Conference iosg::all-in-1_v30

2024.0. "MUTEX after Send to SUBSCRIBERS:" by COPCLU::ELIN (Elin Christensen @DMO, DTN 857-2406) Tue Dec 29 1992 13:39

    
    A customer with a 3-node ALL-IN-1 cluster and 1000 ALL-IN-1 users, is 
    experiencing a hanging process in MUTEX state after Send to SUBSCRIBERS:
    
    It only happens on one node in the cluster. When a user (yesterday it was 
    JJP) on that node sends a mail to SUBSCRIBERS, he gets a subprocess 
    (JJP_1) which will hang in MUTEX state.
    
    We did an ANALYZE/SYSTEM and looked at process JJP_1. 
    It had run out of BYTLM.
    BUFIO byte count/limit   for JJP_1 was   0/20256
    
    At the same time other user processes were hung in RWAST state.
    
    JJP had BYTLM=48000 set in UAF, and SYSGEN parameters PQL_MBYTLM and 
    PQL_DBYTLM were both 40000. The same values were set on the other nodes 
    in the cluster where they have not seen this problem. 
    
    $ SHOW MEMORY/POOL/FULL  showed that NPAGEDYN was too low, Initial Size 
    was approx. 7500000, and Current Size was 8300000.
    
    The system has a Main Memory of 320.00 MB, and there was a lot of free 
    memory.
    
    The other times the customer has had a process in MUTEX state after a 
    Send to SUBSCRIBER:, the customer has rebooted the system, but yesterday 
    while we were looking at the system over a modem, the process JJP_1 
    suddenly disappeared, and other ALL-IN-1 users could work again.
    
    We have recommended that NPAGEDYN should be raised to a minimum of 
    8500000, and that PQL_MBYTLM should be raised to 60000.
    
        1. Why does this problem occur?
        
        2. What should be done in order to prevent it from reoccurring?
        
        3. Has anybody else seen this problem?
        
    Elin

    elin,
    
    did STARS not show anything on this - I remember something like this
    happening in v2.4 but can't remember the details off-hand.
    
    regards,
    
    Andrew.D.Wicks

    Hi Andrew
    
    I have found a STARS article "ALL-IN-1 Process Locked in MUTEX Wait
    State" about an ALL-IN-1 process waiting for BYTLM. 
    That article is about a process that has a very low BYTLM (10000), and
    it is recommended that it gets raised to 36000  (which is the default
    in ALL-IN-1 account creation templates).
    
    At my customer site users have a higher BYTLM than recommended (48000),
    but anyway they run out of BYTLM and hang in MUTEX state when they send
    to SUBSCRIBERS: on a system with 1000 subscribers.
    
    Is sending to SUBSCRIBERS: an operation that requires a particularly
    high BYTLM so that we should be aware of that on large systems?
    
    Elin