HIGH-PRIORITY (must be done to support SETI@home) - Network retry policies can't download file: when to give up? how to retry? exponential backoff can't upload file: when to give up/retry? can't connect to sched server error return from sched server - make scheduling server use fast CGI - proxy support HTTP, Socks Look at other open source code - team system in PHP - credit display in PHP - CPU accounting in the presence of checkpoint/restart core client periodically gets CPU time, accumulates in state file - test versioning mechanisms for core Idea: need to notify user if core becomes out of date. Send messages if either 1) some WU couldn't be sent because core is too old 2) no WUs could be sent because core is too old - Think about issues in update core client What if state file is incompatible? What if app versions are incompatible? - test reporting of app errors - test checksumming of executables - Think about making the set of scheduling servers dynamic one approach: allow "redirect" reply from server (requires old address to still work) second approach: "master URL" returns a file with list of scheduler address and email address of "problems" contact - complete client side and test hi/lo water mark scheme - initialize rsc_fpops and rsc_iops in client WORKUNIT - check server sends correct number of work units - check client requests correct number of seconds of work - measure hardware parameters: CPU speed, #CPUs, memory, disk - define CPU benchmarks - do this for other platforms - Get info periodically rather than always at startup - core client / app trigger file interface core->app init: app preferences name of graphics shared-mem seg recommended graphics size recommended frame rate recommended checkpoint period dynamic: whether to do graphics exit app->core init: actual graphics size dynamic: % done I just checkpointed - implement graphics system Basic idea: core client passes shared-memory segment name to app, recommended size and frame rate. App creates pixmap + dirty flag in shared mem, renders to it, sets dirty flag. Core client (or screensaver) maps shared mem. When dirty, blit pixmap to screen - X11 version It's not clear whether X11 allows you to draw into memory. Does everything have to go through server? - Win32-based version - The API by which the app draws into shared-mem pixmap should be platform-independent. In the first pass we can use our own library based on Win32 primitives. In principle we should be able to use any library capable of offscreen drawing. OpenGL version? - GUI and screensaver around core client design default (core client) display screensaver options move around screen? system tray icon behavior - get idle-only behavior without screensaver Windows UNIX - sched server should return total credit info (user, team?) could display in default core client display - edit user account -------------------------- MEDIUM-PRIORITY (must be done for CM) - review and complete documentation - test sticky files - test multiple-file applications change add.C to support multi-file applications - implement WU/result sequences -------------------------- LOW-PRIORITY - implement checkpoint/restart for file transfers use features of HTTP 1.1 Return an error to server if transfer fails (store in DB in server) - implement and test the batch mechanism - test environment-var mechanism - test alpha/beta/prod for apps - test on a multiprocessor - define logging system on client - server-side estimates of WU time requirements - DB entry for WU has int/FP/RAM/disk/net - server sends only feasible WUs - limit by RAM, disk - test - implement file upload/download requests - preferences finish PHP web interface specific preferences: disk usage % of free space? all except X GB? X GB? minimum checkpoint interval max bytes upload/download per day implement in client hours/days to communicate/compute implement in client control over priorities? - define a notion of "app preferences": arbitrary XML edited on server, passed to app client. This could be used to e.g. select a color scheme for graphics Could pass entire prefs? - NET_XFER_SET:: do_select: prevent one stream from starving others - add the ability to store input data directly in WU, and output directly in result (blob). For projects with small data. -------------------------- DONE (may need test) - mechanism for returning app stderr output to server? store in blob? - add size to FILE_INFO server, client sides - measure bandwidth on each network xfer maintain exponential average, weighted by #bytes xferred