Subj : Re: Thread priority and policy
To   : comp.programming.threads
From : steve
Date : Tue Mar 22 2005 06:48 am

In article <1111468487.585118.195430@z14g2000cwz.googlegroups.com>,
 <hemarangarajan@gmail.com> wrote:
[ Please don't top-post, why will be evident in a moment ]
> i m porting it from Solaris 5.2 version
>so i was able to create the POSIX equvivalent of bounded and detached
>thread.

The Solaris term is "bound", not "bounded".  The equivalent POSIX
term is "system contention scope".

>in my application
>thread A recvs Multicast pkt
>thread B writes it into a journl file (mainly for bk up
>in thread c the rcvd pkt processed  and it is published  on a tcp
>
>i want to make thread A to have the highest priority than thread B and
>C

David asked some important questions that you should consider and
answer:

>David Butenhof wrote:
>> Hema wrote:

[ snip ]

>> > while processing i make the thread A as a RealTime thread by changing
>> > it class to SCHED_RR(shuld i go for SCHED_FIFO or SCHED_RR??)
>>
>> I don't know -- what are the realtime REQUIREMENTS of your application?
>> Presumably there are some, since you're using realtime scheduling.
>> Without knowing WHY and HOW you're depending on priority scheduling, we
>> cannot answer that question for you.

This is the core:  What problem are you attempting to solve by
changing thread priorities?  Is it just that "it seems like a
good idea", or is there some specific deadline that you must
assert priority processing?

Remember, real-time does not equal real-fast.  In fact, the
opposite is usually the case -- real-time systems are not fast,
but they are predictable.

[ snip ]

Also, looking at your threading usage, it seems like you have
designed a system to cause the greatest amount of system overhead
possible; a single packet forces a minimum of two context switches.
That doesn't scale well to higher traffic rates.

Is there some reason the design can't simply be single-threaded?
Based on what you've shared so far, a simple select() loop should
be able to do hundreds of thousands of packets per second, assuming
your journal system can keep up.
-- 
Steve Watt KD6GGD  PP-ASEL-IA          ICBM: 121W 56' 57.8" / 37N 20' 14.9"
 Internet: steve @ Watt.COM                         Whois: SW32
   Free time?  There's no such thing.  It just comes in varying prices...

.