90 likes | 202 Views
This project explores a hybrid approach to searching for AES keys with missing lower-order bits using parallel processing in shared-memory and distributed environments. By evaluating all possible combinations of the missing bits, the method achieves significant performance improvements through effective thread and node management. Implementations include sequential, SMP, and cluster versions, each designed to optimize key search efficiency. The approach also introduces early termination mechanisms when the correct key is identified, allowing for faster completion of the search process.
E N D
Computer Science 320 Introduction to Hybrid SMP/Clusters
Massively Parallel AES Key Search • Inputs: • 128-bit plain text block • 128-bit cipher text block • 256-bit key with N lower-order bits missing • N, the number of missing lower-order key bits • Output: The correct key to encode the text • Method: try all 2N possible missing bit strings to find the key that does encrypt the plain text correctly
Sequential Version for (intcntr = 0; cntr < maxcount; ++cntr){ // Try key } Tries up to 2N possible keys
SMP Version new ParallelTeam().execute(new ParallelRegion(){ public void run() throws Exception{ execute(0, maxcounter, new IntegerForLoop(){ public void run(){ for (intcntr = first; cntr <= last; ++cntr){ // Try key } } }); } }); Splits 2N possible keys evenly among K threads
Cluster Version Range chunk = new Range(0, maxcounter – 1).subrange(size, rank); int lb = chunk.lb(); int ub = chunk.ub(); for (int cntr = lb; cntr <= ub; ++cntr){ // Try key } Splits 2N possible keys evenly among K nodes
Hybrid Version Range chunk = new Range(0, maxcounter – 1).subrange(size, rank); int lb = chunk.lb(); int ub = chunk.ub(); new ParallelTeam().execute(new ParallelRegion(){ public void run() throws Exception{ execute(lb, ub, new IntegerForLoop(){ public void run(){ for (int cntr = first; cntr <= last; ++cntr){ // Try key } } }); } }); Splits 2N possible keys evenly among Kp nodes * Kt threads per node
Specify # Nodes and # Threads Range chunk = new Range(0, maxcounter – 1).subrange(size, rank); int lb = chunk.lb(); int ub = chunk.ub(); new ParallelTeam().execute(new ParallelRegion(){ public void run() throws Exception{ execute(lb, ub, new IntegerForLoop(){ public void run(){ for (int cntr = first; cntr <= last; ++cntr){ // Try key } } }); } }); $ java –Dpj.port=28000 –Dpj.np=8 -Dpj.nt=8 . . .
Improvement: Early Return for (intcntr = first; cntr <= last; ++cntr){ intlsbs = keylsbs | cntr; trialKey[28] = (byte) (lsbs >>> 24); trialKey[29] = (byte) (lsbs >>> 16); trialKey[30] = (byte) (lsbs >>> 8); trialKey[31] = (byte) (lsbs ); cipher.setKey(trialKey); cipher.encrypt(plainText, trialciphertext); if (match(ciphertext, trialciphertext)){ foundKey = new byte[32] System.arraycopy(trialkey, 0, foundkey, 0, 32); } } . . . if (foundKey != null) System.out.println(Hex.toString(foundkey)); Would like to quit this thread, other threads in the same node, and other nodes early, when the correct key is found
Improvement: Sizeup for (intcntr = first; cntr <= last; ++cntr){ intlsbs = keylsbs | cntr; trialKey[28] = (byte) (lsbs >>> 24); trialKey[29] = (byte) (lsbs >>> 16); trialKey[30] = (byte) (lsbs >>> 8); trialKey[31] = (byte) (lsbs ); cipher.setKey(trialKey); cipher.encrypt(plainText, trialciphertext); if (match(ciphertext, trialciphertext)){ foundKey = new byte[32] System.arraycopy(trialkey, 0, foundkey, 0, 32); } if (foundKey != null) System.out.println(Hex.toString(foundkey)); Would also like to extend the size of N beyond 30 bits