Assessment of Deep Sequence Models for Characterization and Prediction of Cloud Workloads
Manasa Pachava, Karthick Seshadri, Chidambaran Kollengode · 2022 IEEE 3rd Global Conference for Advancement in Technology (GCAT) · 2022
Forecasting cloud workloads is necessary for allocating resources when an application request is received at a data center. The workload of a data center can be considered as a time sequence. Workload forecasting of a data center can be done using sequential models like Long Short Term Memory (LSTM) or Gated Recurrent Networks (GRN). The time sequence of workload across VMs is typically tracked at the granularity of groups of VM. In this paper, we use LSTM, GRU, and Encoder-decoder network for forecasting resource requirements and assess the performance of each model. We compare these models in grouped workload settings. We use the Azure workload dataset with approximately 2 × 105VMs and the Bitbrains dataset with 1250 VMs for validating the models.