Conference iosg::all-in-1_v30

    There is now, in place and always has been, an easy way to split the
    load over multiple servers per node (not just cluster), there are also
    guidelines to tune the server, created from the tests the BOSE
    performance group did on the server.  Why don't you see them? Simple.
    for some reason the release notes for the FCS never made it into the
    A1 documentation. I believe this is fixed with a PFR.
    
    To split over multiple servers per node, it's simple, create mutliple
    servers via the IOS UI, giving each a unique (per node) DECnet object
    number and configuration file. Then decide which users you want to 
    have on which servers and have them put the following line into 
    LOGIN.COM:
    
    $ DEFINE OAFC$SRV_OBJ   xxx  (where xxx is the DECnet object number of
    		the server you want them to connect to).
    
    all servers will serve the same set of drawers (i.e. there is only one
    partition.dat per cluster). The users will see NO EFFECT of doing this
    other than possibly better response from the server.
    
    As for Tuning the FCS, there is a chapter in the FCS release notes 
    that has guidelines. Note the default parameters are for a small
    system. The guidelines are along the lines of:
    
    	raise param X by Y amount for every Z users
    
    Sorry, but I don't have access to an online copy of these any longer.
    
    --Bob

    Hi Bob,
                                           
    I like your suggestion of having multiple servers per node and it's
    quite simple to achive that and is very appealing idea.
    
    I have a question concerning a mixed environment ALL-IN-1 IOS and
    TeamLinks clients.  What the customer would like to do is split the
    users between servers ie <node>$SRV73 ALL-IN-1 IOS users and
    <node>$SRV128.  Is this possible ?  How can this be achived ?
    
    A couple of questions I would like to raise are :
    
    1. Would the alternate server startup automatically when ALL-IN-1 IOS is 
       started ?
    
    2. Whould alternate logfiles be created for the servers ie
       oafc$server.log and oafc$server_error.log ? How do we define
       alternate logfiles ?  The logical OAFC$SERVER_LOG if defined would
       cause the primary server to divert it's output to this logfile.
       Unless the logical is checked when the detached process is created.
    
    3. In the case of the OA$FCV process will this be used by all the
       servers ?
    
    My suggestions that I had made previous to your reply to solve some
    server problems was to stop/restart the server at night.  Having more
    than one server would mean fewer users would be effected by the
    stop/start during normal working hours and we would get better
    performance.  We could dedicate servers to departments etc. there by
    getting better performance.
    
    Now one last thing as the saying goes "There is no such thing as a free
    lunch" :-).  So are there any gotcha that we should be aware of ? Also
    management issues we should know about.
    
    Thanks for your help in advance, regards,
    
    Sunil

    ALL-IN-1 already has the capability for users to use a different server
    on their node, by using the logical Bob defined. With the currently 
    released version of TeamLinks V1.1 all clients are forced to use
    object 73. However, I think I can safely say that this mechanism will
    be slightly different for the next version of TeamLinks (RoadRunner),
    where they'll probably be a FCS object number defined in your WIN or
    OFFICE.INI. Therefore, from TeamLinks you will be able to elect what
    server you connect to, on a per user basis.
    
    By running A1V30START.COM, ALL servers on that node are started up
    automatically.
    
    A lot of people seem to think OA$FCV is related to the file cabinet
    server. This process has been around for a number of ALL-IN-1 releases,
    and is used as a mechanism for creating unique filenames on a system. I
    don't believe the filecabinet server uses OA$FCV, as it has it's own
    algorithm for creating the filename. Remember that files created by the
    FCS have a "I" prefix and consist of 8 characters. Unique ALL-IN-1
    filenames have a "Z" prefix and consist of 9 characters.
    
    The idea of running multiple servers and splitting them across
    departments is a good one. It also means that if someone is causing the
    server to crash for whatever reason, the impact is reduced and you have
    a better chance of isolating the problem down to a specific set of
    people. Half the problem with errors reported by the FCS is being able
    to identify how they happened, almost all of them are unreproduceable
    at will.
    
    
    Finally, remember that although you can split users/groups across
    multiple servers on your own node, there is no supported mechanism for
    getting a user/server to broker to another FCS running anything other
    than object 73. By this I mean that if you have a distributed
    environment and a user wishes to edit a remote shared drawer, although
    the local server they are using might be any object number, it will
    only try to connect remotely to a server running as object 73. 
    I hope I've explained that well enough !
    
    Kevin.

    re .3
    
    You can do this now by having the IOS users go to the 128 server and
    have the TL users go to teh 73 server. As of RoadRunner, you will
    be able to have TL users specify which server they want by adding
    a line to win.ini.
    
    The FCS will write to the file defined by the logical OAFC$SERVER.LOG,
    in the absense of this it will write to SYS$MANAGER:OAFC$SERVER.LOG, 
    there is really no need to split them, but you could with a process or
    job logical. Each entry in the file will be tagged with the server name
    that entered it.
    
    The FCS does not use OA$FCV at all.
    
    Each server would take resources, if your system has the resources
    then go for it. As for management, it would not be any harder than
    managing one of them.
    
    Kevin also alluded to another subject to be aware of. There is no
    supported way to get a FCS to broker to any object number other than
    73. All server to server connects will go to a remote server running
    as object 73.  If you don't have the DSO installed, then this is not
    a problem. If you do have it, it still is no real problem, just
    something you may want to consider when figuring out your load balance.
    
    --Bob

    G'day All,
    
    I would like to round off this note to basically give you the benefit
    of a large customer site.  One of our major customers has been using
    multiple FCS for the last few months and they have not had a single
    problem.
    
    Prior to this they had several problems that ment that they had to
    shutdown their server during core time.  By making use of this option
    they have a stable environment where they do not have to shutdown the
    servers and the problems that they had experienced no longer manifest
    themselves.  The added bonus has been performance has improved, that is
    the customers perception and I have no figures to prove it one way or
    the other. Finally management in a multi-server environment was not an
    issue.
    
    Regards,
    
    Sunil