Object Based Hybrid Cloud Storage System

With the arrival of big data era,cloud computing which is capable of handling large amount of data becoming hot spot in academic circle.Distributed storage is the foundation of cloud computing and the improvement of distributed storage layer is the key point to improve the performance of computing clusters.The development of Solid State Drives(SSDs)has brought a revolution to cloud storage architecture.Instead of using a motor to spin disks,SSDs utilize integrated circuit assemblies as memory to store data persistently.Compared with HDDs,SSDs are much faster,quieter,more reliable,less power consumption but much more expensive.Traditional Hard Disk Drive(HDD)architecture is below the requirement of current cloud computing center and utilization of SSDs becomes the key to improve the performance of storage architecture.Howerver the cost of SSD per GB is ten times of HDDs.Most cloud computing center can hardly afford such high cost.The max operation count on SSD is less than it on HDD and data stored on SSDs is volatile,these facts make it impossible for SSD to replace HDD at once.Hybrid storage architecture becomes a trade-off method to gain relative high performance storage cluster with acceptable cost.Although there is already an ocean of researches in hybrid storage,none of them has worked out a mature hybrid storage architecture to perform well under all situations.We design an object-based hybrid storage system called WHOBBS to deal with the problem upon.WHOBBS is built on a hybrid storage layer with SSDs and HDDs to provide block storage service for virtual machines on cloud.Hybrid storage layer is totally transparent to virtual machines upon.WHOBBS splits Virtual Machine Disk Image(VMDI)into a set of objects with the same size and stores them to different storage devices according to their own access pattern.WHOBBS collects and analyzes virtual machine workload in runtime to make the global-optimized objects placement strategy.WHOBBS will carry out data migration if the objects placement strategy has changed to ensure high utilization of SSDs.WHOBBS is built upon Ceph,an open source object storage architecture,and modify its source code to implement hybrid object storage.We run block device and file system I/O benchmark on WHOBBS to verify the improvement of its performance.Our test results show that WHOBBS is effective than original Ceph with real time data analyze and migration.