| blob | 21052eee2f023a2279932dce4a95b291c8002a3d |
1 .\" $NetBSD: conclusions.ms,v 1.3 2000/03/13 23:03:36 soren 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 .\" @(#)conclusions.ms 6.2 (Berkeley) 4/16/91
31 .\"
32 .ds RH Conclusions
33 .NH
34 Conclusions
35 .PP
36 Peak available throughput is only one criterion
37 in most storage system purchasing decisions.
38 Most of the VAX UNIX systems we are familiar with
39 are not I/O bandwidth constrained.
40 Nevertheless, an adequate disk bandwidth is necessary for
41 good performance and especially to preserve snappy
42 response time.
43 All of the disk systems we tested provide more than
44 adequate bandwidth for typical VAX UNIX system application.
45 Perhaps in some I/O-intensive applications such as
46 image processing, more consideration should be given
47 to the peak throughput available.
48 In most situations, we feel that other factors are more
49 important in making a storage choice between the systems we
50 tested.
51 Cost, reliability, availability, and support are some of these
52 factors.
53 The maturity of the technology purchased must also be weighed
54 against the future value and expandability of newer technologies.
55 .PP
56 Two important conclusions about storage systems in general
57 can be drawn from these tests.
58 The first is that buffering can be effective in smoothing
59 the effects of lower bus speeds and bus contention.
60 Even though the UDA50 is located on the relatively slow
61 UNIBUS, its performance is similar to controllers located on
62 the faster processor busses.
63 However, the SC780 with only one sector of buffering shows that
64 little buffering is needed if the underlying bus is fast enough.
65 .PP
66 Placing more intelligence in the controller seems to hinder UNIX system
67 performance more than it helps.
68 Our profiling tests have indicated that UNIX spends about
69 the same percentage of time in the SC780 driver and the UDA50 driver
70 (about 10-14%).
71 Normally UNIX uses a disk sort algorithm that separates reads and
72 writes into two seek order queues.
73 The read queue has priority over the write queue,
74 since reads cause processes to block,
75 while writes can be done asynchronously.
76 This is particularly useful when generating large files,
77 as it allows the disk allocator to read
78 new disk maps and begin doing new allocations
79 while the blocks allocated out of the previous map are written to disk.
80 Because the UDA50 handles all block ordering,
81 and because it keeps all requests in a single queue,
82 there is no way to force the longer seek needed to get the next disk map.
83 This disfunction causes all the writes to be done before the disk map read,
84 which idles the disk until a new set of blocks can be allocated.
85 .PP
86 The additional functionality of the UDA50 controller that allows it
87 to transfer simultaneously from two drives at once tends to make
88 the two drive transfer tests run much more effectively.
89 Tuning for the single drive case works more effectively in the two
90 drive case than when controllers that cannot handle simultaneous
91 transfers are used.
92 .ds RH Acknowledgements
93 .nr H2 1
94 .sp 1
95 .SH
96 \s+2Acknowledgements\s0
97 .PP
98 We thank Paul Massigilia and Bill Grace
99 of Digital Equipment Corp for helping us run our
100 disk tests on their UDA50/RA81.
101 We also thank Rich Notari and Paul Ritkowski
102 of Emulex for making their machines available
103 to us to run our tests of the SC780/Eagles.
104 Dan McKinster, then of Systems Industries,
105 arranged to make their equipment available for the tests.
106 We appreciate the time provided by Bob Gross, Joe Wolf, and
107 Sam Leffler on their machines to refine our benchmarks.
108 Finally we thank our sponsors,
109 the National Science Foundation under grant MCS80-05144,
110 and the Defense Advance Research Projects Agency (DoD) under
111 Arpa Order No. 4031 monitored by Naval Electronic System Command under
112 Contract No. N00039-82-C-0235.
113 .ds RH References
114 .nr H2 1
115 .sp 1
116 .SH
117 \s+2References\s0
118 .LP
119 .IP [McKusick83] 20
120 M. McKusick, W. Joy, S. Leffler, R. Fabry,
121 ``A Fast File System for UNIX'',
122 \fIACM Transactions on Computer Systems 2\fP, 3.
123 pp 181-197, August 1984.
124 .ds RH Appendix A
125 .bp