1. Introduction 2. What Bonnie Tries To Do 3. Result Listing - Introduction 4. Comprehensive Listing Of Results 5. Listing Of "Quality" Results 6. Ranking Of Systems By I/O Category
1. INTRODUCTION
A few years ago, I wrote a Unix filesystem benchmark named Bonnie, which I casually disseminated by mentioning it in this group. It got used in a couple of magazines here and there, and now floats about the internet. I get a couple of requests a week for it; I always send it out with a request for a brief report of results. About 10% of the people send me some results.
I've been feeling guilty, since these people cared enough to share their results with me, and the least I can do is pass them on to the rest of you.
By the way, this is copyright (c) Tim Bray, 1993, in case I want to write it up for Usenix or some such.
2. WHAT BONNIE TRIES TO DO
The source code for Bonnie (a single file of K&R C) starts with 69 lines of comment verbiage explaining what it's all about. To summarize: Bonnie tries very hard to measure the performance, under stress, of certain aspects of Unix filesystem implementations. Bonnie does not measure disk speed. Bonnie does not (by design) measure cache effectiveness. Bonnie tries to find out how fast a Unix filesystem can write a file with putc(3), write(2), and read it with getc(3) and read(2), how fast it can update blocks in a fie with read(2) and write(2), and how many times a second it can lseek(2). Bonnie attempts to defeat caching only by brute force, i.e. by working with a file many times the size of available RAM. There is lots of room for argument as to what is measured and what is IMPORTANT to measure, but some years after Bonnie's birth, I still maintain (with heavy acknowledged bias) that there really isn't a better way to find out how fast the primitive primitives operate when the cache has maxed out and you really have to move bytes back and forth. Which, with any significant big production database application, is the case much of the time.
Send me mail (tbray@watsol.uwaterloo.edu) and I'll send you Bonnie.
3. RESULT LISTING - INTRODUCTION
The following is truly a mixed bag. There are two overwhelming problems:
(1) most of the computers in it are obsolete in the sense that you can't buy them any more (but this will always be true), and (2) in most cases, the file sizes are NOT a substantial multiple of RAM size, so the results are a low-quality measurement of cache effectiveness, rather than a high-quality measurement of byte throughput.
The listing is followed by a section that provides all the information I was given as to what the test environment was. In many cases, this is insufficient information to draw real conclusions. However, there is some really fascinating wheat among all the chaff. In the later sections, I extract the subset of results where I thought the file/RAM ration was high enough to make the results quantitatively interesting.
Some of the ones with problems are still interesting. For example, look at the Cray results; despite a valiant effort by the person who did this work, I don't think it actually measured much about real I/O, but it did say quite a bit about what Crays are like.
I have intentionally not attributed any of the people who did the work here. Some of them are me. Some of them asked not to be named, and some of them might get in trouble if a management type at their company got upset at insufficient levels of release authorization etc. etc. However, Bonnie and I would like to say a heartfelt "THANKS" to all of you who took the time.
THE AUTHOR WISHES TO GO ON RECORD THAT HE FEELS THIS IS RELATIVELY LOW-QUALITY INFORMATION, AND THAT AS PRESENTED IT PROBABLY MISREPRESENTS IN SOME FASHION THE PERFORMANCE OF EVERY MANUFACTURER NAMED HEREIN. FOR HEAVEN'S SAKE, DON'T USE THIS AS A GUIDE IN A PROCUREMENT EXERCISE, AND IF YOU DO, DON'T SAY I TOLD YOU TO!
On the other hand, if you'd like to run Bonnie on your latest & greatest and shoot me some better-looking numbers, well hey.
