cmd/raidz_test/raidz_bench.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or https://opensource.org/licenses/CDDL-1.0.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
  24  */
  25
  26 #include <sys/zfs_context.h>
  27 #include <sys/time.h>
  28 #include <sys/wait.h>
  29 #include <sys/zio.h>
  30 #include <sys/vdev_raidz.h>
  31 #include <sys/vdev_raidz_impl.h>
  32 #include <stdio.h>
  33
  34 #include "raidz_test.h"
  35
  36 #define GEN_BENCH_MEMORY        (((uint64_t)1ULL)<<32)
  37 #define REC_BENCH_MEMORY        (((uint64_t)1ULL)<<29)
  38 #define BENCH_ASHIFT            12
  39 #define MIN_CS_SHIFT            BENCH_ASHIFT
  40 #define MAX_CS_SHIFT            SPA_MAXBLOCKSHIFT
  41
  42 static zio_t zio_bench;
  43 static raidz_map_t *rm_bench;
  44 static size_t max_data_size = SPA_MAXBLOCKSIZE;
  45
  46 static void
  47 bench_init_raidz_map(void)
  48 {
  49         zio_bench.io_offset = 0;
  50         zio_bench.io_size = max_data_size;
  51
  52         /*
  53          * To permit larger column sizes these have to be done
  54          * allocated using aligned alloc instead of zio_abd_buf_alloc
  55          */
  56         zio_bench.io_abd = raidz_alloc(max_data_size);
  57
  58         init_zio_abd(&zio_bench);
  59 }
  60
  61 static void
  62 bench_fini_raidz_maps(void)
  63 {
  64         /* tear down golden zio */
  65         raidz_free(zio_bench.io_abd, max_data_size);
  66         memset(&zio_bench, 0, sizeof (zio_t));
  67 }
  68
  69 static inline void
  70 run_gen_bench_impl(const char *impl)
  71 {
  72         int fn, ncols;
  73         uint64_t ds, iter_cnt, iter, disksize;
  74         hrtime_t start;
  75         double elapsed, d_bw;
  76
  77         /* Benchmark generate functions */
  78         for (fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
  79
  80                 for (ds = MIN_CS_SHIFT; ds <= MAX_CS_SHIFT; ds++) {
  81                         /* create suitable raidz_map */
  82                         ncols = rto_opts.rto_dcols + fn + 1;
  83                         zio_bench.io_size = 1ULL << ds;
  84
  85                         if (rto_opts.rto_expand) {
  86                                 rm_bench = vdev_raidz_map_alloc_expanded(
  87                                     zio_bench.io_abd,
  88                                     zio_bench.io_size, zio_bench.io_offset,
  89                                     rto_opts.rto_ashift, ncols+1, ncols,
  90                                     fn+1, rto_opts.rto_expand_offset);
  91                         } else {
  92                                 rm_bench = vdev_raidz_map_alloc(&zio_bench,
  93                                     BENCH_ASHIFT, ncols, fn+1);
  94                         }
  95
  96                         /* estimate iteration count */
  97                         iter_cnt = GEN_BENCH_MEMORY;
  98                         iter_cnt /= zio_bench.io_size;
  99
 100                         start = gethrtime();
 101                         for (iter = 0; iter < iter_cnt; iter++)
 102                                 vdev_raidz_generate_parity(rm_bench);
 103                         elapsed = NSEC2SEC((double)(gethrtime() - start));
 104
 105                         disksize = (1ULL << ds) / rto_opts.rto_dcols;
 106                         d_bw = (double)iter_cnt * (double)disksize;
 107                         d_bw /= (1024.0 * 1024.0 * elapsed);
 108
 109                         LOG(D_ALL, "%10s, %8s, %zu, %10llu, %lf, %lf, %u\n",
 110                             impl,
 111                             raidz_gen_name[fn],
 112                             rto_opts.rto_dcols,
 113                             (1ULL<<ds),
 114                             d_bw,
 115                             d_bw * (double)(ncols),
 116                             (unsigned)iter_cnt);
 117
 118                         vdev_raidz_map_free(rm_bench);
 119                 }
 120         }
 121 }
 122
 123 static void
 124 run_gen_bench(void)
 125 {
 126         char **impl_name;
 127
 128         LOG(D_INFO, DBLSEP "\nBenchmarking parity generation...\n\n");
 129         LOG(D_ALL, "impl, math, dcols, iosize, disk_bw, total_bw, iter\n");
 130
 131         for (impl_name = (char **)raidz_impl_names; *impl_name != NULL;
 132             impl_name++) {
 133
 134                 if (vdev_raidz_impl_set(*impl_name) != 0)
 135                         continue;
