Hi Loic, as much as I understood for now it should not happen in Jerasure, but it will happen with the default Vandermonde matrix in the Intel library. The reason is the way these matrices are constructed. But to verify this you have to make millions of iterations not only 1 because for large (k,m) there many failure scenario (chunk combinations) to test. Cheers Andreas. ________________________________________ From: Loic Dachary [loic@xxxxxxxxxxx] Sent: 03 July 2014 12:22 To: Andreas Joachim Peters Cc: Ceph Development Subject: Checking of Reed-Solomon Vandermonde parameter combinations Hi Andreas, There are some combinations of K/M parameters for which Reed-Solomon Vandermonde may not be able to recover from the loss of M chunks. Reed-Solomon Cauchy does not suffer from this problem. (I'm paraphrasing a conversation we had in private, feel free to correct me if I'm wrong). In the context of Ceph we are interested in a limited range of parameters so I ran a test with: https://github.com/ceph/ceph/blob/master/src/test/erasure-code/ceph_erasure_code_benchmark.cc#L40 for k in $(seq 2 50) ; do for m in $(seq 1 $k) ; do echo -n "k=$k m=$m " ; ./ceph_erasure_code_benchmark --plugin jerasure --parameter directory=.libs --parameter technique=reed_sol_van --parameter k=$k --parameter m=$m --erasures $m --iterations 1 --workload decode ; done ; done and it does not throw an error. It basically take a range of parameters K=2,M=2 up to K=50,M=50 and decode with M erasures for each of them. Should some of them fail ? Cheers -- Loïc Dachary, Artisan Logiciel Libre -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