4. COMPREHENSIVE LISTING OF RESULTS
(more or less alphabetical by manufacturer - commentary at end of list)
-------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
4M i386 25 130 90.5 199 26.1 179 46.1 121 93.5 400 59.7 10.5 23.8
A3000 64 51 31.5 53 7.6 41 10.1 137 72.5 264 42.3 14.9 24.9
MacIIfx 10 68 20.4 66 2.7 83 3.5 185 49.6 209 4.5 14.5 4.0
BBGL 100 534 65.2 1351 23.4 648 21.8 668 85.1 2586 45.1 91.7 17.7 BBGL 200 534 65.2 1367 23.8 507 18.5 604 77.7 1851 35.0 58.9 12.7 BBGL 750 534 65.7 1236 22.5 419 17.5 564 74.3 1534 32.8 35.0 8.3
Y-MPE1 50 909 99.4 10929 88.8 6545 32.3 747 83.1 18923 71. 8647.1 388+ CRAY_YMPD 50 601 72.1 20875 86.2 6871 56.6 709 86.4 21080 86. 5014.1 386+ Y-MP_C90 50 1137 99.9 16876 33.7 26171 99 1112 99.7 52536 99. 6438.1 798 S-MP 50 1141 95.6 1468 12.6 1485 16.9 1227 98.6 17490 99. 1508.4 98.2 Y-MP_EL 50 208 97 14407100. 2419 38 194 96. 14597 98 1472.4 393+
DEC3100 600 356 97.8 352 14.3 204 8.3 382 95.5 556 13.8 23.2 9.0 DS5K200cx 100 361 45.0 370 4.8 156 5.0 665 94.3 1154 23.6 30.0 6.3 5820 100 415 93.2 416 25.3 199 21.4 417 93.0 424 26.0 28.3 16.3 5820 100 406 97.7 419 26.9 201 21.1 407 93.6 426 25.5 27.0 15.9 brillig 8 124 90.6 366 21.6 202 13.1 210 97.3 483 13.6 39.9 12.9 brillig 8 138 92.3 355 20.7 205 13.2 199 98.3 463 13.5 41.4 12.9 brillig.old 8 124 90.6 366 21.6 202 13.1 210 97.3 483 13.6 39.9 12.9 brillig.new 8 138 92.3 355 20.7 205 13.2 199 98.3 463 13.5 41.4 12.9 gyre 8 59 97.2 284 50.8 99 24.7 75 97.3 320 36.2 15.0 20.6 VAX 8650 200 208 89.0 232 8.3 143 7.4 197 65.0 373 8.6 17.3 4.6 Alpha3000 950 1552 47.4 1571 7.2 917 5.2 2215 67.3 2238 5.6 45.4 2.0
HP710 330 1302 84.1 1299 15.4 90 5.2 1207 84.9 1460 13.2 29.6 4.7 HP9ks730 100 1335 56.5 1232 7.3 79 1.9 1430 67.9 1464 5.7 32.2 2.5 hp9ks750 100 2035 82.4 1710 8.9 465 3.9 2055 95.8 2087 8.2 22.6 2.0 HP720 50 1350 74.1 1263 8.1 128 4.7 1020 64.7 1398 6.7 36.5 4.1
IBM 520 25 481 98.4 6422 99.8 6016 99.6 331 59.7 16960100.01247.8 99.0 IBM 520 35 481 98.6 701 11.6 5047 87.5 499 89.0 16583 99.91230.1 99.6 IBM 520 40 484 98.6 780 12.4 6103 99.5 541 98.5 16732100.01244.1 99.0 IBM 520 45 483 98.5 1099 16.9 985 16.8 555 99.2 16131 98.41217.9 98.7 IBM 520 46 482 98.5 6373 98.9 904 16.3 450 82.2 15964 97.91213.4 99.5 IBM 520 47 482 98.5 2756 43.2 910 16.7 429 77.0 10915 70.8 732.8 63.4 IBM 520 50 473 97.4 1119 17.4 717 13.7 516 96.6 1640 19.3 302.3 29.6 IBM 520 55 482 98.5 1775 27.6 433 9.3 435 83.2 1432 16.9 142.0 17.1 IBM 520 60 482 98.6 2481 39.0 423 9.8 404 78.2 1430 17.2 109.5 14.9 IBM 520 65 480 98.4 1091 17.8 456 10.8 504 97.9 1433 18.4 82.6 12.6 IBM 520 75 480 98.4 1211 20.2 418 9.7 485 93.3 1452 18.4 62.5 10.9 IBM 520 100 473 97.4 1514 24.5 413 9.3 449 86.7 1442 17.1 42.1 8.8 RS6K 100 576 97.5 1360 20.3 603 13.0 496 96.2 1799 21.2 26.9 8.1
