While debugging the performance of my C# application I noticed that it's a lot slower than it's Java equivalence. After looking into it the problem seems to be caused by the encryption/decryption methods.
I am forced to use AES encryption with mode set to CFB8 and no Padding. For Java this is quite straight forward as I can use Cipher.getInstance("AES/CFB8/NoPadding");. In C# I found out that I need to use new RijndaelManaged(). After running tests with the same keys and same data it here are the results:
Java:
- Encrypt: 0.402s
- Decrypt: 0.480s
C#:
- Encrypt: 4.201s
- Decrypt: 3.671s
C# Cipher code:
public ICryptoTransform enc;
public ICryptoTransform dec;
public AesCrypto(byte[] key)
{
enc = Generate(key).CreateEncryptor();
dec = Generate(key).CreateDecryptor();
}
private SymmetricAlgorithm Generate(byte[] key) {
RijndaelManaged cipher = new RijndaelManaged();
cipher.Mode = CipherMode.CFB;
cipher.Padding = PaddingMode.None;
cipher.KeySize = 128;
cipher.FeedbackSize = 8;
cipher.Key = key;
cipher.IV = key;
return cipher;
}
public byte[] Crypt(byte[] buffer, int offset, int count) {
return enc.TransformFinalBlock(buffer, offset, count);
}
C# Test code:
static void Test() {
// Init
var AesCrypto = new AesCrypto(Encoding.UTF8.GetBytes("aaabbbccaaabbbcc"));
var testData = Encoding.UTF8.GetBytes(createDataSize(9000000)); // 9mb test.
// Timer
var stopWatch = new Stopwatch();
stopWatch.Start();
AesCrypto.Crypt(testData, 0, testData.Length);
stopWatch.Stop();
Console.WriteLine("AesCrypto.Crypt took: " + stopWatch.ElapsedMilliseconds);
}
static string createDataSize(int msgSize)
{
StringBuilder sb = new StringBuilder(msgSize);
for (int i = 0; i < msgSize; i++)
{
sb.Append('a');
}
return sb.ToString();
}
Result: "AesCrypto.Crypt took: 3626"
JAVA Cipher code:
public Cryptor(boolean reader) throws CryptingException {
keySpec = new SecretKeySpec(secretKey.getBytes(CHARSET), "AES");
ivSpec = new IvParameterSpec(iv.getBytes(CHARSET));
try {
cipher = Cipher.getInstance("AES/CFB8/NoPadding");
if (reader) cipher.init(Cipher.DECRYPT_MODE, keySpec, ivSpec);
else cipher.init(Cipher.ENCRYPT_MODE, keySpec, ivSpec);
} catch (NoSuchAlgorithmException e) {
throw new SecurityException(e);
} catch (NoSuchPaddingException e) {
throw new SecurityException(e);
}catch (InvalidKeyException e) {
throw new SecurityException(e);
} catch (InvalidAlgorithmParameterException e) {
throw new SecurityException(e);
}
}
public byte[] decrypt(byte[] input) throws CryptingException {
return cipher.update(input);
}
public byte[] encrypt(String input) throws CryptingException {
return cipher.update(input.getBytes());
}
Java Test code:
private static void Test() {
// Init
String data = createDataSize(9392963);
Cryptor writer = new Cryptor(false);
// Timer
Instant starts = Instant.now();
byte[] encrypted = writer.encrypt(data);
Instant ends = Instant.now();
System.out.println("Java Encryption took: " + Duration.between(starts, ends));
}
private static String createDataSize(int msgSize) {
StringBuilder sb = new StringBuilder(msgSize);
for (int i=0; i<msgSize; i++) {
sb.append('a');
}
return sb.toString();
}
Result: "Java Encryption took: PT0.469S"
Possible solution:
After researching this quite a bit it seems that AesCryptoServiceProvider() has around the same performance as Java's equivalence and has a mostly identical results. The problem with this is however that it requires padding, whereas Java's equivalence does not require padding. For example this means if I want to encrypt "abcdab" it would only encrypt "abcd" and return the result for that and keep the rest ("ab") internally. If I use padding I can get it to return the full "abcdab" encrypted, however then it has additional data appended and the symmetric algorithm is de-synced, because in java I could encrypt "abcdab" without any padding.
Question
So finally my question is how would could I make the C# encryption/decryption process just as fast as Java's? Am I doing something wrong with AesCryptoServiceProvider, maybe it's possible for it to not require padding?
