Minor Bug with LS win.

You name it!
grishata
Developer
Posts: 138
Joined: Thu Jan 24, 2008 7:21 pm
Location: Bulgaria
Contact:

Re: Minor Bug with LS win.

Post by grishata » Fri Aug 19, 2011 9:26 am

typewriter wrote:It's now working perfectly on OSX. The only drawback is the Ram-usage on OSX. Each instance eats up another 100MB of RAM. 8 instances are 800MB RAM only for the plugins while on WIN this is only around 100MB for all instances.
I can't reproduce that. Are you sure you are using the latest version? In fantasia choose Actions/Sampler Information, should be 1.0.0.cvs14.

typewriter
Advanced User
Posts: 147
Joined: Fri Sep 26, 2008 9:04 am

Re: Minor Bug with LS win.

Post by typewriter » Fri Aug 19, 2011 10:08 am

Yes, latest version. 1.0.0 csv14.

For demonstration I have currently inserted 24 instances of LS with a total of 384 stereo channels into Reaper 64.

The Reaper project is empty. Only linuxsampler is inserted, there are no samples loaded.

The Ram usage displayed by Reaper is 2470MB, CPU use around 50% on a dual core 3ghz, 4GB RAM Macbook Pro with leopard.

When I remove the LS Plugins CPU use goes down to near 3% and RAM usage is around 69MB.

In general adds every LS instance around 100MB Ram use.

grishata
Developer
Posts: 138
Joined: Thu Jan 24, 2008 7:21 pm
Location: Bulgaria
Contact:

Re: Minor Bug with LS win.

Post by grishata » Fri Aug 19, 2011 11:33 am

What amount of max voices you are using? I guess this could happen if you set max voices to 1000 or something.

typewriter
Advanced User
Posts: 147
Joined: Fri Sep 26, 2008 9:04 am

Re: Minor Bug with LS win.

Post by typewriter » Fri Aug 19, 2011 1:48 pm

Reducing the voice count in Fantasia from 1024 to 100 voices reduces the RAM usage very little. Less than 10mb per instance. 24 instances are still more than 2400MB.

I am testing with the snapshot from aug. 17th.

I'll update to the latest snapshot from today.

Update:

No difference with latest snapshot (Aug. 19).

24 instances (100 voices) still use 2400MB Ram.

grishata
Developer
Posts: 138
Joined: Thu Jan 24, 2008 7:21 pm
Location: Bulgaria
Contact:

Re: Minor Bug with LS win.

Post by grishata » Fri Aug 19, 2011 3:32 pm

I did some profiling and it turned out that most of the memory goes for streams. You also need to set the max streams less than 100, for example. So, I guess this is how it should be (it's not an issue introduced by the recent changes).

typewriter
Advanced User
Posts: 147
Joined: Fri Sep 26, 2008 9:04 am

Re: Minor Bug with LS win.

Post by typewriter » Fri Aug 19, 2011 5:45 pm

The number of streams was always set to 90 all the time.

But I think you are right. When I set the number of streams to 1000 the plugin uses aroung 1000MB instead of 100MB for 90 streams.

It looks like LS reserves 1MB for every stream. So when I insert 8 instances with 90 streams the used Ram would be 8 * 90 * 1 = 720MB and for 24 instances 24 * 90 * 1MB = 2160MB. This is around the results I had while testing.

I thought the numbers of voices and streams you can set up in Fantasia are meant for LS as unity - but they are obviously set up for every single instance.

There seems to be a difference to the Win version. The Ram usage stays there a around 100 MB and does not change significantly when more instances of the LS plugin are used.

User avatar
cuse
Developer
Posts: 319
Joined: Wed Jan 23, 2008 10:07 pm
Location: Germany

Re: Minor Bug with LS win.

Post by cuse » Fri Sep 23, 2011 11:38 pm

Max. voices and disk streams are global settings, yes. And both VST and AU plugins create just one instance of LinuxSampler, no matter how many VSTs you have spawn (in the same VST host application of course). But you also should know, beside the LinuxSampler instance itself, there are also "sampler engine" instances inside LinuxSampler. A sampler engine instance always lives in the context of exactly one audio thread. And each sampler engine instance allocates its own disk thread and thus its own disk streaming buffers. I never actually worked on the AU plugin version of LinuxSampler, but it sounds like the AU plugin has a separate audio thread for each plugin instance.

Post Reply