M2000 125 445 47.0 466 5.8 308 6.2 838 95.5 1424 13.3 32.6 3.5 M2000 125 565 60.1 464 5.8 273 5.1 795 94.0 1391 16.6 32.1 3.7 M2000 125 401 44.2 445 5.4 311 6.3 392 45.0 926 9.1 25.2 3.4
NxtCube 125 241 94.7 347 39.3 253 31.7 246 94.8 772 49.6 27.7 20.8 NxtStn 150 508 95.4 735 55.7 446 42.2 473 88.6 995 50.9 32.1 17.7
Sequent 95 140 97.8 1017 66.2 251 17.7 122 95.7 713 34.9 31.0 15.9
-------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU SGI4D25 1 69 49.5 150 7.1 99 12.8 146 90.3 3657 100.0149.9 60.4 SGI4D25 2 64 46.7 180 9.9 100 13.3 104 76.4 3413 93.3 135.9 54.5 SGI4D25 3 62 45.5 180 10.1 95 11.6 104 77.3 3072 84.0 140.1 56.3 SGI4D25 4 60 43.1 152 8.0 85 10.3 105 77.2 3103 84.8 136.2 54.5 SGI4D25 5 62 45.1 105 6.1 93 12.0 104 77.5 3121 85.4 169.8 68.6 SGI4D25 6 60 43.4 142 7.4 89 9.5 93 69.0 1359 39.8 138.1 56.4 SGI4D25 7 59 43.1 176 10.6 89 11.5 101 74.6 183 8.9 66.3 32.1 SGI4D25 8 57 41.0 119 7.2 90 9.7 90 66.5 190 9.8 47.8 27.3 SGI4D25 9 54 39.0 172 12.4 93 9.8 98 72.8 188 9.3 37.1 23.9 SGI4D25 10 54 39.1 172 13.2 94 9.9 85 62.7 190 9.8 32.6 22.4 SGI4D25 15 53 38.4 115 6.7 67 6.6 77 55.3 135 6.1 13.4 11.1 SGI4D25 20 51 37.1 131 8.9 69 6.7 76 54.4 136 6.0 9.7 8.6 SGI4D25 25 50 36.5 110 7.9 65 6.4 80 56.4 134 5.9 8.2 7.6 SGI4D25 30 50 36.2 143 12.5 65 7.1 80 57.2 134 5.9 7.7 7.3 SGI4D25 35 49 35.7 154 14.4 65 7.1 80 57.4 132 5.8 7.3 6.9 SGI4D25 40 49 35.3 156 14.9 66 6.8 81 57.6 133 6.0 6.6 6.4 SGI4D25 45 46 33.3 142 8.0 69 7.1 82 58.0 134 6.0 6.3 6.1 SGI4D25 50 45 32.9 149 8.8 66 6.6 80 57.8 133 5.9 6.3 6.1 SGI4D25 75 40 29.3 121 10.9 63 6.7 81 57.3 130 5.6 5.4 5.3 SGI4D25 100 36 26.6 139 13.4 61 6.4 80 57.1 127 5.7 5.1 5.0 SGI4D25 125 33 24.3 123 10.6 59 5.7 79 57.0 124 5.5 4.8 4.8 SGI4D 100 454 91.1 1001 24.2 359 15.8 444 90.7 559 15.6 32.4 9.6
sony3870 34 898 97.9 1224 12.5 5399 99.5 874 99.2 7786 100 470.2 54.7 sony3870 100 885 97.3 1634 19.0 451 8.5 447 50.8 1015 13.9 47.5 6.3 sony3870 180 879 97.1 1220 18.8 449 9.2 488 55.6 1015 14.0 33.2 4.7 sony3870 600 878 96.5 971 24.2 428 8.1 522 59.2 946 13.1 29.1 4.4
