First, some relevant background info: I've got a CoreAudio-based low-latency audio processing application that does various mixing and special effects on audio that is coming from an input device on a purpose-dedicated Mac (running the latest version of MacOS) and delivers the results back to one of the Mac's local audio devices.
In order to obtain the best/most reliable low-latency performance, this app is designed to hook in to CoreAudio's low-level audio-rendering callback (via AudioDeviceCreateIOProcID(), AudioDeviceStart(), etc) and every time the callback-function is called (from the CoreAudio's realtime context), it reads the incoming audio frames (e.g. 128 frames, 64 samples per frame), does the necessary math, and writes out the outgoing samples.
This all works quite well, but from everything I've read, Apple's CoreAudio implementation has an unwritten de-facto requirement that all real-time audio operations happen in a single thread. There are good reasons for this which I acknowledge (mainly that outside of SIMD/SSE/AVX instructions, which I already use, almost all of the mechanisms you might employ to co-ordinate parallelized behavior are not real-time-safe and therefore trying to use them would result in intermittently glitchy audio).
However, my co-workers and I are greedy, and nevertheless we'd like to do many more math-operations per sample-buffer than even the fastest single core could reliably execute in the brief time-window that is necessary to avoid audio-underruns and glitching.
My co-worker (who is fairly experienced at real-time audio processing on embedded/purpose-built Linux hardware) tells me that under Linux it is possible for a program to requisition exclusive access for one or more CPU cores, such that the OS will never try to use them for anything else. Once he has done this, he can run "bare metal" style code on that CPU that simply busy-waits/polls on an atomic variable until the "real" audio thread updates it to let the dedicated core know it's time to do its thing; at that point the dedicated core will run its math routines on the input samples and generate its output in a (hopefully) finite amount of time, at which point the "real" audio thread can gather the results (more busy-waiting/polling here) and incorporate them back into the outgoing audio buffer.
My question is, is this approach worth attempting under MacOS/X? (i.e. can a MacOS/X program, even one with root access, convince MacOS to give it exclusive access to some cores, and if so, will big ugly busy-waiting/polling loops on those cores (including the polling-loops necessary to synchronize the CoreAudio callback-thread relative to their input/output requirements) yield results that are reliably real-time enough that you might someday want to use them in front of a paying audience?)
It seems like something that might be possible in principle, but before I spend too much time banging my head against whatever walls might exist there, I'd like some input about whether this is an avenue worth pursuing on this platform.
can a MacOS/X program, even one with root access, convince MacOS to give it exclusive access to some cores
I don't know about that, but you can use as many cores / real-time threads as you want for your calculations, using whatever synchronisation methods you need to make it work, then pass the audio to your IOProc using a lock free ring buffer, like TPCircularBuffer.
But your question reminded me of a new macOS 11/iOS 14 API I've been meaning to try, the Audio Workgroups API (2020 WWDC Video).
My understanding is that this API lets you "bless" your non-IOProc real-time threads with audio real-time thread properties or at least cooperate better with the audio thread.
The documents distinguish between the threads working in parallel (this sounds like your case) and working asynchronously (this sounds like my proposal), I don't know which case is better for you.
I still don't know what happens in practice when you use Audio Workgroups, whether they opt you in to good stuff or opt you out of bad stuff, but if they're not the hammer you're seeking, they may have some useful hammer-like properties.
Related
We have an application that uses waveXXX() and mixerXXX() functions to handle the audio I/O to and from some instruments (think: oscilloscope or electronics rather than musical instruments, not that it much matters). It's finally time to stop deploying it on Windows XP, and move it to Windows 7 and/or 8.
From reading a variety of material on WASAPI, it sounds like the bulk of the application (based on waveXXX() functions) might actually work fine, but the mixer() stuff used to set master output volume, line in volume, and mute the microphone will definitely have to change, and use IAudioEndPointVolume calls instead.
Is it possible to change only the mixerXXX() calls? Is it desirable?
Logically, this application requires exclusive use of its audio endpoints (speaker out, line in). If I want to ensure exclusive access through software, would that force me to rewrite all the waveXXX() code too? (The alternative is to warn users that other audio applications may interfere with this one).
