A Load Awareness Object Based Hybird Distributed Storage System

In the cloud environment,virtualization technology has being widely used.The block storage system in the service of the virtual machines plays an important role in the cloud storage.But it is more and more difficult for traditional hard disk and storage architecture,such as SAN and NAS,to meet new requirements.A new kind of storage device–SSD with high performance starts a revolution in the field of storage.People hope that it can improve the overall performance of the storage system.But due to its cost,the data center in PB level cannot afford the bill of replacing all storage device with SDD.So the solution of hybrid storage system is a good one with both high performance and high cost performance.This paper proposes a load awareness object based hybrid distributed storage system – BOBBS.It can provide efficient block storage service to virtual machines in the cloud environment.And it has high extensibility.Its underlying architecture is an object hybrid distributed storage system with SSD and HDD.There are two scheduling system in BOBBS.In the underlying storage architecture,BOBBS creates two storage pools – SSD POOL and HDD POOL.All objects from VMDI are placed in these two pools.They will migrate across pools according to their temperature and most hot object will be in the SSD POOL.The migration needs the monitor node to collect,analyze and calculate object heat value and design the scheduling strategy.With the increasing of the scale of the system,this node will easily become the bottleneck of system's performance and single fault node,so we change it into a distributed system:Firstly,we design a heat model for monitor nodes to represent their heat.This model combines the history data and current monitoring data to put forward a forecasting formula.For a single monitor node,it can help the node to know its heat value through calculating a variety of performance data.Secondly,we design the monitor cycle and its distributed topological structure.We build up the communication and exchanging data mechanism on the cycle,which is the foundation structure of the following scheduling algorithm and dynamic migration mechanismThirdly,we design the diffusion scheduling algorithm.Monitors will use this algorithm to analyze their heat value and run the migration to diffuse hot node's load to surrounding nodes with low heat value.In the end,all monitor nodes will have a balanced heat value.Finally,we run block device and file system I/O benchmark on BOBBS and WHOBBS to verify the improvement of BOBBS's performance.We also verify the dynamic diffusion scheduling algorithm and mechanism of the BOBBS.Our test results show that BOBBS is more effective than the WHOBBS.So the diffusion scheduling algorithm works.