21 Fractal Analysis and Programming of Elastic Systems …
317
require to compute or even know f (x), but it computes the slope, that is, v = v τ2 −
v τ1 , where v τ2 − v τ1 is a slop. If a slope is positive, then max(v) will increase. If a
slope is negative, then max(v) will decrease.
The idea is to identify a slope between the neighboring points. The algorithm is
mentioned below.
Algorithm 1. Gradient Ascent for Big Data ordered in time
1. V is a random initial vector ordered at discrete-time T = [0,..., τn]
max(vτ) = 0 is an initial value of random vector max/min vτ V
2. for each τ in T
3. if max(vτ) < vτ
4. max(vτ) = vτ
5. end for each
6. return max(vτ)
The algorithm runs until found max(v τ ) or min(v τ ).
21.4.2 Distributed Algorithm: Gradient Ascent for Big Data
Ordered in Time
Our objective is to construct the distributed CCM that captures a relationship between
the computing time m of Algorithm 1 (A1) and capacity c. For solution of this issue,
the following assumptions are made:
• Let vector V has about 2000 M (Millions) values and changeable during some
period of time, V = 2000 M.
• Let m be the computation time, and c be the capacity of the system that handles
by u[m], see Eq. 21.6.
• Let a server be encapsulated in container e 0 . The capacity of the server c =
2 MWIPS (Millions of Whetstone Instruction Per Second [24, 25]) is a constant,
and each server encapsulated in container has the same capacity.
• Algorithm A1 consists from three operations in one loop (w = 3): For each τ in T
= [0, …, τ n ], condition operator is less then “<”, and assignment operator is “=”.
Thus, the computation time m of Algorithm 1 with u[m] containers is presented
by Eq. 21.1.
m = V /(c ∗ u[m]/w)
(21.11)
Figure 21.8 shows a container-component tree, where the containers are marked
by capacity c.
For the computation time m, vector V needs u[m] = V : w/(m : c) =
2000 M : 3/(2 MWIPS : u[m]) containers. The calculation of algorithm can be planned
317
require to compute or even know f (x), but it computes the slope, that is, v = v τ2 −
v τ1 , where v τ2 − v τ1 is a slop. If a slope is positive, then max(v) will increase. If a
slope is negative, then max(v) will decrease.
The idea is to identify a slope between the neighboring points. The algorithm is
mentioned below.
Algorithm 1. Gradient Ascent for Big Data ordered in time
1. V is a random initial vector ordered at discrete-time T = [0,..., τn]
max(vτ) = 0 is an initial value of random vector max/min vτ V
2. for each τ in T
3. if max(vτ) < vτ
4. max(vτ) = vτ
5. end for each
6. return max(vτ)
The algorithm runs until found max(v τ ) or min(v τ ).
21.4.2 Distributed Algorithm: Gradient Ascent for Big Data
Ordered in Time
Our objective is to construct the distributed CCM that captures a relationship between
the computing time m of Algorithm 1 (A1) and capacity c. For solution of this issue,
the following assumptions are made:
• Let vector V has about 2000 M (Millions) values and changeable during some
period of time, V = 2000 M.
• Let m be the computation time, and c be the capacity of the system that handles
by u[m], see Eq. 21.6.
• Let a server be encapsulated in container e 0 . The capacity of the server c =
2 MWIPS (Millions of Whetstone Instruction Per Second [24, 25]) is a constant,
and each server encapsulated in container has the same capacity.
• Algorithm A1 consists from three operations in one loop (w = 3): For each τ in T
= [0, …, τ n ], condition operator is less then “<”, and assignment operator is “=”.
Thus, the computation time m of Algorithm 1 with u[m] containers is presented
by Eq. 21.1.
m = V /(c ∗ u[m]/w)
(21.11)
Figure 21.8 shows a container-component tree, where the containers are marked
by capacity c.
For the computation time m, vector V needs u[m] = V : w/(m : c) =
2000 M : 3/(2 MWIPS : u[m]) containers. The calculation of algorithm can be planned