4/75 100 914 98.8 2524 44.6 612 24.1 690 93.8 3048 69.5 73.6 15.5 4/75 200 909 98.8 2492 44.9 535 21.9 705 95.9 3250 73.5 61.3 13.0 4/75 250 891 95.8 1480 26.1 547 21.0 696 93.6 2139 45.0 43.4 9.3 4/490 100 676 97.4 2047 49.0 780 23.3 587 90.2 1980 30.5 60.0 9.8 SSrv2 1 776 91.1 1268 26.0 1339 37.9 825100.0 12427 97.11037.5 98.8 SSrv2 2 822 97.2 1248 27.4 1248 33.5 827 99.8 12506 97.7 730.4 69.2 SSrv2 3 828 98.0 1240 28.3 1160 31.4 828100.1 12565 98.2 668.5 63.5 SSrv2 4 831 98.5 1166 25.1 1227 46.4 771100.0 5962100.4 618.5 70.4 SSrv2 5 835 98.9 1204 26.8 1170 42.5 770100.0 5932 99.7 593.6 75.4 SSrv2 6 818 96.9 1207 26.7 1206 45.0 770100.0 5964100.0 579.9 78.6 SSrv2 7 828 98.0 1209 26.5 1198 45.1 756 98.2 5965 99.9 586.5 81.1 SSrv2 8 839 99.1 1211 27.1 1178 43.9 770 99.9 5935100.0 558.7 79.6 SSrv2 9 839 99.3 1191 26.5 1209 44.2 769100.0 5933 99.8 542.8 80.2 SSrv2 10 838 99.4 1180 26.5 1204 45.0 769100.0 5929100.2 563.1 83.1 SSrv2 15 829 98.4 1201 27.1 1200 45.2 762 99.0 5935100.1 542.8 85.2 SSrv2 20 827 98.3 1193 27.6 910 33.5 749 98.5 4296 69.7 171.1 28.3 SSrv2 30 816 97.5 1152 26.6 507 19.6 679 91.7 2168 41.4 82.8 14.5 SSrv2 40 817 97.8 1180 27.8 524 20.1 693 93.9 2183 42.8 64.1 11.9 SSrv2 50 814 97.5 1175 27.5 449 17.1 679 92.8 1853 41.3 51.6 10.1 SSrv2 60 819 98.1 1177 28.0 430 16.8 674 93.2 1610 39.8 45.6 9.5 SSrv2 70 815 97.7 1180 27.9 402 15.9 669 93.0 1482 34.4 41.5 8.6 SSrv2 80 820 98.2 1176 28.2 397 15.5 665 92.9 1409 32.1 39.7 8.5 SSrv2 90 817 98.0 1181 28.0 390 15.5 663 92.8 1364 33.5 38.3 7.9 SSrv2 100 820 98.2 1178 28.2 351 14.1 665 93.5 1297 30.8 36.7 7.7 SSrv2 150 811 97.6 1170 27.7 337 13.4 649 91.9 1282 31.1 33.0 7.6 SSrv2 200 817 97.9 1169 28.1 271 10.6 655 93.1 1284 31.5 32.0 6.7 SS490 1 810 96.6 3666 64.4 1860 43.6 692100.1 14922102.0 586.5 33.4 SS490 2 832 98.8 4378 79.1 4025100.2 693 99.9 15183103.8 462.7 26.8 SS490 3 809 95.9 4309 81.4 3697 93.9 693100.1 15199 99.0 637.2 36.3 SS490 4 836 99.6 4402 87.1 3397100.4 665 97.8 8894 99.9 588.1 37.2 SS490 5 837 99.6 4276 89.4 3337 99.7 670 98.7 8834100.1 570.6 38.7 SS490 6 835 99.6 4243 90.5 3302100.0 669 98.7 8686100.4 570.8 40.5 SS490 7 836 99.6 4183 92.2 3290101.0 677100.0 8693100.7 531.7 39.7 SS490 8 827 98.5 4130 92.8 3235 99.9 677100.0 8697 99.8 517.0 38.6 SS490 9 838 99.8 4069 94.5 3179100.0 674 99.1 8676 99.8 623.5 48.3 SS490 10 827 98.6 3977 95.2 3141100.0 677100.0 8683100.1 562.0 43.4 SS490 15 838 99.9 3641 96.5 2982100.0 