once, maybe there also is something additional going on with other softsynths, samplers etc.

    Also : What audiointerface do you use ? There can be large differences in CPU and overhead demand between the models. If you get clicks and dropouts, try using a larger latency.

    IMO you should be able to basically make music with an i5 laptop and a fast conventional HD using the Special Edition.

    But hey, there are people here on the forum discussing about 3x Raid0 WD Velociraptors or multiple SSDs. It really depends on what you want to do.


  • Of course I installed the HDD drive within the Notebook. The usb 2.0 interface only brings it to 35 MB/s because although theoretically it should be 60 MB/s with HI-speed, the USB 2.0 protocol has in contrast to Firewire 400 a much too big security overlay redundancy. I wrote USB 2.0 in my setting because there is already usb 3.0 and eSATA available for notebooks, which might help me out now. Within one year usb 2.0 will be what is now usb 1.1. It might be possible that an Apple chipset will be 2 or 3 MB/s more effective but the problem here is not the chipset but the usb 2.0 protocol. What's the hardware recommendation of the people of VSL? For a flute, for a violin and a clarinet and for big orchestra?