| blob | 1b50d1a0d5849bca22eb4a5e1d0982aaaffd9164 |
1 .\" $NetBSD: results.ms,v 1.2 1998/01/09 06:41:02 perry Exp $
2 .\"
3 .\" Copyright (c) 1983 The Regents of the University of California.
4 .\" All rights reserved.
5 .\"
6 .\" Redistribution and use in source and binary forms, with or without
7 .\" modification, are permitted provided that the following conditions
8 .\" are met:
9 .\" 1. Redistributions of source code must retain the above copyright
10 .\" notice, this list of conditions and the following disclaimer.
11 .\" 2. Redistributions in binary form must reproduce the above copyright
12 .\" notice, this list of conditions and the following disclaimer in the
13 .\" documentation and/or other materials provided with the distribution.
14 .\" 3. Neither the name of the University nor the names of its contributors
15 .\" may be used to endorse or promote products derived from this software
16 .\" without specific prior written permission.
17 .\"
18 .\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 .\" SUCH DAMAGE.
29 .\"
30 .\" @(#)results.ms 6.2 (Berkeley) 4/16/91
31 .\"
32 .ds RH Results
33 .NH
34 Results
35 .PP
36 The following tables indicate the results of our
37 test runs.
38 Note that each table contains results for tests run
39 on two varieties of 4.2BSD file systems.
40 The first set of results is always for a file system
41 with a basic blocking factor of eight Kilobytes and a
42 fragment size of 1 Kilobyte. The second sets of measurements
43 are for file systems with a four Kilobyte block size and a
44 one Kilobyte fragment size.
45 The values in parenthesis indicate the percentage of CPU
46 time used by the test program.
47 In the case of the two disk arm tests,
48 the value in parenthesis indicates the sum of the percentage
49 of the test programs that were run.
50 Entries of ``n. m.'' indicate this value was not measured.
51 .DS
52 .TS
53 box,center;
54 c s s s s
55 c s s s s
56 c s s s s
57 l | l s | l s
58 l | l s | l s
59 l | l l | l l
60 l | c c | c c.
61 4.2BSD File Systems Tests - \fBVAX 11/750\fR
62 =
63 Logically Sequential Transfers
64 from an \fB8K/1K\fR 4.2BSD File System (Kbytes/sec.)
65 _
66 Test Emulex SC750/Eagle UDA50/RA81
68 1 Drive 2 Drives 1 Drive 2 Drives
69 _
70 read_8192 490 (69%) 620 (96%) 310 (44%) 520 (65%)
71 write_4096 380 (99%) 370 (99%) 370 (97%) 360 (98%)
72 write_8192 470 (99%) 470 (99%) 320 (71%) 410 (83%)
73 rewrite_8192 650 (99%) 620 (99%) 310 (50%) 450 (70%)
74 =
75 .T&
76 c s s s s
77 c s s s s
78 l | l s | l s
79 l | l s | l s
80 l | l l | l l
81 l | c c | c c.
82 Logically Sequential Transfers
83 from \fB4K/1K\fR 4.2BSD File System (Kbytes/sec.)
84 _
85 Test Emulex SC750/Eagle UDA50/RA81
87 1 Drive 2 Drives 1 Drive 2 Drives
88 _
89 read_8192 300 (60%) 400 (84%) 210 (42%) 340 (77%)
90 write_4096 320 (98%) 320 (98%) 220 (67%) 290 (99%)
91 write_8192 340 (98%) 340 (99%) 220 (65%) 310 (98%)
92 rewrite_8192 450 (99%) 450 (98%) 230 (47%) 340 (78%)
93 .TE
94 .DE
95 .PP
96 Note that the rate of write operations on the VAX 11/750 are ultimately
97 CPU limited in some cases.
98 The write rates saturate the CPU at a lower bandwidth than the reads
99 because they must do disk allocation in addition to moving the data
100 from the user program to the disk.
101 The UDA50/RA81 saturates the CPU at a lower transfer rate for a given
102 operation than the SC750/Eagle because
103 it causes more memory contention with the CPU.
104 We do not know if this contention is caused by
105 the UNIBUS controller or the UDA50.
106 .PP
107 The following table reports the results of test runs on a VAX 11/780
108 with 4 Megabytes of main memory.
109 .DS
110 .TS
111 box,center;
112 c s s s s s s
113 c s s s s s s
114 c s s s s s s
115 l | l s | l s | l s
116 l | l s | l s | l s
117 l | l l | l l | l l
118 l | c c | c c | c c.
119 4.2BSD File Systems Tests - \fBVAX 11/780\fR
120 =
121 Logically Sequential Transfers
122 from an \fB8K/1K\fR 4.2BSD File System (Kbytes/sec.)