678100.0 8671 99.9 579.3 46.2 SS490 20 830 99.0 2486 64.6 2872100.0 685 99.7 8648100.1 565.3 46.5 SS490 30 832 99.3 2748 74.1 1812 55.7 530 78.3 2471 32.7 224.3 19.9 SS490 40 828 99.1 2351 61.8 856 24.6 611 92.6 2171 34.3 93.8 9.3 SS490 50 827 99.0 2278 61.8 952 27.9 625 95.1 2138 32.9 66.6 7.0 SS490 60 820 98.2 2224 58.6 936 27.3 621 94.6 2110 32.2 56.3 5.8 SS490 70 824 98.6 2323 61.9 935 27.2 618 94.4 2179 32.8 50.0 5.6 SS490 80 825 98.8 2394 64.0 934 27.7 615 93.7 2165 33.1 46.0 5.2 SS490 90 826 98.9 2316 60.6 928 27.2 621 94.7 2192 33.9 43.2 5.5 SS490 100 822 98.7 2257 60.5 936 27.4 626 95.5 2174 33.3 40.9 4.7 SS490 150 822 98.7 2251 60.2 936 27.5 626 95.5 2182 33.4 35.7 4.6 SS490 200 824 98.8 2243 56.1 934 27.2 627 95.7 2167 33.7 33.5 4.4 SS1+ 100 362 83.3 743 39.0 247 18.3 310 82.3 857 45.0 31.6 13.0 Sun4/260 95 270 90.4 645 26.8 305 20.0 278 94.3 811 28.8 37.3 15.4
Notes (from those who ran the benchmarks, unedited):
4M i386: 25 Mhz, Running Interactive Unix, 4M RAM. AMI motherboard, ordinary Seagate disk.
A3000: An Amiga 3000, running a very early version of the Amiga Unix product. I have heard that they have considerably souped up the I/O, which is clearly a good idea.
MacIIfx: Apple MacIntosh IIfx, A/UX 2.0, Internal hard drive.
BBGL: A 40 Mhz SPARC in a Black Box with a Green Label, 64 Mb RAM, HP 2GB SCSI disk on unspecified controller, SunOS 4.1.1. Note the monster sequential read value due to some of Larry McVoy's code dropping into optimal mode with enough RAM to spare.
Cray notes: I ran Bonnie on each system and watched for the asymptotic performacne. One all systems it was reached by the time I run the 50 Mbyte tests case.
Note that for the Cray Research systems, sinlgle spindle disk transfers typically go at 20 Mbyte/s. The C90 results included disk caching to a solid state disk.
(I don't believe some of the above. -- T. Bray)
Y-MPE1: CRAY Y-MP w/ model E IOS, UNICOS 7.C, /tmp DD60 Disk CRAY_YMPD: CRAY Y-MP model D, UNICOS 7.0, /tmp DD60 Y-MP_C90: CRAY Y-MP C90 w/ model E IOS, ldcaching, UNICOS 8.0.0, 5700 units (20 Mbytes) ldcache S-MP: CRAY S-MP, CRS CRSOS 5.0.4, Disk ? (/tmp) Y-MP_EL: CRAY Y-MP EL, UNICOS 6.1.6, Disk ? (/tmp, striped device)
DEC3100: Really old DECstation3100, 24Mb ram, Ultrix 3.1, external Fuji 800M SCSI.
DS5K200cx: DECstation 500-200/CX, 16MB RAM, 90MB swap, Ultrix4.0, RZ57 1GB SCSI disk, on 250 Mb partition 'f', system not particularly idle.
