| KYLIN SecStor system is based on the YINHE KYLIN operating system which is developed by Computer College of National University of Defense Technology. Datum is encrypted and stored centrally in the server, and can be encrypted and decrypted transparently in kernel level by the server operating system, which support strongbox-based file encryption and securely sharing.KYLIN SecStor system uses central datum storage style because of single server, which can not meet the need of numerous user requests. As a result, we want to explore a system expanding strategy.Samba[1] is the data output subsystem of KYLIN SecStor system, so the thesis think that it is the first step of system expanding to investigate and deploy high performance samba distributed file system. Now this system can work well but not enough, it consists of a metadata server which is called the root server and a group of back-end storage servers. Root server can republic share volumes which are called root directories, whose function is to supply a global virtual view for users. It achieves this by means of making symbol links, the link name is called link and the real volume is called link target, one link can map multiple link targets, every link target is a redundant replica. When clients access the file system, the root server will return the referral list which contains the symbol link targets .After the clients get the referral list, they will access the link targets following the order stored in the referral list until they find the first alive link target or fail if all the link targets are dead. So, for the most part, all the accesses will concentrate on some data storage server, which will bring about load incline among the servers. What is more, because KYLIN SecStor system supplies transparent encryption storage service, data modifying and updating are the fundamental functions, which will bring about the inconsistency among the replicas in distributed storage system. However, now samba has not done anything to ensure the replica consistency. The thesis's aim is to do some research about the load balancing and replica data consistency of samba distributed file system.For load balancing, this thesis has put forward a load balancing strategy which is based on the performance of back-end servers. This strategy mainly consists of two sub-strategies, which are called back-end load feedback strategy and front-end load scheduling strategy. The back-end storage servers periodically feed back the load index, which is represented by the throughput per second of servers'network interface card. While the front-end root server computes the available load indexes of back-end servers according to the peak load indexes and current load indexes, then use the available load indexes to do the load scheduling. In samba distributed file system, it means that before root server returns the referral list which contains the symbol links targets, it should reorder the referral list to make that the larger the target server's current available load index is, the topper the symbol link target is.For replica consistency, this thesis has the strategy below. Firstly, update operations must be done in the master replica,which can be ensured by locating the users that have the updating rights on the master replica. As a result we can resolve the write-write conflicts. Secondly, for the inconsistency because of read-write conflicts, we have the master replica push-mode updating propagation strategy, which is that the master replica will positively propagate the updating to the other replicas as soon as it has updated, which will make all replicas come to consistent.We have tested the improved samba distributed file system by the software which is called netbench, as a result, the overall throughput and average response time to clients both have improved heavily. |
PDF Full Text Request