Processes: Code Migration Chapter 3 Will Cameron CSC 8530 Dr. Schragger

Overview What is Code Migration What is Code Migration What does it offer a distributed system? What does it offer a distributed system? Models for Code Migration Models for Code Migration Weak Mobility and Strong Mobility Weak Mobility and Strong Mobility Sender and Receiver Initiated Sender and Receiver Initiated Migration and Local Resources Migration and Local Resources Binding Categories Binding Categories Heterogeneous Migration Heterogeneous Migration

Code Migration “Moving programs between machines with the intention to have those programs be executed at the target” (Tanenbaum and Steen) “Moving programs between machines with the intention to have those programs be executed at the target” (Tanenbaum and Steen) Traditionally code migration in distributed systems took place as process migration Traditionally code migration in distributed systems took place as process migration The execution status of a program, pending signals, other parts of the environment must be moved as well The execution status of a program, pending signals, other parts of the environment must be moved as well Entire process moved from one machine to another Entire process moved from one machine to another Costly and intricate task, only done for good reason: performance Costly and intricate task, only done for good reason: performance

Why Code Migration? Overall system performance can be improved if processes are moved from heavily-loaded to lightly-loaded machines Overall system performance can be improved if processes are moved from heavily-loaded to lightly-loaded machines –Where ‘load’ is CPU queue length, CPU utilization Load distribution algorithms played important role Load distribution algorithms played important role –Allocation and redistribution of tasks with respect to a set of processors –In modern DS optimizing computing capacity is less an issue than trying to minimize communication –Migration to a safer environment

Performance –It generally makes sense to process data close to where those data reside (get client server split slide) Example: Client-server system where the server manages a huge DB Example: Client-server system where the server manages a huge DB –If a client needs to do many db operations involving lots of queries –It may be better to transport part of the client application to the server and send only the results over the network –Otherwise it may be swamped with the transfer of data from the server to client Example: Migrating parts of the server to the client Example: Migrating parts of the server to the client –Clients need to fill in forms translated into a series database operations –Process the form on the client, sending only the completed form to the server –Reduce large number of small messages across the network –Client perceives better performance while the server spends less time on form processing and communication

Multitiered Architectures (1) Alternative client-server organizations (a) – (e). Alternative client-server organizations (a) – (e). 1-29

Parallelism –Code Migration can provide parallelism –But without usual parallel difficulties due to the independence of code copies Fewer worries concerning the overriding of a shared address space, etc Fewer worries concerning the overriding of a shared address space, etc Examples Examples –Implementing a web search as a small mobile program that moves from site to site –Make several copies of it, send off to different sites to speed search

Flexibility –Traditional Distributed Applications –Partition the application into different parts –Decide in advance what machine each part should be executed –Lead to multi-tiered client-server applications in chapter 1 –If code can move between different machines Can dynamically configure distributed systems Can dynamically configure distributed systems –Example: Server interface to a file system Client side implementation of file system interface needs to be linked to client application Client side implementation of file system interface needs to be linked to client application Thus the software would need to be readily available to the client at the time the client application is developed Thus the software would need to be readily available to the client at the time the client application is developed Alternative: server provide client implementation when the client binds to the server Alternative: server provide client implementation when the client binds to the server Client dynamically downloads the implementation, sets it up, invokes server Client dynamically downloads the implementation, sets it up, invokes server Requirements Requirements –Protocol for downloading and initializing code is standardized –Downloaded code must be compatible with client machine Security issue: (bring in Applets, security, etc) Security issue: (bring in Applets, security, etc) –Trusting the downloaded code only implements the advertised interface

Reasons for Migrating Code The principle of dynamically configuring a client to communicate to a server. The client first fetches the necessary software, and then invokes the server. The principle of dynamically configuring a client to communicate to a server. The client first fetches the necessary software, and then invokes the server.

Models for Code Migration –Framework of three segments Code Segment Code Segment –Contains the set of instructions that make up the program being executed Resource Segment Resource Segment –References to external resources needed by the process  Files, printers, devices, other processes Execution (Process) Segment Execution (Process) Segment –Store the current execution state of a process  Private data, stack, program counter

Weak Mobility –Transfer only the Code Segment with some initialization data Program always starts from its initial state Program always starts from its initial state –Java applets Benefit Benefit –Simplicity –Just requires portable, executable code for the target Matters whether the code is merely executed by target process or a new process is created Matters whether the code is merely executed by target process or a new process is created –Applets are simply downloaded and executed in browser space –No need for new process, communication at target –Drawback: target process needs to be protected by malicious or inadvertent code executions –OS solution: create a separate process

Strong Mobility –Execution segment transferred as well Running process stopped, moved to another machine, resumed Running process stopped, moved to another machine, resumed More powerful than weak, much harder to implement More powerful than weak, much harder to implement –Or Remote Cloning Yields an exact copy of the original process but on a different machine Yields an exact copy of the original process but on a different machine Executed in parallel to original Executed in parallel to original –UNIX: fork off a child process onto a remote machine –Benefit: –Model closely resembles existing, but on another machine –Simple way to improve distribution transparency

Models for Code Migration Alternatives for code migration. Alternatives for code migration.