5820: DEC 5820, >64M RAM, HSC70 DEC's fastest (at the time) disks.
brillig, gyre: Brillig is a VAX 8600 with an Emulex SC780 controller and Eagles, gyre an 8250 with CDC 9771 disks on an SC41/MS Unibus MSCP. In both cases, an oil-stained Chris Torek is hunched under the machine tweaking the 4.3/Tahoe/Reno/ad-hoc kernel.
VAX 8650: 32M, fairly vanilla 4.3bsd, massbus/eagles.
Alpha3000: A DEC AXP 3000, OSF1 "1.2 10 alpha", 96M memory, rz26 disk, just-newfs'ed with big (16384) blocksize.
HP 710: Little info; I ran across it in the office just long enough to run Bonnie, then it vanished. It was all stock HP stuff as of spring 1992. It was pretty well empty so the 330 Mb filesize probably will tell an HP expert what kind of a disk that was. There was no visible external disk box.
HP720, OS ?, Disk ? (/tmp)
HP9ks730, hp9ks750: a 730 with untunde file I/O system and an 730 will tuned file I/O system; both machines have identical CPUs running at the same clock speed (66 MHz).
IBM 520: RS/6000 mod 520, AIX 3.1. A 320MB SCSI (12.5ms access time, 2MB/s transfer rate) and a 670 SCSI (18 ms access time, 1.875MB/s transfer rate). The logical volume group includes both physical volumes, but I guess it would be the 670 because the 320 was pretty filled up long before I extended my /usr filesystem, which is the one I used. Filesystem size is 503,808KB, 77% full. Compiler is xlc with -O flag only.
RS6K: An anonymous RS/6000 sitting in a "deskside" style box, 32Mb, some SCSI.
M2000: A MIPS M/2000 running UMIPS in early 1991. 64Mb RAM, MIPS disks.
NxtCube, NxtStation: NeXT Cube and Station respectively. All right out of the box as shipped by NeXT; both monochrome, if it matters. Internal disk.
Sequent: A Symmetry with 10 386's. Gobs and gobs of memory, Sequent SCSI disks.
SGI4D25: Silicon Graphics Personal Iris 4D/25 Turbo with 1 20 MHZ IP6 Processor, FPU: MIPS R2010A/R3010 VLSI Floating Point Chip Revision: 2.0, CPU: MIPS R2000A/R3000 Processor Chip Revision: 2.0, Data cache size: 32 Kbytes, Instruction cache size: 64 Kbytes, Main memory size: 16 Mbytes, Integral Ethernet controller: Version 0, Graphics board: GR1.2 Bit-plane, Z-buffer, Turbo options installed, Integral SCSI controller 0: Version WD33C93A, Tape drive: unit 7 on SCSI controller 0: QIC 150, Disk drive: unit 1 on SCSI controller 0
sony3870: A Sony NEWS 3870, 64Mb RAM, 1.2Gb harddisk, mips 3000 (and a mips 3010 co-processor) at 25MHz, 68030 I/O-processor at 25MHz, NEWS-OS 4.0R
4/75: Sun 4/75 with 2 internal 32 M ram. The 100M & 200M runs are on an elite SCSI. The 250 added an Imprimis Wren 7 SCSI.
4/490: Sun 4/490, 64M RAM, older 3 MB/sec IPI drives.
SSrv2: Sun SPARCserver 2 with 32MB RAM, a 207MB internal disk, a 669MB external disk (both SCSI)--we tested the latter--running SunOS 4.1.1, SPARCcompilers 1.0beta2.
SS490: Sun SPARCserver 490 with 32MB RAM, a 1GB external IPI-2 disk (which we tested), running SunOS 4.1, SPARCcompilers 1.0beta2.
SS1+: SPARCstation 1+, no further info.
Sun4/260: Just about the first SPARC ever shipped. 32M memory, SunOS 4.1.something, external Fuji 1.2G disks.
5. LISTING OF "QUALITY" RESULTS
Some of the results above are much more equal than others. Those that follow used files that are at least big enough that there was a good chance that lots of real I/O was being done.