123 _
124 Test Emulex SC780/Eagle UDA50/RA81 Sys. Ind. 9900/Eagle
126 1 Drive 2 Drives 1 Drive 2 Drives 1 Drive 2 Drives
127 _
128 read_8192 560 (70%) 480 (58%) 360 (45%) 540 (72%) 340 (41%) 520 (66%)
129 write_4096 440 (98%) 440 (98%) 380 (99%) 480 (96%) 490 (96%) 440 (84%)
130 write_8192 490 (98%) 490 (98%) 220 (58%)* 480 (92%) 490 (80%) 430 (72%)
131 rewrite_8192 760 (100%) 560 (72%) 220 (50%)* 180 (52%)* 490 (60%) 520 (62%)
132 =
133 .T&
134 c s s s s s s
135 c s s s s s s
136 l | l s | l s | l s
137 l | l s | l s | l s
138 l | l l | l l | l l
139 l | c c | c c | c c.
140 Logically Sequential Transfers
141 from an \fB4K/1K\fR 4.2BSD File System (Kbytes/sec.)
142 _
143 Test Emulex SC780/Eagle UDA50/RA81 Sys. Ind. 9900/Eagle
145 1 Drive 2 Drives 1 Drive 2 Drives 1 Drive 2 Drives
146 _
147 read_8192 490 (77%) 370 (66%) n.m. n.m. 200 (31%) 370 (56%)
148 write_4096 380 (98%) 370 (98%) n.m. n.m. 200 (46%) 370 (88%)
149 write_8192 380 (99%) 370 (97%) n.m. n.m. 200 (45%) 320 (76%)
150 rewrite_8192 490 (87%) 350 (66%) n.m. n.m. 200 (31%) 300 (46%)
151 .TE
152 * the operation of the hardware was suspect during these tests.
153 .DE
154 .PP
155 The dropoff in reading and writing rates for the two drive SC780/Eagle
156 tests are probably due to the file system using insufficient
157 rotational delay for these tests.
158 We have not fully investigated these times.
159 .PP
160 The following table compares data rates on VAX 11/750s directly
161 with those of VAX 11/780s using the UDA50/RA81 storage system.
162 .DS
163 .TS
164 box,center;
165 c s s s s
166 c s s s s
167 c s s s s
168 l | l s | l s
169 l | l s | l s
170 l | l l | l l
171 l | c c | c c.
172 4.2BSD File Systems Tests - \fBDEC UDA50 - 750 vs. 780\fR
173 =
174 Logically Sequential Transfers
175 from an \fB8K/1K\fR 4.2BSD File System (Kbytes/sec.)
176 _
177 Test VAX 11/750 UNIBUS VAX 11/780 UNIBUS
179 1 Drive 2 Drives 1 Drive 2 Drives
180 _
181 read_8192 310 (44%) 520 (84%) 360 (45%) 540 (72%)
182 write_4096 370 (97%) 360 (100%) 380 (99%) 480 (96%)
183 write_8192 320 (71%) 410 (96%) 220 (58%)* 480 (92%)
184 rewrite_8192 310 (50%) 450 (80%) 220 (50%)* 180 (52%)*
185 =
186 .T&
187 c s s s s
188 c s s s s
189 l | l s | l s
190 l | l s | l s
191 l | l l | l l
192 l | c c | c c.
193 Logically Sequential Transfers
194 from an \fB4K/1K\fR 4.2BSD File System (Kbytes/sec.)
195 _
196 Test VAX 11/750 UNIBUS VAX 11/780 UNIBUS
198 1 Drive 2 Drives 1 Drive 2 Drives
199 _
200 read_8192 210 (42%) 342 (77%) n.m. n.m.
201 write_4096 215 (67%) 294 (99%) n.m. n.m.
202 write_8192 215 (65%) 305 (98%) n.m. n.m.
203 rewrite_8192 227 (47%) 336 (78%) n.m. n.m.
204 .TE
205 * the operation of the hardware was suspect during these tests.
206 .DE
207 .PP
208 The higher throughput available on VAX 11/780s is due to a number
209 of factors.
210 The larger main memory size allows a larger file system cache.
211 The block allocation routines run faster, raising the upper limit
212 on the data rates in writing new files.
213 .PP
214 The next table makes the same comparison using an Emulex controller
215 on both systems.
216 .DS
217 .TS
218 box, center;
219 c s s s s
220 c s s s s
221 c s s s s
222 l | l s | l s
223 l | l s | l s
224 l | l l | l l
225 l | c c | c c.
226 4.2BSD File Systems Tests - \fBEmulex - 750 vs. 780\fR
227 =
228 Logically Sequential Transfers
229 from an \fB8K/1K\fR 4.2BSD File System (Kbytes/sec.)
