On Mon, Jun 16, 2008 at 1:22 PM, C?sar Davi <cesargxn at gmail.com> wrote: > Hi Benny, > > I am developing a SIP gateway of 8 channels. > I'm using siprtp demo, as is the example that most applies to my case > (at least I think), since the audio processing is done by an external > DSP (according to your article - "integrating third party media stack > with PJSIP), so, the ARM do the treatment of SIP signaling and the DSP > do audio processing. When I 8 simultaneous calls, the thread SIP to > consume 96% of power CPU (SIP signaling processing only, without > audio). Here's what I got when running siprtp with 8 calls on an ARM9 180MHz (AT91SAM9260) platform: real 1m17.984s user 0m2.560s sys 0m6.910s If my calculation is correct that's a ~12% average CPU usage. That's using latest pjsip and the following configure command: ./aconfigure --host=arm-926-linux-gnu CFLAGS='-O3 -msoft-float -DNDEBUG -mcpu=arm926ej-s -march=armv5te' --disable-floating-point So something is wrong in your setup IMO. Cheers Benny > I thank sincerely for their help. > > Cesar > > Date: Fri, 13 Jun 2008 22:59:17 +0100 > From: "Benny Prijono" <bennylp@xxxxxxxxx> > Subject: Re: PJSIP in ARM - CPU consumption > To: "pjsip list" <pjsip at lists.pjsip.org> > Message-ID: > <1879720d0806131459w65e5f582y72f5dc9ad1aed24a at mail.gmail.com> > Content-Type: text/plain; charset=ISO-8859-1 > > On Fri, Jun 13, 2008 at 9:11 PM, C?sar Davi <cesargxn at gmail.com> wrote: >> Hi All, >> >> I wonder if anyone has used the PJSIP in an embedded environment, with >> low requirements of memory and processing power (something around >> 200Mhz). For what I have seen the stack is behaving well in desktop, >> but I have not seen much discussion on embedded environments. I > > Really? This list has been running long enough long enough (more than > 3 years now) and I've heard many discussions about embedded use of > pjsip. About CPU requirement, I know for sure people who have > production hardware running pjsip on a 20 MHz MIPS (can't name them > for political reason unfortunately). And don't forget about mobile > devices as well, which from device capability is not dissimilar from > embedded. So I don't see any problems with this. It's just a matter > of configuring it correctly, and of course we all know this is never > straightforward with any embedded developments. > >> personally have faced some problems regarding the performance in an >> application using ARM9 and may be forced to abandon the use of PJSIP >> in my project. >> > > What is your expectation anyway? If you expect to run 8 simulteneous > speex calls on your 200MHz ARM9 then probably it will never happen. > And as I told you, siprtp is thread based so this probably is not the > ideal approach for your development. And why are you using siprtp to > measure up pjsip. What are you trying to do in the first place? > > I'm just saying that there are plenty of options with pjsip. There are > plenty of features. And each features come with their own cost (CPU, > memory) of course. So it's always a matter of choosing the right > feature to have, to fit in the capability of the device. And you > haven't explained what you're trying to do, so we can't give good > suggestions for now. > > Of course nobody forces anybody to use pjsip. There are plenty of > other choices. It's just a pity if you have to abandon pjsip based on > wrong experimentation. > > Cheers > Benny >