My recommendation:
If you need exclusive access, convert everything to WASAPI
If you are using line-in, convert everything to WASAPI
If you have time, convert everything to WASAPI
If you are strictly only using speaker and microphone in shared mode, replace mixerXXX() with the ISimpleAudioVolume interface (and several other interfaces to get to it), then test whether existing waveXXX() code behaves as you need it to. Then test each time hardware, OS or audio drivers change. Better still, just convert to WASAPI.
In my case, exclusive speaker output is critical - this drives the instrument that generates a related input signal. I guess I don't mind if another application wants to share access to that incoming signal, but logically it is a system that wants an exclusive contract with its audio endpoints.
That exclusivity requires that I obtain an IMMDevice instance for both speaker output and line-in input, Activate() the IAudioClient interface on them and Initialize() both using AUDCLNT_SHAREMODE_EXCLUSIVE (see also this answer).
But have I actually selected line-in by such a process? Probably not. All I can be sure of is annoying any other applications who were previously sharing my endpoints by cutting them off.
Having done this much, it's really not clear what will happen to waveInXXX() calls - maybe they'll take from line-in, maybe from microphone - maybe it depends on how the hardware vendor implements their end of the deal. It's also never been clear to me whether line-in and microphone are always multiplexed (i.e. selectable), always mixed (i.e. you can only simulate selection by muting the other one) or there is no standard one can rely on.
Because of factors like that, it's a gamble not to use WASAPI throughout.
I’m working on a packet reshaping project in Linux using the BeagleBone Black. Basically, packets are received on one VLAN, modified, and then are sent out on a different VLAN. This process is bidirectional - the VLANs are not designated as being input-only or output-only. It’s similar to a network bridge, but packets are altered (sometimes fairly significantly) in-transit.
I’ve tried two different methods for accomplishing this:
Creating a user space application that opens raw sockets on both
interfaces. All packet processing (including bridging) is handled in
the application.
Setting up a software bridge (using the kernel
bridge module) and adding a kernel module that installs a netfilter
hook in post routing (NF_BR_POST_ROUTING). All packet processing is
handled in the kernel.
The second option appears to be around 4 times faster than the first option. I’d like to understand more about why this is. I’ve tried brainstorming a bit and wondered if there is a substantial performance hit in rapidly switching between kernel and user space, or maybe something about the socket interface is inherently slow?
I think the user application is fairly optimized (for example, I’m using PACKET_MMAP), but it’s possible that it could be optimized further. I ran perf on the application and noticed that it was spending a good deal of time (35%) in v7_flush_kern_dcache_area, so perhaps this is a likely candidate. If there are any other suggestions on common ways to optimize packet processing I can give them a try.
Context switches are expensive and kernel to user space switches imply a context switch. You can see this article for exact numbers, but the stated durations are all in the order of microseconds.
You can also use lmbench to benchmark the real cost of context switches on your particular cpu.
The performance of the user space application depends on the used syscall to monitor the sockets too. The fastest syscall is epoll() when you need to handle a lot of sockets. select() will perform very poor, if you handle a lot of sockets.
See this post explaining it:
Why is epoll faster than select?
the problem I try to deal with it is the saving of big number (millions) of small files (up to 50KB), which are sent via network. The saving is done sequential: server receives a file or a dir (via network), it saves it on disk; the next one arrives, it's saved etc.
Apparently, the performance is not acceptable, if multiple server processes coexist (let's say I have 5 processes which all read from network and write at the same time), because the I/O scheduler doesn't manage to merge efficiently the I/O writes.
A suggested solution is to implement some sort of buffering: each server process should have a 50MB cache, in which it should write the current file, do a chdir etc; when the buffer is full, it should be synced to disk, therefore obtaining an I/O burst.
My questions to you:
1) I know that already exists a buffer mechanism (disk buffer); do you think that the above scenario is going to add some improvement? (the design is much more complicated and it's not easy to implement a simple test case)
2) do you have any suggestions, where to look if I would implement this?
Many thanks.
You're going to need to do better than
"apparently the performance is not acceptable".
Specifically
How are you measuring it? Do you have an exact, reproducible figure
What is your target?
In order to do optimisation, you need two things- a method of measuring it (a metric) and a target (so you know when to stop, or how useful or useless a particular technique is).
Without either, you're sunk, I'm afraid.
How important are those writes? I have three suggestions (which can be combined), but one of them is a lot of work, and one of them is less safe...