230 _
231 Test VAX 11/750 CMI Bus VAX 11/780 SBI Bus
233 1 Drive 2 Drives 1 Drive 2 Drives
234 _
235 read_8192 490 (69%) 620 (96%) 560 (70%) 480 (58%)
236 write_4096 380 (99%) 370 (99%) 440 (98%) 440 (98%)
237 write_8192 470 (99%) 470 (99%) 490 (98%) 490 (98%)
238 rewrite_8192 650 (99%) 620 (99%) 760 (100%) 560 (72%)
239 =
240 .T&
241 c s s s s
242 c s s s s
243 l | l s | l s
244 l | l s | l s
245 l | l l | l l
246 l | c c | c c.
247 Logically Sequential Transfers
248 from an \fB4K/1K\fR 4.2BSD File System (Kbytes/sec.)
249 _
250 Test VAX 11/750 CMI Bus VAX 11/780 SBI Bus
252 1 Drive 2 Drives 1 Drive 2 Drives
253 _
254 read_8192 300 (60%) 400 (84%) 490 (77%) 370 (66%)
255 write_4096 320 (98%) 320 (98%) 380 (98%) 370 (98%)
256 write_8192 340 (98%) 340 (99%) 380 (99%) 370 (97%)
257 rewrite_8192 450 (99%) 450 (98%) 490 (87%) 350 (66%)
258 .TE
259 .DE
260 .PP
261 The following table illustrates the evolution of our testing
262 process as both hardware and software problems effecting
263 the performance of the Emulex SC780 were corrected.
264 The software change was suggested to us by George Goble
265 of Purdue University.
266 .PP
267 The 4.2BSD handler for RH750/RH780 interfaced disk drives
268 contains several constants which to determine how
269 much time is provided between an interrupt signaling the completion
270 of a positioning command and the subsequent start of a data transfer
271 operation. These lead times are expressed as sectors of rotational delay.
272 If they are too small, an extra complete rotation will often be required
273 between a seek and subsequent read or write operation.
274 The higher bit rate and rotational speed of the 2351A Fujitsu
275 disk drives required
276 increasing these constants.
277 .PP
278 The hardware change involved allowing for slightly longer
279 delays in arbitrating for cycles on the SBI bus by
280 starting the bus arbitration cycle a little further ahead of
281 when the data was ready for transfer.
282 Finally we had to increase the rotational delay between consecutive
283 blocks in the file because
284 the higher bandwidth from the disk generated more memory contention,
285 which slowed down the processor.
286 .DS
287 .TS
288 box,center,expand;
289 c s s s s s s
290 c s s s s s s
291 c s s s s s s
292 l | l s | l s | l s
293 l | l s | l s | l s
294 l | l s | l s | l s
295 l | c c | c c | c c
296 l | c c | c c | c c.
297 4.2BSD File Systems Tests - \fBEmulex SC780 Disk Controller Evolution\fR
298 =
299 Logically Sequential Transfers
300 from an \fB8K/1K\fR 4.2BSD File System (Kbytes/sec.)
301 _
302 Test Inadequate Search Lead OK Search Lead OK Search Lead
303 Initial SBI Arbitration Init SBI Arb. Improved SBI Arb.
305 1 Drive 2 Drives 1 Drive 2 Drives 1 Drive 2 Drives
306 _
307 read_8192 320 370 440 (60%) n.m. 560 (70%) 480 (58%)
308 write_4096 250 270 300 (63%) n.m. 440 (98%) 440 (98%)
309 write_8192 250 280 340 (60%) n.m. 490 (98%) 490 (98%)
310 rewrite_8192 250 290 380 (48%) n.m. 760 (100%) 560 (72%)
311 =
312 .T&
313 c s s s s s s
314 c s s s s s s
315 l | l s | l s | l s
316 l | l s | l s | l s
317 l | l s | l s | l s
318 l | c c | c c | c c
319 l | c c | c c | c c.
320 Logically Sequential Transfers
321 from an \fB4K/1K\fR 4.2BSD File System (Kbytes/sec.)
322 _
323 Test Inadequate Search Lead OK Search Lead OK Search Lead
324 Initial SBI Arbitration Init SBI Arb. Improved SBI Arb.
326 1 Drive 2 Drives 1 Drive 2 Drives 1 Drive 2 Drives
327 _
328 read_8192 200 220 280 n.m. 490 (77%) 370 (66%)
329 write_4096 180 190 300 n.m. 380 (98%) 370 (98%)
330 write_8192 180 200 320 n.m. 380 (99%) 370 (97%)
331 rewrite_8192 190 200 340 n.m. 490 (87%) 350 (66%)
332 .TE
333 .DE
334 .ds RH Conclusions
335 .bp