Re: Question about "fd" token

From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Tue Jun 16 2009 - 13:24:57 PDT

  • Next message: 李宏亮: "Re: Re: Question about "fd" token"
    The application may call cr_register_callback() at any time. So, the
    call that spawns the callback thread could happen either before of after
    the first time a checkpoint is requested for any given process.
    
    As you say the checkpoint command adds to the task list all the threads
    of the target process.
    If the callback thread was created by the application before the
    checkpoint request, then it will be included. If the checkpoint request
    is issued before the application has registered any thread-context
    callbacks then there is no callback thread to checkpoint (and no work
    for it to do if it did exist).
    
    While most tasks that one would want to do in a callback are best suited
    to use of the callback thread, there are certain things that one might
    like to do in a signal-context handler. For instance anything related to
    thread-specific data is best dealt with in a signal-context handler
    where the thread can manipulate its own data with out the need for locks.
    
    -Paul
    
    李宏亮 wrote:
    > Hello, Professor:
    >
    > Thank you very much for the previous answer.
    >
    > It's my fault about the way of expressing. Last time I really want to
    > know:
    >
    > I want to know how phase1 tasks are added to the task list.
    >
    > becase when we issue the command "checkpoint -p", which means the
    > "target_type == CR_SCOPE_PROC". so invoke "build_req_proc(pid_t
    > target)" to add all threads belongs to this process to the task list.
    > but the phase1 task, or we call it "callback thread" is not created
    > until we execute "cri_register_thread()", so the "callback thread" is
    > not belonged to the target process,then not added through invoking
    > build_req_proc(), so how it is added to the task list?
    >
    > By the way, the paper "BLCR's design & implementation" say that
    > "thread context callback " is a better way to implement BLCR's
    > callback mechanism. So is it enough for BLCR if we use only thread
    > context callback?
    >
    >
    >
    >
    >
    >
    >
    >
    >
    >
    > ===============================================
    > 快来和我一起享受TOM免费邮箱吧!看看除了1.5G,还有什么?
    > <http://bjcgi.163.net/cgi-bin/newreg.cgi?%0Arf=050602>
    > ===============================================
    >
    
    
    -- 
    Paul H. Hargrove                          PHHargrove_at_lbl_dot_gov
    Future Technologies Group
    HPC Research Department                   Tel: +1-510-495-2352
    Lawrence Berkeley National Laboratory     Fax: +1-510-486-6900
    

  • Next message: 李宏亮: "Re: Re: Question about "fd" token"