| Author |
Message
|
| sebastia |
 Posted: Tue Sep 09, 2008 2:46 am Post subject: what is bad if DISCINT = 0 in a ? |
 |
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
Is there anything wrong if we code
DISCINT=0
in a TYPE(CLUSSDR) channel ?
I mean, I understand this parameter in the old RDSI links,
but now ... the cost of maintaining the channel always active
is close to zero.
Or ... the other way :
is there any benefit to have DISCINT=6000
in a cluster sender channel ?
Thanks. Seb. |
|
| Back to top |
|
 |
| AkankshA |
| Posted: Tue Sep 09, 2008 2:55 am Post subject: |
|
|
Grand Master
Joined: 12 Jan 2006
Posts: 1494
Location: Singapore
|
In a MQ cluster, cluster channels are triggered automatically. No point having discint =0
for normal p2p also i always prefer having triggering for starting the channels rather then having discint=0.. it does eat up resources..
_________________
Cheers |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Sep 09, 2008 3:00 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
I see triggering does eat resources ...
If the channel is already active, there is NO triggering done => it is cheaper
What resources are wasted ...
... when the channel is connected but without data ?
Some "keepalive" traffic ?
If we have NO DATA TO SEND, the channel is empty,
so sending those "keepalive" packets means no problem
to the "data" traffic,
as THERE IS NO DATA TRAFFIC ...
Your opinion, please ... |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Tue Sep 09, 2008 3:04 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
An Inactive channel can't be effected by temporary network problems.
In a cluster do you want every channel to every QM running constantly? Running channels do use resources.
The CLUSNDR is only used once to connect to the FR or if you issue REFRESH CLUSTER REPOS(YES). Why do you want that channel running forever? Actually, I think that channel immedialty stops and is restarted based on the FR's CLUSRCVR, so a CLUSSNDR's DISCINT value is irrelevant. I think.
Or did you mean DISCINT=0 on the CLUSRCVRs in your original question?
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Tue Sep 09, 2008 3:08 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
| sebastia wrote: |
Some "keepalive" traffic ?
|
Heartbeats. OK, those are tiny, but still.
And a running channel uses up threads in amqrmppa. It uses up memory. I think the performance support packs have stats on how much memory a running chanel takes.
An Inactive channel can't be effected by temporary network problems.
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Sep 09, 2008 3:24 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
Peter - I (am glad to) have say you are right.
(a) I saw the communications resources,
but not the computer's resources.
(b) And my actual customer has LOTS of "temporary LAN problems" !
( I can't say how did you know it .... ( )
Last week it took me almost 1 hour to copy MB 6.1 Toolkit source product
from one PC to the next one ... (1,5 GB)
OK - no more DISCINT=0 in the cluster SENDER channels ...
What is your opinion on "normal" Sender channels,
where I try to avoid the triggering that occurs when
next message comes and finds the channel disconnected ?
Seb. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Tue Sep 09, 2008 3:59 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
| sebastia wrote: |
What is your opinion on "normal" Sender channels,
where I try to avoid the triggering that occurs when
next message comes and finds the channel disconnected ?
Seb. |
Same opinion. Very few apps can't tolerate the second or so it takes to start up a channel, which will probably be happening only once a day in the morning as business starts.
Having said that, I do have a couple of QM to QM links that service apps that can't tolerate that overhead, especially when you consider the multiple hops round trip. On those channels I set DISCINT to 300,000. That will keep the channel running through the weekend. Everywhere else I use 900, a number I picked as a good compromise between constant retriggering and channels running too long into the night with nothing to do.
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Sep 09, 2008 4:12 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
|
| Back to top |
|
|
| Mr Butcher |
| Posted: Tue Sep 09, 2008 4:33 am Post subject: |
|
|
Padawan
Joined: 23 May 2005
Posts: 1716
|
i like to add another aspect that is an argument for using discint = 0.
we do have data that flows close to deadlines across channels. if we use triggered channels, we will recognize connection problems at the time the channel is started, which is then very close to the deadline.
if we start the channels at night or keep them running all day, we detect problems communicating with the remote end quite early (remote queuemanager not available, bad network changes, ....) which gives us time to investigate and fix the error a long time before the deadline is reached.
for triggered channels, we sometimes have less then 5 minutes, which is not enough to fix everything.
