Corrigenda for Book

Corrigenda for Guerrilla Capacity Planning Book

Last updated May 1, 2010

Page Spotter Correction
 p. 31  Author   Table in Example 3.5: Last entry s.b. 24 sigdigs. Must move decimal point 23 places to right.
 p. 32  M. Berger   para 2, line 4: "To help help rectify..." Elide 2nd "help"
  The Author   Algortihm 3.2: The assignment of XYZ needs to be made clearer. See note 1.1 below.
 p. 49  T. Wilson   Line above Fig. 4.5: "grapically" s.b. graphically
  T. Wilson   Def. 4.3: "executes in purely sequential of serial fashion" of s.b. or
 p. 51  P. Stalder   Fig. 4.6 caption s.b. ...serial fraction σ = 0.10 corresponds to...
 p. 52  Author   Eqn. (4.19): The S is shorthand for S(p). Using Sp might be a better choice.
 p. 56  Author   Line 5 from bottom: σ−1 s.b. σ−1
  T. Wilson   Sec. 4.3.5: [Gustafson 1992] citation missing from bibliography and date s.b. 1988. See note 1.3 below.
 p. 77  M. Berger   First line: "we _ more than the three data points..." Insert "need"
 p. 116  R. Hamilton  Sect. 6.9 Line 7 from bottom. "be confiigured..." s.b. configured
 p. 135  M. Berger  Sentence before Sect. 7.4.4: "...became available to the high-priority VMs" s.b. low-priority
  T. Wilson   First complete para: "With the in mind..." the s.b. this
 p. 147  M. Berger  Period missing at end of 2nd para.
 p. 149  M. Berger  Sect. 8.3.2, Line 8: "...measurements than than are generally available..." Elide 2nd "than"
 p. 152  Author   Eqn.(8.1) should only have 4 random variables, not 6 X's. Two r.v.'s (MinU and MaxU in Fig. 8.5) are used to filter the data.
 The VBA code in Appendix E.2 is correct.
 p. 152  R. Hamilton   2nd line in Sect. 8.6 "We new present ..." s.b. now
 p. 154  R. Schmitt  Last line s.b. "...larger values of Ueff are estimated..."
 p. 159  R. Hamilton   Table 8.1: Last digit is missing in each of the ∆CLK numbers. s.b. 36677 and 42340, respectively.
 p. 161  R. Hamilton   Example: 8.3, 1st line: equation should read UC(20) = 246.97%
 p. 165  R. Hamilton   Para 2, line 6. "Mnay of these..." s.b. Many
 p. 217  Author   D s.b. S in equations A.23 and A.24.
 p. 241  R. Hamilton  Line above Section F.3.2: "...reduces to Amdahl's law, as expcted." s.b. ...expected.

1  NOTES

1.1  Significant Digits and Rounding

To round the example number 7.245 to 3 sigdigs using Algortihm 3.2, start by rewriting the number without the decimal point. In a more tabular form: 
1 2 3 4 5
7 2 4 5 _
_ _ x y z
the first row shows the position of each of the digits 7 2 4 5 of our number in the second row. The last row shows the alignment of the X, Y and Z labels in the rounding algorithm with X set in the 3rd position because we want only 3 sigdigs. The 5th digit is a blank (denoted _ ) in the original number.

1.2  J2EE and WebLogic Scalability

This is not a correction but additional information pertaining to Section 7.4.4, pp. 135 ff.

The following version numbers were supplied by Jamie Rybicki on Mon, Mar 24, 2008:
BEA WebLogic 9.1 with the BEA JRockit JDK 5.0 Update 6 (R26.4.0-63) 32 bit java virtual machine.
Regarding the WebLogic listen-threads hypothesis, originally due to Drew Sliwkowski (circa December 2005), 
Michael Ducy wrote on 3/24/08 12:06 PM: 
    This makes sense now that you tell me they are running the JRockit JVM.
    JRockit has the concept of "Thin Threads." Basically this is where M
    Java threads are ran inside N OS threads (or the MxN threading model,
    where M > N.) The JVM handles scheduling, synchronization, etc of these
    thin threads within the OS threads. I found an old JRockit developer
    article that indicated that N would most likely be set to the number of
    processors in the system.  This article was written when JRockit thin
    threads were still experimental. I am guessing BEA did some testing and
    found that 2N x CPUs performed better.
The interested reader may be able to find more in details in [J2EE Performance].

1.3  Scaled Speedup

This is not a correction but additional information pertaining to Section 4.3.5. p. 56.

The following diagram is useful for understanding the origin of eqn.(4.29).
The parallel portion of the work in Fig. 4.5 is first scaled up in proportion to the number of available processors: (1−σ) → (1−σ) p.
Then, it follows that the executions times are:

T1
= (1−σ) p T1 + σT1 (1st row of diagram)
(1)
Tp
=  (1−σ) p
p
 T1 + σT1 (2nd row of diagram) 
(2)
The ratio of these two times gives eqn.(4.29).
This kind of [Anti-Amdahl scale-up] has more recently been proposed as a way to optimize the throughput of multicores [Break Amdahl].

References

[J2EE Performance]
Peter Zadrozny, Philip Aston, and Ted Osborne,
J2EE Performance Testing with BEA WebLogic Server,
Expert Press, 2002.
[Anti-Amdahl scale-up]
John Gustafson,
"Reevaluating Amdahl's Law,"
Comm. ACM. 31(5): 532-533, 1988.
http://www.scl.ameslab.gov/Publications/Gus/AmdahlsLaw/Amdahls.html
[Break Amdahl]
Herb Sutter,
"Break Amdahl's Law!"
http://www.ddj.com/hpc-high-performance-computing/205900309


File translated from TEX by TTH, version 3.38.
On 1 May 2010, 15:56.