Journaling
I'm guessing you're seeing some poor performance due in part to the journaling common to most modern Linux filesystems. The journaling causes barriers to be inserted into the IO queue when file metadata is written. You can try turning down the safety (and maybe turning up the speed) with mount(8) options barrier=0 and data=writeback.
But if there is a crash, the journal might not be able to prevent a lengthy fsck(8). And there's a chance the fsck(8) will wind up throwing away your data when fixing the problem. On the one hand, it's not a step to take lightly, on the other hand, back in the old days, we ran our ext2 filesystems in async mode without a journal both ways in the snow and we liked it.
IO Scheduler elevator
Another possibility is to swap the IO elevator; see Documentation/block/switching-sched.txt in the Linux kernel source tree. The short version is that deadline, noop, as, and cfq are available. cfq is the kernel default, and probably what your system is using. You can check:
$ cat /sys/block/sda/queue/scheduler
noop deadline [cfq]
The most important parts from the file:
As of the Linux 2.6.10 kernel, it is now possible to change the
IO scheduler for a given block device on the fly (thus making it possible,
for instance, to set the CFQ scheduler for the system default, but
set a specific device to use the deadline or noop schedulers - which
can improve that device's throughput).
To set a specific scheduler, simply do this:
echo SCHEDNAME > /sys/block/DEV/queue/scheduler
where SCHEDNAME is the name of a defined IO scheduler, and DEV is the
device name (hda, hdb, sga, or whatever you happen to have).
The list of defined schedulers can be found by simply doing
a "cat /sys/block/DEV/queue/scheduler" - the list of valid names
will be displayed, with the currently selected scheduler in brackets:
# cat /sys/block/hda/queue/scheduler
noop deadline [cfq]
# echo deadline > /sys/block/hda/queue/scheduler
# cat /sys/block/hda/queue/scheduler
noop [deadline] cfq
Changing the scheduler might be worthwhile, but depending upon the barriers inserted into the queue by the journaling requirements, there might not be much reordering possible. Still, it is less likely to lose your data, so it might be the first step.
Application changes
Another possibility is to drastically change your application to bundle files itself, and write fewer, larger, files to disk. I know it sounds strange, but (a) the iD development team packaged their maps, textures, objects, etc., into giant zip files that they would read into the program with a few system calls, unpack, and run with, because they found the performance much better than reading a few hundred or few thousand smaller files. Load times between levels was drastically shorter. (b) The Gnome desktop team and KDE desktop teams took different approaches to loading their icons and resource files: the KDE team packages their many small files into larger packages of some sort, and the Gnome team did not. The Gnome team had longer startup delays and were hoping the kernel could make some efforts to improve their startup time. The kernel team kept suggesting the fewer, larger, files approach.
Creating/renaming a file, syncing it, having lots of files in a directory and having lots of files (with tail waste) are some of the slow operations in your scenario. However to avoid them it would only help to write lesser files (for example writing out archives, concatenated file or similiar). I would actually try a (limited) parallel async or sync approach. The IO scheduler and caches are typically quite good.
I'm looking into making some software that makes the keyboard function like a piano (e.g., the user presses the 'W' key and the speakers play a D note). I'll probably be using OpenAL. I understand the basics of digital audio, but playing real-time audio in response to key presses poses some problems I'm having trouble solving.
Here is the problem: Let's say I have 10 audio buffers, and each buffer holds one second of audio data. If I have to fill buffers before they are played through the speakers, then I would would be filling buffers one or two seconds before they are played. That means that whenever the user tries to play a note, there will be a one or two second delay between pressing the key and the note being played.
How do you get around this problem? Do you just make the buffers as small as possible, and fill them as late as possible? Is there some trick that I am missing?
Most software synthesizers don't use multiple buffers at all.
They just use one single, small ringbuffer that is constantly played.
A high priority thread will as often as possible check the current play-position and fill the free part (e.g. the part that has been played since the last time your thread was running) of the ringbuffer with sound data.
This will give you a constant latency that is only bound by the size of your ring-buffer and the output latency of your soundcard (usually not that much).
You can lower your latency even further:
In case of a new note to be played (e.g. the user has just pressed a key) you check the current play position within the ring-buffer, add some samples for safety, and then re-render the sound data with the new sound-settings applied.
This becomes tricky if you have time-based effects running (delay lines, reverb and so on), but it's doable. Just keep track of the last 10 states of your time based effects every millisecond or so. That'll make it possible to get back 10 milliseconds in time.