-------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 4M-i386 25 130 90.5 199 26.1 179 46.1 121 93.5 400 59.7 10.5 23.8 A3000 64 51 31.5 53 7.6 41 10.1 137 72.5 264 42.3 14.9 24.9 BBGL 750 534 65.7 1236 22.5 419 17.5 564 74.3 1534 32.8 35.0 8.3 DEC3100 600 356 97.8 352 14.3 204 8.3 382 95.5 556 13.8 23.2 9.0 DS5K200cx 100 361 45.0 370 4.8 156 5.0 665 94.3 1154 23.6 30.0 6.3 VAX-8650 200 208 89.0 232 8.3 143 7.4 197 65.0 373 8.6 17.3 4.6 Alpha3000 950 1552 47.4 1571 7.2 917 5.2 2215 67.3 2238 5.6 45.4 2.0 HP710 330 1302 84.1 1299 15.4 90 5.2 1207 84.9 1460 13.2 29.6 4.7 RS6K 100 576 97.5 1360 20.3 603 13.0 496 96.2 1799 21.2 26.9 8.1 M2000 125 445 47.0 466 5.8 308 6.2 838 95.5 1424 13.3 32.6 3.5 NxtCube 125 241 94.7 347 39.3 253 31.7 246 94.8 772 49.6 27.7 20.8 NxtStn 150 508 95.4 735 55.7 446 42.2 473 88.6 995 50.9 32.1 17.7 Sequent 95 140 97.8 1017 66.2 251 17.7 122 95.7 713 34.9 31.0 15.9 SGI4D 100 454 91.1 1001 24.2 359 15.8 444 90.7 559 15.6 32.4 9.6 sony3870 600 878 96.5 971 24.2 428 8.1 522 59.2 946 13.1 29.1 4.4 SSrv2 200 817 97.9 1169 28.1 271 10.6 655 93.1 1284 31.5 32.0 6.7 SS490 200 824 98.8 2243 56.1 934 27.2 627 95.7 2167 33.7 33.5 4.4 SS1+ 100 362 83.3 743 39.0 247 18.3 310 82.3 857 45.0 31.6 13.0
Qualitatively, one notes that: The DEC Alpha is an I/O monster. The HP 710 is amazingly fast for its size, maybe the price-performance champ. Likewise the NeXTs, R.I.P. The big Suns are excellent given that they are now 1.5 generations old; would like to see numbers for a big Sparc 10; also the new RS/6000s, whatever they're called. Also I suspect that a fast 486 or Pentium running SCO with a superduper disk controller might really embarrass some of the big boys. Machines that can seek more than 35 times a second on big files are VERRRRRY rare.
6. RANKING OF SYSTEMS BY I/O CATEGORY
In descending order of:
Per-char write speed: Alpha3000 HP710 sony3870 SS490 SSrv2 RS6K BBGL NxtStn SGI4D M2000 SS1+ DS5K200cx DEC3100 NxtCube VAX-8650 Sequent 4M-i386 A3000
Block write speed: SS490 Alpha3000 RS6K HP710 BBGL SSrv2 Sequent SGI4D sony3870 SS1+ NxtStn M2000 DS5K200cx DEC3100 NxtCube VAX-8650 4M-i386 A3000
In-place update: SS490 Alpha3000 RS6K NxtStn sony3870 BBGL SGI4D M2000 SSrv2 NxtCube Sequent SS1+ DEC3100 4M-i386 DS5K200cx VAX-8650 HP710 A3000
Per-char read speed: Alpha3000 HP710 M2000 DS5K200cx SSrv2 SS490 BBGL sony3870 RS6K NxtStn SGI4D DEC3100 SS1+ NxtCube VAX-8650 A3000 Sequent 4M-i386
Block read speed: Alpha3000 SS490 RS6K BBGL HP710 M2000 SSrv2 DS5K200cx NxtStn sony3870 SS1+ NxtCube Sequent SGI4D DEC3100 4M-i386 VAX-8650 A3000
Seeks/sec: Alpha3000 BBGL SS490 M2000 SGI4D NxtStn SSrv2 SS1+ Sequent DS5K200cx HP710 sony3870 NxtCube RS6K DEC3100 VAX-8650 A3000 4M-i386