Sender-Initiated migration –Migration initiated at the machine where the code currently resides or being executed uploading programs to a server: Dist. File System uploading programs to a server: Dist. File System Sending a search program across the Internet to a Web database server to perform queries Sending a search program across the Internet to a Web database server to perform queries –Often requires the client be previously registered and authenticated at that server Server must know all its clients b/c client will want disk: security Server must know all its clients b/c client will want disk: security

Retriever-Initiated migration –Initiative for code migration taken by the target machine Java applets Java applets –Simpler to implement, client takes initiative, downloading done anonymously downloading done anonymously server not interested in the client’s resources server not interested in the client’s resources Code migration to client done just for improving client side performance Code migration to client done just for improving client side performance Just server memory and network connections need be protected Just server memory and network connections need be protected

Migration and Local Resources This makes code migration difficult, resource segment cannot be simply transferred without being changed This makes code migration difficult, resource segment cannot be simply transferred without being changed –Process reference to a specific TCP port to communicate to another process  When moved to another location, must give up port and request a new one

Binding Categories –Binding by identifier Not a problem: absolute URL, still valid Not a problem: absolute URL, still valid FTP server by internet address FTP server by internet address Local communication endpoints also binding by identifier Local communication endpoints also binding by identifier –Bind by value Weaker process-to-resource binding, only the value is needed Weaker process-to-resource binding, only the value is needed Execution not effected if another resource provide the same value Execution not effected if another resource provide the same value Example: program rely on standard libraries, C, Java Example: program rely on standard libraries, C, Java Always available but exact location could differ Always available but exact location could differ –Bind by type Process indicates it only needs a resource of a given type Process indicates it only needs a resource of a given type References to local devices: monitors, printers.. References to local devices: monitors, printers..

The Resource-to-Machine Bindings –Often need to change the references to resources but cannot affect the kind of process-to-resource binding Can a reference be changed, depends on whether that resource can be moved and code Can a reference be changed, depends on whether that resource can be moved and code –Unattached resources can be easily moved, typically data files associated with program to be migrated –A fastened resource can only be moved with relatively high cost  Examples: local databases, complete web sites  In theory not dependent on machine, but infeasible to move –Fixed Resources  Intimately bound to a specific machine or environment  often local devices, local communication endpoint Establishing global references can be costly Establishing global references can be costly

Migration and Local Resources GR: Global Reference GR: Global Reference MV: Move the resource MV: Move the resource CP: Copy the value of the resource CP: Copy the value of the resource RB: Rebind process to locally available resource RB: Rebind process to locally available resource Actions to be taken with respect to the references to local resources when migrating code to another machine. Actions to be taken with respect to the references to local resources when migrating code to another machine. UnattachedFastenedFixed By identifier By value By type MV (or GR) CP ( or MV, GR) RB (or GR, CP) GR (or MV) GR (or CP) RB (or GR, CP) GRGR RB (or GR) Resource-to machine binding Process-to- resource binding

Heterogeneous Migration –Realistic situation in DS –Migration requires that each platform is supported for code segment –Issues Weak Mobility Weak Mobility –No runtime info to transfer, just recompile the source code, but generate different segments, one per platform Strong mobility Strong mobility –Transfer of the execution segment: PC, registers, stack, etc  Highly dependent on platform  Same OS and architecture –Solution  Only allow code migration at specific execution points: when the next function, method, or subroutine is called (C, Java)  Migration stack  The runtime system maintains a machine independent program stack  Updated when a subroutine is called or when execution returns from one  Identifier for called subroutine  Jump Label pushed as well  The address where execution should continue when the caller returns

Migration in Heterogeneous Systems The principle of maintaining a migration stack to support migration of an execution segment in a heterogeneous environment The principle of maintaining a migration stack to support migration of an execution segment in a heterogeneous environment 3-15

Heterogeneous Migration Heterogeneous Migration –If migration takes place when a subroutine is called Runtime system marshals all global program-specific data to form execution segment Runtime system marshals all global program-specific data to form execution segment Machine specific data and current stack ignored Machine specific data and current stack ignored Marshalled data transferred to destination along with the migration stack Marshalled data transferred to destination along with the migration stack –Only works if compiler generates code to update the migration stack whenever a subroutine is entered or exited –Generates labels in the caller’s code allowing a return from a subroutine to be implemented as a (machine-independent) jump –Suitable runtime system needed –Virtual Machines 1970s: Pascal machine independent intermediate code for an abstract virtual machine on many platforms 1970s: Pascal machine independent intermediate code for an abstract virtual machine on many platforms Did not catch on with other languages: C Did not catch on with other languages: C 1990s: Scripting Languages and Java 1990s: Scripting Languages and Java –All such systems:  Rely on a virtual machine to directly interpret code (scripting languages  Or interprets intermediate code generated by compiler (Java bytecode) –Drawback: we are stuck with a specific language

Conclusions Code Migration is an interesting way to improve distributed efficiency, through a sort of multiprograming provided by parallelism Code Migration is an interesting way to improve distributed efficiency, through a sort of multiprograming provided by parallelism Its various manifestations raise issues of security and process migration which must be addressed Its various manifestations raise issues of security and process migration which must be addressed Virtual Machines provide a nice solution to heterogeneous migration issues Virtual Machines provide a nice solution to heterogeneous migration issues