 136
 137                 run_gen_bench_impl(*impl_name);
 138         }
 139 }
 140
 141 static void
 142 run_rec_bench_impl(const char *impl)
 143 {
 144         int fn, ncols, nbad;
 145         uint64_t ds, iter_cnt, iter, disksize;
 146         hrtime_t start;
 147         double elapsed, d_bw;
 148         static const int tgt[7][3] = {
 149                 {1, 2, 3},      /* rec_p:   bad QR & D[0]       */
 150                 {0, 2, 3},      /* rec_q:   bad PR & D[0]       */
 151                 {0, 1, 3},      /* rec_r:   bad PQ & D[0]       */
 152                 {2, 3, 4},      /* rec_pq:  bad R  & D[0][1]    */
 153                 {1, 3, 4},      /* rec_pr:  bad Q  & D[0][1]    */
 154                 {0, 3, 4},      /* rec_qr:  bad P  & D[0][1]    */
 155                 {3, 4, 5}       /* rec_pqr: bad    & D[0][1][2] */
 156         };
 157
 158         for (fn = 0; fn < RAIDZ_REC_NUM; fn++) {
 159                 for (ds = MIN_CS_SHIFT; ds <= MAX_CS_SHIFT; ds++) {
 160
 161                         /* create suitable raidz_map */
 162                         ncols = rto_opts.rto_dcols + PARITY_PQR;
 163                         zio_bench.io_size = 1ULL << ds;
 164
 165                         /*
 166                          * raidz block is too short to test
 167                          * the requested method
 168                          */
 169                         if (zio_bench.io_size / rto_opts.rto_dcols <
 170                             (1ULL << BENCH_ASHIFT))
 171                                 continue;
 172
 173                         if (rto_opts.rto_expand) {
 174                                 rm_bench = vdev_raidz_map_alloc_expanded(
 175                                     zio_bench.io_abd,
 176                                     zio_bench.io_size, zio_bench.io_offset,
 177                                     BENCH_ASHIFT, ncols+1, ncols,
 178                                     PARITY_PQR, rto_opts.rto_expand_offset);
 179                         } else {
 180                                 rm_bench = vdev_raidz_map_alloc(&zio_bench,
 181                                     BENCH_ASHIFT, ncols, PARITY_PQR);
 182                         }
 183
 184                         /* estimate iteration count */
 185                         iter_cnt = (REC_BENCH_MEMORY);
 186                         iter_cnt /= zio_bench.io_size;
 187
 188                         /* calculate how many bad columns there are */
 189                         nbad = MIN(3, raidz_ncols(rm_bench) -
 190                             raidz_parity(rm_bench));
 191
 192                         start = gethrtime();
 193                         for (iter = 0; iter < iter_cnt; iter++)
 194                                 vdev_raidz_reconstruct(rm_bench, tgt[fn], nbad);
 195                         elapsed = NSEC2SEC((double)(gethrtime() - start));
 196
 197                         disksize = (1ULL << ds) / rto_opts.rto_dcols;
 198                         d_bw = (double)iter_cnt * (double)(disksize);
 199                         d_bw /= (1024.0 * 1024.0 * elapsed);
 200
 201                         LOG(D_ALL, "%10s, %8s, %zu, %10llu, %lf, %lf, %u\n",
 202                             impl,
 203                             raidz_rec_name[fn],
 204                             rto_opts.rto_dcols,
 205                             (1ULL<<ds),
 206                             d_bw,
 207                             d_bw * (double)ncols,
 208                             (unsigned)iter_cnt);
 209
 210                         vdev_raidz_map_free(rm_bench);
 211                 }
 212         }
 213 }
 214
 215 static void
 216 run_rec_bench(void)
 217 {
 218         char **impl_name;
 219
 220         LOG(D_INFO, DBLSEP "\nBenchmarking data reconstruction...\n\n");
 221         LOG(D_ALL, "impl, math, dcols, iosize, disk_bw, total_bw, iter\n");
 222
 223         for (impl_name = (char **)raidz_impl_names; *impl_name != NULL;
 224             impl_name++) {
 225
 226                 if (vdev_raidz_impl_set(*impl_name) != 0)
 227                         continue;
 228
 229                 run_rec_bench_impl(*impl_name);
 230         }
 231 }
 232
 233 void
 234 run_raidz_benchmark(void)
 235 {
 236         bench_init_raidz_map();
 237
 238         run_gen_bench();
 239         run_rec_bench();
 240
 241         bench_fini_raidz_maps();
 242 }