For some applications with lots of customers connected, we also keep the channels to the customer running all time, which then is like doing some monitoring for them. so it is quite often we have to call a customer telling him "your queuemanager is not started", getting a surprised "oh really" because the customer is not doing any monitoring at all. Well, of course it is not our task to do the monitoring for the customer, but it is a bit customer friendly if we are able to help fixing a problem at the customer end before it gets expensive (for him).
_________________
Regards, Butcher |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Sep 09, 2008 5:36 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
You see, Peter -
"An Inactive channel can't be effected by temporary network problems"
is BAD sometimes !
Thanks, mr Butcher - I also prefer to do some "preventive"
or "anticipating" monitoring
instead of letting the customer call me with the problem ...
So, DISCINT=0 at "normal" (sender) channels ..
and DISCINT> 0 for sender cluster channels
Sounds great to me ... |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Tue Sep 09, 2008 7:30 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
there is no right answer. But, when that other side takes down their QM for scheduled maintenance (that they conviniently forgot to tell you about!) at 2 AM and your channel goes retrying, and presumambly pages you....
DISCINT = 0, never. Honestly, you are saying you want the channel to run forever?!
DISCINT = a large number to cover several days, ok if dealing with apps that can't deal with a 1 second delay
DISCINT = ~15 or so minutes for all others
An MQ guru years ago told me to use big numbers instead of zero for channels you want running all the time, especially in clusters, as it allows "orphaned" channels to gracefully time out and go away. While that may no longer be a requirement, I still don't like to code F-O-R-E-V-E-R for anything (Expiry included!!!)
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| gbaddeley |
| Posted: Tue Sep 09, 2008 8:45 pm Post subject: |
|
|
Jedi Knight
Joined: 25 Mar 2003
Posts: 2538
Location: Melbourne, Australia
|
My 2c is that DISCINT should be set to at least twice the longest interval between normal flows of messages. This will keep the channel running all the time it is needed (eg. during business hours) and save resources and susceptibility to network outages (eg. overnight) when the channel isn't needed.
eg. If at least one message is transmitted every 5 mins on average, set DISCINT(600).
_________________
Glenn |
|
| Back to top |
|
|
| sebastia |
| Posted: Tue Sep 09, 2008 11:13 pm Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
mr gbaddeley - I never had an idea of how often do messages flow
thry my channels ... I have no way to calculate the figure you are saying.
mr Peter - I understand your point that amwerr01.log can become full
because of this "forever alive" channel.
But I never use linear log ...
On the other side, when a channel becomes iddle and disconnects,
there is also an entry in this file !
So maybe discint=0 can produce LESS text in amqerr01.log !!!
I agree on large numbers for cluster channels.
Have to think on every real customer about "normal" sender channels.
Cheers. Seb. |
|
| Back to top |
|
|
| KeeferG |
| Posted: Wed Sep 10, 2008 12:54 am Post subject: |
|
|
Master
Joined: 15 Oct 2004
Posts: 215
Location: Basingstoke, UK
|
DISCINT=0 should never be set. As Peter points out forever is a very long time and will mean channels started accidently will run for ever.
Historically the recommendation was twice the expected interval between messages because channels would use up resource, leave open handles etc. After the introduction of HBINT to tidy up these connections after a short period of inactivity, the recommendations for DISCINT was for as long as necesary but never 0. If you speak to any of the Hursley guys they will tell you that channels coming down is perfect normal and applicaitons should really allow for asynchronous communication. Of course developers being who they are will tend to want their life to be simple and will always request neverending channels as they then assume they do not have to worry about incactive channels.
It becomes even more entertain when the applications check for destination availabilty before deciding which queue they want to put the message to and complain when a message gets stuck on the transmit queue becuase they hard code destinations rather than use clustering to do the routing for them.
_________________
Keith Guttridge
-----------------
Using MQ since 1995 |
|
| Back to top |
|
|
| sebastia |
| Posted: Wed Sep 10, 2008 1:09 am Post subject: |
|
|
Grand Master
Joined: 07 Oct 2004
Posts: 1003
|
| Thanks for your comentary, Keith. |
|
| Back to top |
|
|
|
|
