[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference virke::mrmemo

Title:VAX MAILGATE for MEMO
Moderator:STKHLM::OLSSON
Created:Sat Feb 25 1989
Last Modified:Tue May 14 1996
Last Successful Update:Fri Jun 06 1997
Number of topics:216
Total number of notes:933

166.0. "mrmemo and cluster environment" by MLN08::MALACRIDA () Thu Mar 04 1993 11:41

 It looks like MRMEMO can have multiple servers connected to the same MEMO
on the same or different machine (right ?).

 So, let suppose there is a dual-host where I would like to start a
server on one machine and a second server on the other one, using of course
two different logical units.

 I would like to know if:

 - this configuration does make sense

 - if I can expect a sort of load balancing of the traffic sent from MEMO
   (I actually have some doubt here ...)

 - if some performance improvement can be expected 

 - if one of the two machine fails, the one remaining takes over the whole
   job in a transparent way

Thanks,

giovanni 
    
T.RTitleUserPersonal
Name
DateLines
166.1No load balancing but failoverSTKHLM::OLSSONAnders Olsson, SIP SwedenThu Mar 04 1993 18:5675
.0> It looks like MRMEMO can have multiple servers connected to the same MEMO
.0> on the same or different machine (right ?).

    Right. It's supported to configure up to 9 MRMEMO servers in a system.
    It doesn't matter whether the different servers are connected to the
    same or to different MEMO systems. But not to the *same* MEMO/GWY port!

.0> So, let suppose there is a dual-host where I would like to start a
.0> server on one machine and a second server on the other one, using of course
.0> two different logical units.
.0> 
.0> I would like to know if:
.0> 
.0>  - this configuration does make sense

    That depends on what you are trying to achieve. One restriction is that
    the MRMEMO servers must use different MR mailboxes. So you can NOT do 
    this:
	                    ----- MRMEMO ---
	MEMO/Gateway port <                  > MR mailbox
	                    ----- MRMEMO ---

.0>  - if I can expect a sort of load balancing of the traffic sent from MEMO
.0>    (I actually have some doubt here ...)

    No, not in any transpararent way. In the same way as only one MRMEMO
    server at a time kan use an MR mailbox, only one server at a time can
    be connected to a MEMO/Gateway distribution port. There is no "fan-out"
    mechanism in MEMO that allows messages to a distribution port to be
    fetched by several gateways.

    You can, however, set up different distribution ports in MEMO but this 
    requires that the MEMO users use the different addresses. I heard about 
    an installation that wanted to use DDS for address lookups (MEMO to MR
    recipient validation) but still allow X.400 messages to be sent to MRX
    for arbitrary recipients (not present in DDS). They solved it by running 
    two MRMEMO servers. When a MEMO user wants to send an ordinary message 
    it is addressed to VAX.xxxxx.yyyyy and when a message is to be sent to 
    X.400, another distribution port is used - X400.G=xxxxx..S=yyyyyy...etc.

	MEMO/Gateway port VAX. <----- MRMEMO ---> MR mailbox
	MEMO/Gateway port X400.<----- MRMEMO ---> MR mailbox

.0>  - if some performance improvement can be expected 

    There will be a performance improvement unless there is a bottle neck
    somewhere. If, for example, the SNA link from the SNA/Gateway is already
    saturated, there isn't much to gain. 

    Internally in the MRMEMO server there is very little parallel activities
    going on. E.g. while an SNA message is being sent, no Message Router
    activity is taking place. So if you have spare resources, you can
    increase the throughput by running parallel servers (limited by the
    restrictions above).

.0>  - if one of the two machine fails, the one remaining takes over the whole
.0>    job in a transparent way

    Since the parallel transparent setup is not possible, this is not 
    applicable but...

    MRMEMO *does* support a crude way of failover through the use of standby 
    servers. By running more than one server having the *SAME* server 
    parameters specifying a specific SNA session address, only one server 
    will be able to get hold of the SNA session. If the server having the 
    session goes away, the standby server will get the session and take over.

    This is described in the Operator's Manual (V2.1), section 2.5.1.

    This causes the standby server to log a failed connection attempt every
    clock_tick (default 30 seconds) so you would have make room for a large
    log file or increase the clock_tick interval time (DEFINE/CLOCK_TICK). 

    Anders