With the WinAPI, you can only get so far in terms of latency. Usually you can't get below 40-50ms which is quite nasty. The solution is to implement ASIO support in your app, and make the user run something like Asio4All in the background. This brings the latency down to 5ms but at a cost: other apps can't play sound at the same time.
I know this because I'm a FL Studio user.
The solution is small buffers, filled frequently by a real-time thread. How small you make the buffers (or how full you let the buffer become with a ring-buffer) is constrained by scheduling latency of your operating system. You'll probably find 10ms to be acceptable.
There are some nasty gotchas in here for the uninitiated - particularly with regards to software architecture and thread-safety.
You could try having a look at Juce - which is a cross-platform framework for writing audio software, and in particular - audio plugins such as SoftSynths and effects. It includes software for both sample plug-ins and hosts. It is in the host that issues with threading are mostly dealt with.
Here's what I want to do:
I want to allow the user to give my program some sound data (through a mic input), then hold it for 250ms, then output it back out through the speakers.
I have done this already using Java Sound API. The problem is that it's sorta slow. It takes a minimum of about 1-2 seconds from the time the sound is made to the time the sound is heard again from the speakers, and I haven't even tried to implement delay logic yet. Theoretically there should be no delay, but there is. I understand that you have to wait for the sound card to fill up its buffer or whatever, and the sample size and sampling rate have something to do with this.
My question is this: Should I continue down the Java path trying to do this? I want to get the delay down to like 100ms if possible. Does anyone have experience using the ASIO driver with Java? Supposedly it's faster..
Also, I'm a .NET guy. Does this make sense to do with .NET instead? What about C++? I'm looking for the right technology to use here, and maybe a good example of how to read/write to audio input/output streams using your suggested technology platform. Thanks for your help!
I've used JavaSound in the past and found it wonderfully flaky (and it keeps changing between VM releases). If you like C#, use it, just use the DirectX APIs. Here's an example of doing kind of what you want to do using DirectSound and C#. You could use the Effects plugins to perform your 250 ms echo.
http://blogs.microsoft.co.il/blogs/tamir/archive/2008/12/25/capturing-and-streaming-sound-by-using-directsound-with-c.aspx
You may want to look into JACK, an audio API designed for low-latency sound processing. Additionally, Google turns up this nifty presentation [PDF] about using JACK with Java.
Theoretically there should be no delay, but there is.
Well, it's impossible to have zero delay. The best you can hope for is an unnoticeable delay (in terms of human perception). It might help if you describe your basic algorithm for reading & writing the sound data, so people can identify possible problems.
A potential issue with using a garbage-collected language like Java is that the GC will periodically run, interrupting your processing for some arbitrary amount of time. However, I'd be surprised if it's >100ms in normal usage. If GC is a problem, most JVMs provide alternate collection algorithms you can try.
If you choose to go down the C/C++ path, I highly recommend using PortAudio ( http://portaudio.com/ ). It works with almost everything on multiple platforms and it gives you low-level control of the sound drivers without actually having to deal with the various sound driver technology that is around.
I've used PortAudio on multiple projects, and it is a real joy to use. And the license is permissive.
If low latency is your goal, you can't beat C.
libsoundio is a low-level C library for real-time audio input and output. It even comes with an example program that does exactly what you want - piping the microphone input to the speakers output.
It's possible with JavaSound to get end-to-end latency in the ballpark of 100-150ms.
The primary cause of latency is the buffer sizes of the capture and playback lines. The bufferSize is set when opening the lines:
capture: TargetDataLine#open(AudioFormat format, int bufferSize)
playback: SourceDataLine#open(AudioFormat format, int bufferSize)
If the buffer is too big it will cause excess latency, but if it's too small it will cause stuttery playback. So you need to find a balance for your applications needs and your computing power.
The default buffer size can be checked with DataLine#getBufferSize when calling #open(AudioFormat format). The default size will vary based on the AudioFormat and seems to be geared for high latency, stutter free playback applications (e.g. internet streaming). If you're developing a low latency application, the default buffer size is much too large and should be changed.
In my testing with a 16-bit PCM AudioFormat, a buffer size of 1024 bytes has been pretty close to ideal for low latency.
The second and often overlooked cause of audio latency is any other activity being done in the capture or playback threads. For example, logging messages to console can introduce 10's of ms of latency. Turn it off.