Research Of Physical Memory Sharing Technology In Cloud Environment

Memory is an exclusive device to computers and can only be used by local hosts. Memory capacity is limited and unchangeable in the condition of no change to memory hardware configuration. Even though the capacity of physical memory of single host can be terabytes with the development of technology, it makes the host much more expensive than usual. Besides, the quantity of computation task is dynamic for enterprises, to satisfy the computation requirement only by improving the performance of host hardware will lead to higher running cost and lower rate in resources' utilization.Distributed system and parallel computing system use middleware to combine the cluster hosts together logically, which makes it achieves that computation task is stored separately in the memory of cluster hosts and handled simultaneously. Middleware runs in user space, which doesn't hide implementation details to users. As a result, some applications should be rebuilt or rewritten, which brings a lot of inconvenience. Meanwhile, some tasks cannot be divided into pieces because their inner dependencies may cause some other problems such as synchronization.Virtualization technology makes it possible that the memory capacity of virtual machines can be adjusted dynamically, which shows the possibility that memory capacity can be changed by software. Inspired by the situation above, a new kind of technology for memory sharing is proposed: one host can rent its free memory to other hosts in the same cluster whose free memory is not enough through network. In other words, memory is not anymore exclusive for just one host, it also acts as public resources and can be shared by all the hosts among the cluster. The memory capacity of hosts would not stay fixed, it can be changed according to the computation requirements. Those computation tasks that couldn't be handled by distributed system can be processed in one host without caring the limitation of memory capacity. The related works are as follows:1. By analyzing computer architecture and memory management in Linux kernel, the model of memory sharing technology is proposed. One host can borrow and use the memory from other hosts in the same cluster through network. All the procedures are processed in OS kernel layer, implementation details are hidden to users and applications. Users would not feel any difference between remote memory and local memory.2. Based on memory sharing model and memory dynamic adjustment technology for virtual machines, algorithms for memory autonomous extension and shrink are put forward. Host memory capacity can be adjusted according to the requirement, and no external intervention is needed.3. In average case, 25% memory pages can't be used and would be wasted when using buddy algorithm. It would be more serious for huge page systems which are used in virtualization environments. The modified buddy algorithm is adapted to the allocation of arbitrary page number. The linear property is improved and no page wasting appears.4. Due to the situation that network performance is the bottleneck for the performance of I/O in memory sharing technology, shared memory is regard as the swap space in the role of block devices. Meanwhile, page swap algorithm is changed to reduce page turbulence and page fault.5. Differences are superficially analyzed on performance between memory sharing technology with distributed system and parallel computing system.