China Grid and Related Dependability Research - Grid Computing: Infrastructure, Service, and Applications - page 101

If Δ intervalmax = 0, the QoS cannot be achieved. P L is t he probabilit y of message
loss and P L can be simply computed as in Equation 4.6:

CC -

P =

To Re

(4.6)

L

C

To

C To is the count of total messages that are sent in a sample window, and C Re
is the count of the received message in the sample window.

Step 2: [get Δ interval ]
Let

U
D

T

D

D+ ( Tj

U

-D )

2

interval

'

f ( D ) = D

tr D interval

(4.7)

interval interval

U 2

D+ PT ( -D j )

j = 1

tr L D interval

i nd the largest Δ interval which is less than Δ intervalmax and with f (Δ interval ) ≥
T L

MR . Such a Δ interval always exists.

Step 3: [estimate arrival time of next heartbeat message]
Compute m and s with Equation 4.3 by using the message behavior in the
sliding window. Estimate the next message arrival time as

ET =m+s W (4.8)

n + 1 s

Here, W s is the window size of heartbeat messages.

Step 4: [get freshness point t n +1 ]
Compute the freshness point:

U

t= ET + -D T (4.9)

n + 1 n + 1 D interval

If no fresh message is received in t n +1 time, suspect the monitored process,
that is,

D= t - T (4.10)

timeout n + 1 now

With Algorithm 4.1, the failure detector between two processes can be
adaptive to the system conditions. Next the algorithm organizing the
deployment of the failure detectors in grids is presented. The details are
shown in Algorithm 4.2.