Thanks for the benchmarking. Let's see if the impact could be minimized.
Xuelei On 5/15/2019 10:30 AM, Martin Balao wrote:
Hi Xuelei, I've developed a JMH benchmark to measure the impact of Webrev.00 for 8223482. This benchmark measures TLS renegotiations on FIPS (SunPKCS11 + NSS + FIPS) and NON-FIPS (all security providers enabled) TLS 1.2 scenarios. WITHOUT 8223482 FIX ============================================================ Benchmark (testMode) Mode Cnt Score Error Units SupportedCiphersuites.test_TLS12Communication FIPS thrpt 10 199.620 ± 3.795 ops/s SupportedCiphersuites.test_TLS12Communication NON_FIPS thrpt 10 592.222 ± 15.944 ops/s WITH 8223482 FIX (Webrev.00) ============================================================ Benchmark (testMode) Mode Cnt Score Error Units SupportedCiphersuites.test_TLS12Communication FIPS thrpt 10 202.215 ± 3.343 ops/s SupportedCiphersuites.test_TLS12Communication NON_FIPS thrpt 10 428.161 ± 11.767 ops/s More information: * Full results: http://cr.openjdk.java.net/~mbalao/webrevs/8223482/benchmark_results_v0 * Benchmark code: http://cr.openjdk.java.net/~mbalao/webrevs/8223482/ciphersuites_benchmark_v0.tar.gz There is a performance penalty of ~28% in NON-FIPS mode. I think I can improve this number, with some trade-offs. Keep you posted. Thanks, Martin.-
