[Spread-users] Buffers

[Jonathan Stanton]

On Tue, May 29, 2001 at 10:46:57AM -0400, Theo Schlossnagle wrote:
> On Tuesday, May 29, 2001, at 10:04  AM, Xabier Vázquez Gallardo wrote:
> > Custom code.
> 
> Okay.  You amy not want to buffer so much.  The reason for buffering in 
> most applications is to reduce disk I/O.  You will  have no disk I/O 
> with Spread, so a more steady stream of messages will probably be more 
> healthy for all apps involved.

A tiny bit of bursting could be good (like 3-4 httpd log messages at a
time) as it helps to make sure we can send full ethernet frames, but
beyond that I don't think it should help and it might hurt.

 > 
> >>> sometimes get disconnected from spread. How can I increse the spread
> >>> buffer?
> >>> Changing
> >>>
> >>> #define         MAX_SESSION_MESSAGES    1000
> >>> in spread_params.h?
> >
> > Can I change it into 10000? How will this change affect to spread?
> 
> You actually want to ask this message on the Spread list.  One of the 
> Spread authors will answer it better than I could.  (I Cced the list so 
> they can just chime in :-)
> 
> I _think_ you can increase it as high as you like, but it allows on 
> group member to be up to MAX_SESSION_MESSAGES behind another member.  
> This is not good in most cases, though it may not matter in your logging 
> scenario.  There could be other undesirable effects of raising this too 
> high that I am not aware of.  Jonathan?  Yair?

Yes that is the limit to raise. There is no harm to raising it except the
memory useage of the Spread daemon will increase as you are asking it to
buffer more messges for slow clients and that clients who are slow can be
further 'behind' in real time from the current messages being sent (again
just because lots are buffered)

We know of people who have raised the limit to several tens of thousands
in special circumstances without a problem.

Jonathan

-- 
-------------------------------------------------------
Jonathan R. Stanton         jonathan at cs.jhu.edu
Dept. of Computer Science   
Johns Hopkins University    
-------------------------------------------------------