Tharindu Rukshan Bamunuarachchi wrote:
Hi all/jim,
On 9/17/06, Jim Cromie <jim.cromie@xxxxxxxxx> wrote:
Tharindu Rukshan Bamunuarachchi wrote:
> Hi all,
>
> What is the best method of improving the real time behaviour of an
> application.
>
'best' needs more elaboration.
- what constraints ?
We have trading system with 4-7 hops connected in a loop. When a
network packet reach each host, we need minimum latecey to wake up
application and do the processing. Processin time is very samll and
usually takes around 50-60microseconds.
I dont get that.
Best I can tell, 7 hops gets me entirely out of qwest.net's net,
and onto somone elses backbone, and a ping to my own wireless router takes
64 bytes from priv1 (192.168.1.1): icmp_seq=1 ttl=64 time=1.03 ms
64 bytes from priv1 (192.168.1.1): icmp_seq=2 ttl=64 time=1.04 ms
64 bytes from priv1 (192.168.1.1): icmp_seq=3 ttl=64 time=2.64 ms
64 bytes from priv1 (192.168.1.1): icmp_seq=4 ttl=64 time=3.71 ms
If youre measuring 50us latencies and regarding that as important,
how could 4-7 hops on *any* network be tolerable ?
on a 1.7ghz laptop, pinging myself takes 60us!
64 bytes from harpo.jimc.earth (127.0.0.1): icmp_seq=2 ttl=64 time=0.060 ms
64 bytes from harpo.jimc.earth (127.0.0.1): icmp_seq=3 ttl=64 time=0.055 ms
I have tested this with two HZ values;
1. 1000 set to HZ
2. 1500 set to HZ
for 1000HZ i got around 2000 microsecond latencey in best case
for 1500HZ i got around 1300 microsecond latencey in best case
old / new kernels ?
This new 2.6.9 kernel (Actually RHEL 4 update 2)
keeping old apps ?
We have minimized other appllications. Only application per host.
how hard/soft is your RT reqs ?
I think we need soft RT
I must say, Im somewhat puzzled ..
considering 1- trading system + 2- a 2.6.9 RHEL kernel,
this seems like a requirement put upon you by marketing in the hope of
making
it more palatable to big (corporate) buyers.
Perhaps that explains your need to 'turn the HZ knob to eleven', when
its only labeled to 10.
trading system does sound like soft-RT. The plane doesnt crash if youre
100ms late on the rudder / flaps / important things.
Why arent you more concerned with thruput ?
Youre clearly killing thruput when the system cant even take care of its
own business (at HZ=4000)
CONFIG_PREEMPT_RT is in recent mainline, as is PREEMPT_VOLUNTARY
I have settled PREEMPT_RT option
what is settled ?
Im pretty sure 2.6.9 had no PREEMPT_RT icbw.
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
