IS-15-011 Ha Tuan Minh A High-speed Scheduling Method of Virtual Machine Placement with Search Parameter Estimation by Ha Tuan Minh, Ryoji Kobayashi, Masaki Samejima (Osaka University) Abstract We address a scheduling problem of dynamically provisioning virtual machines (VMs) in order to minimize power consumption on servers. Our research group has already developed a scheduling method of k VMs on n servers every λ minutes based on predicted power consumption during the following T minutes. We propose a parameter estimation method of k, n, λ based on the following CPU resource usage to make schedules more efficiently. The proposed method determines appropriate parameter values for past time-series data of CPU resource usage. The appropriate parameter values for scheduling are estimated by collaborative filtering with multiple similar time-series data and the appropriate parameter values of it. Performing a simulation experiment with an assumption of 200 servers and 400 VMs, we have confirmed that the proposed method can reduce the power consumption by 2.5% and computational time by 88% compared to our conventional method.,, (Virtual machine dynamic placement problem, Branch and bound, Collaborative filtering, Scheduling problem ) 1. [1] (Virtual Machine:VM) [2, 3] [4] VM VM VM VM [5 7] VM VM [8,9] [10] [11] VM VM VM (VM ) T CPU k VM n λ(< T) [12] [13,14] λ, k, n λ k, n CPU λ λ k, n VM λ, k, n CPU CPU 2. VM 21 VM CPU [11] VM CPU 1/5
Minimize { T N } (C i(t) rate i + B i E i)+m(t) E M t=1 i=1 subject to C i(t) = j J i (t) V j(t) C i(t) 100 1 (C i(t) > 0) B i = 0 (C i(t) =0) C i(t) i t CPU [%]rate i i CPU 1% [W/%]E i [W] B i (1 0 ) E M 1 [W] m(t) t C i(t) i j( J i(t)) VM CPU V j 100% J i(t) 22 VM 1 VM [13, 14] T VM CPU [15,16] λ VM T CPU CPU CPU T VM VM k n 1 k VM Fig. 1. VMs 1 VM Scheduling problem for dynamically provisioning VM λ VM CPU λ, k, n λ λ k, n VM λ, k, n CPU 23 2 CPU T CPU () λ VM k n CPU T CPU s( S) T D T,s 2 Fig. 2. Issues on estimating parameter values 2/5
1 VM s VM 2 VM VM 3 D T,s D T,s 3. VM 31 CPU [17] 3 VM k n λ λ CPU λ λ k, n λ T s D T,s k, n T λ s D T λ,s λ, k, n VM λ 32 λ s T D T,s λ λ 4 λ k, n λ R λ,s D T,s R λ,s λ P λ P λ λ 1 P λ = s R λ,s D T,s 4 70 1 60 2 λ 1 1 2 2 1 λ 1 0.005 λ 1 λ 1 Fig. 3. 3 Outline of the proposed method Fig. 4. 4 λ Estimation of rescheduling time λ 33 VM k n s T λ D T λ,s VM k n k, n 5 3.2 λ k, n k, n R k,n,s λ λ k, n P k.n P k.n k, n 1 P k,n = s R k,n,s D T λ,s 5 60 50 4 T T λ k 2 n 3 1 1 2 2 k 2 n 3 0.006 k 2 n 3 3/5
Fig. 6. 6 Comparison of power consumption 5 VM k n Fig. 5. Estimation of the numbers of target VMs k and servers n in rescheduling 4. 41 CPU Intel Core i7 3.4GHz 8GB VM CPU Web [18,19] rate i 0.5W/% 0.1 E ie M 5W30W 200 VM400 16 T =4024 K = {1, 2, 3, 4},N = {1, 2,..., 8}, Λ={5, 10, 15, 20} 9 CPU (λ = 20) (k =3,n=6,λ= 20) 42 24 CV (=/ ) λ 6 7 8 24 CPU VM 16% 2.5% 8 0.4 94% λ =5 0.4 80% λ =5 λ λ 7 Fig. 7. Comparison of average computational time of scheduling Fig. 8. of λ 8 λ Coefficient of variation and time-series variation 88% 8 211 λ =5 152 VM 5. CPU VM 4/5
VM λ VM k n CPU λ, k, n VM CPU 200 VM400 2.5% 3.3% CPU 88% λ JSPS 24360154 1 P. Thibodeau, Data centers are the new polluters, Computerworld (2014) 2 R. Uhlig, G. Neiger, D. Rodgers, A. L. Santoni, F. C. Martins, A. V. Anderson, and L. Smith, Intel virtualization technology, Computer, Vol. 38, No. 5, pp. 48 56 (2005) 3 M. Stillwell, V. Frederic, and C. Henri, Dynamic fractional resource scheduling for hpc workloads, IEEE International Symposium on Parallel & Distributed Processing(IPDPS 10), pp. 1 12 (2010) 4 N. Kim, J. Cho, and E. Seo, Energy-Based Accounting and Scheduling of Virtual Machines in a Cloud System, In Proc. of IEEE/ACM International Conference on Green Computing and Communications, pp. 176 181 (2011) 5 H. Liu, H. Jin, C.Z. Xu, and X. Lia. Performance and energy modeling for live migration of virtual machines, Cluster computing, Vol. 16, No. 2, pp. 249 264 (2013) 6 S. Akoush, R. Sohan, A. Rice, A.W. Moore, and A. Hopper, Predicting the performance of virtual machine migration, In Proc. of IEEE International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS), pp. 37 46 (2010) 7 W. Voorsluys, J. Broberg, S. Venugopal, and R. Buyya, Cost of virtual machine live migration in clouds, A performance evaluation, Cloud Computing, pp. 254 265 (2009) 8 N. Bansal, C. Alberto, and S. Maxim, Improved approximation algorithms for multidimensional bin packing problems, IEEE 4th Annual Symposium on Foundations of Computer Science(FOCS 06), pp. 697 708 (2006) 9 S. S. Seiden, On the online bin packing problem, Journal of the ACM (JACM), pp. 640 67 (2002) 10,, No. 2010- EVA-33, Vol. 4, pp.1-7 (2010) 11,, Vol. 5, No. 5, pp. 33 42 (2012) 12 H.H. Greenberg, A branch-bound solution to the general scheduling problem, Operations Research, Vol. 16, No. 2, pp. 353 361 (1968) 13 R. Kobayashi, K. Sato, M. Samejima, and N. Komoda, A Prediction-based Scheduling Method for Dynamic Placement of Virtual Machines in Data Center, In Proc. of International Conference on Applications of Computer Engineering (ACE), pp. 19 24 (2014) 14 26 pp. 1791 1792 (2014) 15 G. Kitagawa and W. Gersch, A smoothness priors long AR model method for spectral estimation, IEEE Transactions on Automatic Control, Vol. 30, No. 1, pp. 3-8 (1985) 16 B. E Hansen, The grid bootstrap and the autoregressive model, Review of Economics and Statistics, pp. 594 607 (1999) 17 P. Resnick, N. Iacovou, M. Suchak, P. Bergstrom and J. Riedl, GroupLens: An Open Architecture for of Netnews, In Proc. of Computer-Supported Cooperative Work(CSCW), pp. 175 186 (1994) 18 U. Guido, G. Pierre, and M.V. Steen, Wikipedia workload analysis for decentralized hosting, Computer Networks, Vol. 53, No. 11, pp. 1830 1845 (2009) 19 J. Allspaw, - (2014) 5/5