Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fall back properly if chosen factory class isn't working. #28

Merged
merged 6 commits into from
Sep 22, 2020
106 changes: 65 additions & 41 deletions src/main/java/org/jitsi/srtp/crypto/Aes.java
Original file line number Diff line number Diff line change
Expand Up @@ -266,8 +266,7 @@ public void run(Cipher cipher) throws Exception
private static CipherFactory benchmark(
CipherFactory[] factories,
int keySize,
String transformation,
boolean warmup
String transformation
)
{
long minTime = Long.MAX_VALUE;
Expand All @@ -282,6 +281,14 @@ private static CipherFactory benchmark(
if (factory == null)
continue;

// The user may have specified a specific CipherFactory class
// (name) through setFactoryClassName(String), Practically, FACTORY_CLASS_NAME may override
// minFactory and, consequently, it may appear that the benchmark is
// unnecessary. Technically though, the specified CipherFactory may
// malfunction. That is why FACTORY_CLASS_NAME is selected after it has
// proven itself functional.
boolean chosenFactoryClass = (factory.getClass().equals(Aes.factoryClass));
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Extra parens on purpose?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, no, I reorganized this from an if statement.


try
{
Cipher cipher = factory.createCipher(transformation);
Expand All @@ -295,18 +302,24 @@ private static CipherFactory benchmark(
}
else
{
/* If this is the user-chosen factory class, we don't really
* need to benchmark it; just verify that it's working.
*/
int numWarmups = chosenFactoryClass ? 0 : NUM_WARMUPS;
int numBenchmarks = chosenFactoryClass ? 1 : NUM_BENCHMARKS;

BenchmarkOperation benchmark = BenchmarkOperation.getBenchmark(transformation, keySize);
if (warmup)
if (!chosenFactoryClass)
{
// Let the JVM "warm up" (do JIT compilation and the like)
for (int i = 0; i < NUM_WARMUPS; i++)
for (int i = 0; i < numWarmups; i++)
{
benchmark.run(cipher);
}
}

long startTime = System.nanoTime();
for (int i = 0; i < NUM_BENCHMARKS; i++)
for (int i = 0; i < numBenchmarks; i++)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should line 329 use numBenchmarks instead of NUM_BENCHMARKS? I guess in practice it doesn't matter

{
benchmark.run(cipher);
}
Expand Down Expand Up @@ -335,6 +348,19 @@ private static CipherFactory benchmark(
else if (t instanceof ThreadDeath)
throw (ThreadDeath) t;
}

if (chosenFactoryClass)
{
if (minFactory != null)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure I understand this well. If this loop iteration is for the user-configured factory class, return the fastest factory. If a previous non-user-configured factory was faster, we short-circuit and return it (the faster one)?

Should that return factory instead of minFactory?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Well, elsewhere in the code I reorganize the list so the user-configured factory class is always first, so we avoid profiling factories we don't need to. But I agree that's fragile and confusing.

I've reorganized it now, is it clearer?

{
return minFactory;
}
else
{
logger.warn("Chosen factory class \"" + FACTORY_CLASS_NAME +
"\" not working for " + transformation);
}
}
}

if (log.length() != 0)
Expand Down Expand Up @@ -537,26 +563,47 @@ else if (t instanceof ThreadDeath)

// If FACTORY_CLASS_NAME does not specify a well-known Class, add the
// new Class to FACTORY_CLASSES.
if (add && (factoryClass != null))
if (factoryClass != null)
{
for (Class<?> clazz : factoryClasses)
Class<?>[] newFactoryClasses;
if (add)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think all of this is equivalent to factoryClasses = ArrayUtils.add(factoryClasses, Class.class, factoryClass); (from jitsi-utils)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not quite - I want to put the new factory class first, whereas ArrayUtils.add would seem to put it last.

{
if (factoryClass.equals(clazz))
for (Class<?> clazz : factoryClasses)
{
add = false;
break;
if (factoryClass.equals(clazz))
{
add = false;
break;
}
}
}
if (add)
{
Class<?>[] newFactoryClasses
= new Class<?>[1 + factoryClasses.length];
if (add)
{
newFactoryClasses
= new Class<?>[1 + factoryClasses.length];

newFactoryClasses[0] = factoryClass;
System.arraycopy(
newFactoryClasses[0] = factoryClass;
System.arraycopy(
factoryClasses, 0,
newFactoryClasses, 1,
factoryClasses.length);
}
}
else
{
/* Otherwise, move the FACTORY_CLASS to the beginning of the list. */
newFactoryClasses
= new Class<?>[factoryClasses.length];

newFactoryClasses[0] = factoryClass;
int i = 1;
for (Class<?> clazz : factoryClasses)
{
if (!factoryClass.equals(clazz))
{
newFactoryClasses[i] = clazz;
i++;
}
}
factoryClasses = newFactoryClasses;
}
}
Expand Down Expand Up @@ -639,30 +686,7 @@ private static CipherFactory getCipherFactory(String transformation, boolean war
// Benchmark the StreamCiphers provided by the available
// StreamCipherFactories in order to select the fastest-performing
// CipherFactory.
CipherFactory minFactory = benchmark(factories, keySize, transformation, warmup);

// The user may have specified a specific CipherFactory class
// (name) through setFactoryClassName(String), Practically, FACTORY_CLASS_NAME may override
// minFactory and, consequently, it may appear that the benchmark is
// unnecessary. Technically though, the specified CipherFactory may
// malfunction. That is why FACTORY_CLASS_NAME is selected after it has
// proven itself functional.
{
Class<? extends CipherFactory> factoryClass = Aes.factoryClass;

if (factoryClass != null)
{
for (CipherFactory factory : factories)
{
if ((factory != null)
&& factory.getClass().equals(factoryClass))
{
minFactory = factory;
break;
}
}
}
}
CipherFactory minFactory = benchmark(factories, keySize, transformation);

return minFactory;
}
Expand Down