Sorry, I forgot to specify my workload :
./imaptest-1.0.0/src/imaptest secs=300 clients=800 seed=123 userfile=/tmp/userlist_imaptest no_tracking mbox=/tmp/dovecot-crlf
with dovecot-crlf : http://www.dovecot.org/tmp/dovecot-crlf
Kind regards, rub
2015-01-14 15:50 GMT+01:00 rub zorghy gdrub13@gmail.com:
Hi,
The measurements were created under the following conditions :
• operating system : Red Hat Enterprise Linux Server release 6.6 (Santiago) kernel in version 2.6.32-504.el6.x86_64 • virtual server (VMware) with an Intel(R) Xeon(R) 4vCPU E5649 x 2,526 GHz (2 cores per virtual socket) and 4 GB RAM • 7,200 RPM SATA 1TB (FC SAN IBM System Storage N3400) • all file systems had been formated in ext4 • Dovecot v.2.0.9 • imaptest v.1.0.0 • LDAP 389 Directory Server v.1.2.2 • free pagecache, dentries and inodes: echo 3 > /proc/sys/vm/drop_caches and services restarted before the test
"Tuning" based upon the book « *Dovecot: POP3/IMAP servers for enterprises and ISPs* » Peer Heinlein :
---%<------------------------------------------
imap {} (10-master.conf). service imap { [...] process_limit = 8192 [...] }
auth {} service auth { [...] client_limit = 9516 [...] }
process_min_avail = 16
---%<------------------------------------------
mail_max_userip_connections = 100
The test results shows :
Logi List Stat Sele Fetc Fet2 Stor Dele Expu Appe Logo 100% 50% 50% 100% 100% 100% 50% 100% 100% 100% 100% 30% 5% 324 142 128 263 299 405 92 153 320 336 335 800/800 212 92 92 182 166 233 59 87 178 201 202 800/800
- 4051 stalled for 16 secs in command: 8 APPEND "INBOX" () {3251+} 315 160 134 300 275 400 85 136 300 317 317 800/800
- 4051 stalled for 17 secs in command: 8 APPEND "INBOX" () {3251+} 247 130 163 292 251 366 99 115 239 240 240 800/800 222 122 117 234 279 383 112 147 263 222 222 800/800 188 93 94 177 168 232 65 87 162 196 195 800/800 290 126 119 260 262 383 102 126 283 282 283 800/800 240 125 107 243 259 353 102 130 253 256 256 800/800 276 158 144 291 280 400 101 149 266 261 261 800/800 254 95 89 189 235 313 78 118 243 253 253 800/800 82 829 926 861 722 593 455 628 1695 2252 4 ms/cmd avg
Logi List Stat Sele Fetc Fet2 Stor Dele Expu Appe Logo 100% 50% 50% 100% 100% 100% 50% 100% 100% 100% 100% 30% 5% 148 59 55 126 119 174 43 64 130 147 147 800/800
- 8770 stalled for 16 secs in command: 4 FETCH 1:2 (INTERNALDATE UID INTERNALDATE) 261 128 139 247 252 369 101 125 258 269 269 800/800 302 159 158 307 282 398 109 126 296 298 297 800/800 234 96 118 239 270 369 98 138 357 428 427 800/800 361 213 220 450 383 596 151 197 240 165 166 800/800 192 93 92 206 256 352 104 118 297 289 290 800/800
- 9806 stalled for 16 secs in command: 4 FETCH 1 (RFC822.SIZE) 245 85 72 162 128 190 52 73 157 174 173 800/800
- 9806 stalled for 17 secs in command: 4 FETCH 1 (RFC822.SIZE) 294 152 165 348 360 526 136 179 313 308 309 800/800 272 87 79 174 164 229 54 85 218 230 230 800/800 218 152 152 310 325 467 117 152 265 236 221 798/800 119 830 852 859 707 566 511 702 1672 2161 6 ms/cmd avg
I have three questions :
how to interpret each row except the last one ?
how to explain the stalled commands ?
why benchmark shows too few read disk I/O ?
Thanks so much,
rub