Coarse-Grain QoS-Aware Dynamic Instance Provisioning for Interactive Workload in the Cloud

Algorithm 1

The learning Algorithm to characterize the Percentile QoS Constraint.

Input: 𝐼𝑇𝐸, 𝑁, and SLA specification 𝐷_th; {𝐼𝑇𝐸 is the number of iterations and 𝑁 is the

  number of decision points in a day.}

Output: VP_table;

(1) Create VP_table and initialize each item in VP_table to 0;

(2) Create 𝑝[𝑖][𝑘] and counter; {𝑝[𝑖][𝑘] is a sample of QoS violation ratio of using 𝑘 VM

    instances in phase 𝑖, and counter logs the number of delay violations in a phase.}

(3) for  𝜂=1 to 𝐼𝑇𝐸  do

(4)  for  𝑖=1 to 𝑁  do

(5)   for  𝑘= MIN_NUM to MAX_NUM do

(6)    Log response time 𝑑_𝑖 for each incoming request;

(7)    if   𝑑_𝑖>𝐷_th  then

(8)     counter++;

(9)   end if

(10)      Calculate an unbiased sample of delay violation probability 𝑝[𝑖][𝑘]=counter/𝑛^𝜂_𝑖,

      where 𝑛^𝜂_𝑖 is the total number of requests arrived in phase 𝑖, iteration 𝜂;

(11)       VP_table[𝑖][𝑘]←𝜂−1/𝜂×VP_table[𝑖][𝑘]+1/𝜂𝑝[𝑖][𝑘];

(12)   end for {Loop 𝑘;}

(13)  end for {Loop 𝑖;}

(14) end for {Loop 𝜂;}