Gentoo Archives: gentoo-sparc

From: Keith M Wesolowski <wesolows@××××××××.org>
To: Zhang Weiwu <zhangweiwu@××××××.com>
Cc: Andrew Gaffney <agaffney@g.o>, gentoo-sparc@l.g.o
Subject: Re: [gentoo-sparc] what ultrasparc compare likely such Pentium box on desktop performance
Date: Thu, 02 Dec 2004 16:44:52
Message-Id: 20041202164447.GA27826@foobazco.org
In Reply to: Re: [gentoo-sparc] what ultrasparc compare likely such Pentium box on desktop performance by Zhang Weiwu
1 On Thu, Dec 02, 2004 at 06:19:25PM +0800, Zhang Weiwu wrote:
2
3 > Here, again, I wanna be suggested by you experts.
4 >
5 > 1) I am a saleman (luckly, I am just technical enough to use gentoo, not
6 > any more). My work mostly involve mozilla and openoffice, acroreader etc.
7
8 Ok...
9
10 > I want to have a sparc desktop now. We are using in the company very
11 > complicated web-based database and customer management system, I usually
12 > (with my notebook) wait 8 seconds untill one webpage fully loaded,
13 > because each page is too complicated with javascript and lots of forms
14 > and tables. (Network is very fast.) I hate to let my customer wait 8
15 > seconds on the phone untill I could see the customer's record, because 8
16
17 If you can't or won't analyse your workload at a deep technical level
18 to understand its resource requirements (memory? CPU? S-cache? disk
19 I/O?), just go to your local retailer and ask for the most powerful
20 Opteron workstation he sells. It may or may not meet your needs, and
21 it will cost you dearly, but at least you'll know you have all the
22 performance available in a modern desktop computer. Buying less
23 capable systems than the state of the art is only recommended for
24 people with sufficient understanding of exactly what resources their
25 applications require. While you're done an excellent job of
26 describing the problem you are experiencing, you cannot determine this
27 without assistance from local technical personnel who can observe your
28 application, your system, and the behaviour of both. Simple things
29 like whether there is a lot of disk noise or the existence of spyware
30 on Windows systems are crucial to performance and to understanding
31 resource requirements. Tools like vmstat(1M), iostat(1M), and if
32 you're fortunate enough to have it, dtrace(1M), can be needed to
33 understand the problem fully. This knowledge is neither expected nor
34 demanded of sales personnel, which is why your company should provide
35 knowledgeable IT staff who can assist you in translating your
36 detailed, but nontechnical, explanation of the problem into the kind
37 of information that can lead to a well-informed purchase of
38 technology.
39
40 > A) is U2 with dual 400Mhz fast enough for me?
41 > B) how about U60 with dual CPU?
42
43 I doubt very much that either system would be fast enough for you,
44 though it does depend on _why_ your current solution is not performing
45 well; see below.
46
47 > I don't want to buy anything too fast, I always have other better way to
48 > spend my money:)
49
50 You just told us your important workload, and that it needs to be MUCH
51 faster than what your existing solution delivers. Well, web page
52 rendering can only be accelerated in 3 ways: (1) fix the web page to
53 reduce its complexity, or improve browsers' rendering engines. I
54 assume this is not an option. (2) More memory. Any desktop today
55 should have at absolute minimum 512MB, preferably 1GB. Not because
56 it's really needed, but see (1). Most laptops don't have this. Any
57 UltraSPARC system supports it, as do all modern desktop PCs and
58 workstations and most modern laptops. (3) More CPU power. Assuming
59 you've already maxed out memory, this is the only real option.
60
61 The problem with all this is that we don't have enough information to
62 know _why_ your current solution is too slow.
63
64 It's possible that your laptop has 64MB memory and a typical laptop
65 disk running at 80 rpm or whatever, so you're constantly swapping to a
66 horribly slow medium. That could certainly explain your problem, and
67 an U2 would probably alleviate it somewhat if it has sufficient
68 memory. So would a 256MB SODIMM from the local parts shop ((2)
69 above). Then again, it might be that the javascript is executing on
70 CPU in a single thread for a long time; in this case the solution
71 would be a faster single CPU ((3) above). Of course, it could also be
72 something more complicated still, like pathological browser or
73 windowing system behaviour triggered by the particular pages you view.
74 Diagnosing this slowness is beyond the scope of this mailing list, so
75 I'd suggest you consult a local expert to understand better what
76 system resources required by your application are lacking in your
77 current solution. Only when you understand that can you determine
78 conclusively whether Alternate Solution X or Y will be better for you.
79 This is what I meant by "understanding your workload."
80
81 Mozilla and most other modern browsers are indeed multithreaded, so
82 SMP sometimes helps somewhat. Of course, how much it will help
83 depends on the bottleneck. Not all operations are multithreaded. But
84 unless you have real considerations other than raw speed, such as
85 aesthetics, a desire to do systems programming, or applications
86 supported only on SPARC, I don't see any compelling reason to choose a
87 U2 or U60 in any configuration over a comparably-priced Athlon64 or,
88 if you absolutely must reduce cost, Athlon, box.
89
90 --
91 Keith M Wesolowski
92 "Site launched. Many things not yet working." --Hector Urtubia
93
94 --
95 gentoo-sparc@g.o mailing list