Presentation on theme: "MADFS: The Mobile Agent- based Distributed Network File system Presented by : Hailong Hou Instructor: Yanqing Zhang."— Presentation transcript:
MADFS: The Mobile Agent- based Distributed Network File system Presented by : Hailong Hou Instructor: Yanqing Zhang
Outline INTRODUCTION THE SYSTEM STRUCTURE OF MADFS HIERARCHICAL AND CONVERGENT CACHE COHERENCY MECHANISM THE CACHE COHERENCY MECHANISM IN MADFS
Introduction The conventional distributed file system is designed for LAN environment. They always play poor performance in WAN. In this paper we present a novel distributed file system: The Mobile Agent-based Distributed File System (MADFS). The objective of MADFS is to reduce the overhead of network transfer and cache management inherent to the distribution of a distributed files system in WAN. The MADFS organizes hosts into a hierarchical structure, and uses mobile agents as the underlying facility for transmission, communication and synchronization.
Introduction also present a novel cache coherency mechanism for MADFS: Hierarchical and Convergent Cache Coherency Mechanism (HCCM).
THE SYSTEM STRUCTURE OF MADFS
MADFS is divided into a number of domains in which are connected through high speed LAN and linked to each other through low speed WAN. Each domain is composed of number of hosts. In MADFS, a domain acts as the major domain and is in charge of the all others domains in MADFS. Every server MADFS run the environment for mobile agent and the whole MADFS is a large platform for mobile agent.
In MADFS, all agents can be classified as following: 1) IA (Interface Agent). IA runs on the client host and accepts the file system calls sent by client. Then, IA processes these calls by dispatching, controlling or coordinating with other agents. 2) WA (Working Agent). WA accepts orders or applications from IA, moves to the target server to execute file operation, and then return the result of execution.
3) DMA (Domain Manage Agent). DMA is responsible for name, cache, property and space management and access control management in a domain. DMA can duplicate itself and actively move to target server in order to be close to the data to gain higher processing performance. 4) MMA (Main Management Agent). MMA is responsible for the management and coordination of all DMAs in MADFS. MMA and DMA can cooperate with each other to accomplish the management work in MADFS.
HIERARCHICAL AND CONVERGENT CACHE COHERENCY MECHANISM MADFS is designed for WAN, and then the design goal of MADFS is not only supporting the UNIX semantic but also reducing the traffic in network, in particular reducing the traffic in WAN. For that purpose, authors designed a Hierarchical-and- Convergent Cache Coherency Mechanism (HCCM). HCCM can reduce the communication in WAN, with supporting UNIX semantic.
In MADFS, every file has the “Read Lock” and “Write Lock”. The read or write operation only can be performed if a host has obtained the read lock or write lock. A host can cache the file in local buffer if it obtain the lock, the server can assign the host to release the lock. if it is write lock when the host is releasing, the host will write back to the server.
There is only one Main Consistence Server (MCS) in MADFS, and each domain has a Domain Consistence Server (DCS). The cooperation between MCS and DCS can guarantee to achieve the cache coherence in MADFS. The file lock management in MADFS is divided into two layers: DCS is in charge of domain and MCS manages all DCS.
As Fig. 3 showing, when a host wants to apply a file lock, it will first send a request to the DCS. After DCS receiving the request, it will check whether it has this lock. If the DCS do not have the lock, it will send the request to the MCS. MCS will assign the lock to the DCS as soon as receiving the request. As DCS obtaining the lock from MCS, it then assigns the lock to the host.
THE CACHE COHERENCY MECHANISM IN MADFS.
As Fig. 4 showing, let Host5, Host6 and Host7 have the read lock of file.dat, and at this moment Host3 want to apply the write lock of file.dat. The whole process of applying lock and releasing lock is as follow: (1) Host3 firstly creates an agent A, and then dispatches A to DCS1 with the task of applying the write lock of file.dat. (2) When A arriving at DCS1, A will check the lock record table of DCS1, and finds that there is no host that has the write lock of file.dat. Thus, A moves to MCS, and applies the write lock to MCS.
(3) When arriving at MCS, A will check the lock table on MCS. Let’s suppose that A finds DCS1 and DCS2 have the read lock of file.dat. According as the multiple-reading/single- writing protocol, DCS1 and DCS2 must release the read lock, thus A duplicates itself into two agents (assume they are A1 and A2), and A1 move to DCS1, A2 move to DCS2. (4) After A1 arriving DCS1, it will check the lock table of DCS1, and finds the Host5 in domain1 has the read lock of file.dat. Then, A1 moves to Host5, and requires Host5 to release the read lock. Host5 will release the read lock as soon as receiving the request form A1. While receiving the acknowledgement of releasing read lock from Host5, A1 will move back to DCS1, and then move back to MCS. A2 does the same process in domain2, and moves back to DCS2, MCS after receiving the releasing lock acknowledgement from Host6 and Host7.
(5) A1 and A2 exchange the information while arriving at MCS and A2 will terminate after having exchanged information with A1. A1 deletes the records that DCS1 and DCS2 have the read lock of file.dat, and add a record that DCS1 has the write lock of file.dat. After that, A1 moves to DCS1. (6) While A1 arriving DCS1, it will add a record in the lock table of DCS1 that Host3 has the write lock of file.dat. (7) A1 moves back to Host3 and informs host3 that the write lock request of file.dat is success. While Host3 getting the write lock, it will retrieve the data of file.dat from the remote server and save the data in local buffer. Then, Host3 can read and write file.dat in local buffer without communicating with server.
From the steps described above, we can find that MADFS can efficaciously reduce the communication overload in WAN and the cache coherence management overhead in DCS and MCS by using the mobility, replicability and cooperatility of mobile agent.
Reference Jun Lu; Bin Du; Yi Zhu; DaiWei Li; “MADFS: The Mobile Agent-Based Distributed Network File System”, Intelligent Systems, GCIS '09. WRI Global Congress onVolume 1, May 2009 Page(s):