Partially Connected Operation

L.B. Hustonlhuston@citi.umich.eduP. Honeymanhoney@citi.umich.eduCenter for Information Technology IntegrationUniversity of MichiganAnn ArborABSTRACTRPC latencies and other network-related delays can frustrate mobile users of a distributedfile system. Disconnected operation improves matters, but fails to use networking oppor-tunities to their full advantage. In this paper we describe partially connected operation,an extension of disconnected operation that resolves cache misses and preserves clientcache consistency, but does not incur the write latencies of a fully connected client.Benchmarks of partially connected mode over a slow network indicate overall systemperformance comparable to fully connected operation over Ethernet.1. IntroductionAn important advantage of a distributed comput-ing environment is on-demand access to distri-buted data. Disconnected operation [8,11], aform of optimistic replication that allows the useof cached data when file servers are unavailable,has proved successful at providing this access tomobile users. Disconnected operation is espe-cially successful at hiding network deficiencies bydeferring and logging all mutating operations,replaying them later.Distributed systems are often designed to work inenvironments that provide high data rates and lowlatencies, but these assumptions are generallyinvalid in a mobile environment. Here, discon-nected operation has broad applicability, but issomething of a blunt instrument: by treating net-works as either available or unavailable, discon-nected operation does not account for the varyingdegrees of network quality encountered bymobile users.For example, even though AFS [6] caches aggres-sively and has good support for low-speed

[1]  John A. Kunze,et al.  A trace-driven analysis of the UNIX 4.2 BSD file system , 1985, SOSP '85.

[2]  Ben Shneiderman,et al.  Designing the user interface: supplemental materials , 1987 .

[3]  Mahadev Satyanarayanan,et al.  Scale and performance in a distributed file system , 1987, SOSP '87.

[4]  John H. Howard,et al.  On Overview of the Andrew File System , 1988, USENIX Winter.

[5]  Michael L. Kazar,et al.  Synchronization and Caching Issues in the Andrew File System , 1988, USENIX Winter.

[6]  Van Jacobson,et al.  Compressing TCP/IP Headers for Low-Speed Serial Links , 1990, RFC.

[7]  Mary Baker,et al.  Measurements of a distributed file system , 1991, SOSP '91.

[8]  Peter Honeyman,et al.  The Rx Hex , 1993 .

[9]  Peter Honeyman,et al.  Disconnected Operation for AFS , 1993, Symposium on Mobile and Location-Independent Computing.

[10]  Mahadev Satyanarayanan,et al.  Experience with Disconnected Operation in a Mobile Computing Environment , 1994, Mobidata.

[11]  Maria Ebling,et al.  Overcoming the Network Bottleneck in Mobile Computing , 1994, 1994 First Workshop on Mobile Computing Systems and Applications.

[12]  Garret Swart,et al.  A coherent distributed file cache with directory write-behind , 1994, TOCS.

[13]  Mahadev Satyanarayanan,et al.  Disconnected Operation in the Coda File System , 1999, Mobidata.

[14]  William E. Weihl,et al.  Lottery scheduling: flexible proportional-share resource management , 1994, OSDI '94.

[15]  James J. Kistler Disconnected Operation in a Distributed File System , 1995, Lecture Notes in Computer Science.

[16]  Mahadev Satyanarayanan,et al.  A Programming Interface for Application-Aware Adaptation in Mobile Computing , 1995, Comput. Syst..

[17]  L. B. Huston,et al.  Peephole Log Optimization , 1995 .

[18]  Carl Staelin,et al.  Idleness is Not Sloth , 1995, USENIX.