Compiling Asterisk on Ubuntu server

I wanted to test something on Asterisk for a project and need it a new clean install.
Instead of setting up another machine i decided to use the power of my desktop machine ( an intel quad core with 8GBram) and go for a VM.
I decided to use QEMU for the VM as i am using it for embedded development for sometime now and it has proven a very reliable tool (plus its free :)
I created an Ubuntu server on the VM, but then i realized that the asterisk packages shipping with Ubuntu server were rather old.
So i downloaded asterisk, asterisk-addons, libpri and zaptel and started the building process.
For those of you who decide to go the same way, following are the packages you need to get asterisk with realtime (aka mysql support) to build.


sudo apt-get install build-essential speex libspeex-dev gsm-utils \
libgsm1-dev libgsmme-dev libgsm1 libgsm-tools \
libncurses5-dev ncurses-term libncurses5-dbg libsqlxx-dev doxygen

You will also need to install the linux kernel headers
so do a 'uname -r" and then

sudo apt-get install linux-headers-your-kernel-version-here

Once all these get install, follow the standard procedure to build libpri,zaptel,asterisk and asterisk-addons

Altec Telecoms - Charges beyond the "grave"

To many of you who do not know it, Altec Telecoms *WAS* a Greek VOIP provider that went bankrupt a couple of months ago.
I was using their service, a pre-paid one, and i was paying using PayPal for a DID line plus the costs of the calls i was making.
When it wend bankrupt, services to customers terminated and i had something in the range of 30 Euros pre-paid to them, which i pressumed that had gone to the "never-never land"
Then i started receiving these emails
To those of you that the first part looks...Greek, don't worry it is :)

English text follows


Αγαπητέ / αγαπητή SKORONEOS,

Με το μήνυμα αυτό σας ενημερώνουμε ότι η ισχύς του i-CallIn αριθμού 2116875170 που έχετε δεσμεύσει, θα λήξει την 13/12/2008, οπότε και θα αφαιρέσουμε το ποσό των 1 € από το υπόλοιπό σας για να ανανεώσουμε την ισχύ του για τον επόμενο μήνα.

Σε περίπτωση που το υπόλοιπο του λογαριασμού σας δεν επαρκεί, ο αριθμός θα απενεργοποιηθεί (δηλαδή δεν θα μπορείτε να δέχεστε κλήσεις σε αυτόν) αλλά δεν θα αποδεσμευθεί. Θα εξακολουθήσει να είναι δεσμευμένος για εσάς για τους επόμενους 6 μήνες.

Εναλλακτικά, μπορείτε να ενεργοποιήσετε / ανανεώσετε το i-Call Unlimited για ένα μήνα επιλέγοντας αυτόν τον αριθμό, οπότε θα ανανεώσουμε την ισχύ του για ένα μήνα εντελώς δωρεάν! Περισσότερα για το i-Call Unlimited διαβάστε στο site του i-Call (http://www.i-call.gr/unlimited.aspx).

Σας ευχαριστούμε!


Dear SKORONEOS,

With this message we are informing you that your i-CallIn number 2116875170 will expire on 13/12/2008. On that date, we will deduct from your account's "wallet" the amount of 1 € to renew this number for the next month.

In case your account's balance is insufficient, the number will be deactivated (you will not be able to receive calls on it), but it will not be released from you. It will continue to be reserved for you for the next 6 months.

Alternatively, you can activate / renew i-Call Unlimited and select this number, in which case we will renew the number for an additional month free of charge! Read more about i-Call Unlimited at the i-Call site (http://www.i-call.gr/unlimited.aspx).

Thank you!

A charge notification from a bankrupt company, that has shutdown operations !!!???? WOW !

Well it looks like Altec Telecoms is now in a "zombi" state, as the people who worked there have "taken over" the installations in order to "protect the company assets until the bankruptcy court comes out with a decision" i.e get some money as compensation.

The weird think is that although i can not connect to their sip.server, they still have internet connection and can send emails and also their TDM lines are working, so when you call the DID assigned to you get directed to the voicemail...

I thank God, i decided to go with the pre-pay plan and pay with Paypal and not with a contract/monthly credit card charge, as i believe i would have seen charges been applied to the credit-card and then go though the pains of contacting the bank for canceling them....

My .2 euro cent advice to Altec Telecom customers paying with credit card... check your credit card statments.

Preparing for my son's baptizing

There is a lot of activity in the house as we are preparing for my son's baptizing on November 9th.
We are expecting family from all over Greece and friends from abroad so preparations are underway for sometime now, but we are now moving to "overdrive"

OSLEC code optimization

While working on my first CUDA app, i had a look at OSLEC as i though it would be a good candidate for parallel execution..
OSLEC is an Open Source echo canceler, written by David Row and used by asterisk.
After profiling the code,I found 2 places that i thought could benefit from some optimization. These were 2 loops that OSLEC spends considerable time in.
The first looked like that, and its part of the FIR filter in OSLEC

int i;
int offset1;
int offset2;

fir->history[fir->curr_pos] = sample;

offset2 = fir->curr_pos;
offset1 = fir->taps - offset2;
y = 0;
for (i = fir->taps - 1; i >= offset1; i--)
y += fir->coeffs[i]*fir->history[i - offset1];
for ( ; i >= 0; i--)
y += fir->coeffs[i]*fir->history[i + offset2];

By unrolling it it becomes like that

int i;
int offset1;
int offset2;
int position;

fir->history[fir->curr_pos] = sample;

offset2 = fir->curr_pos;
offset1 = fir->taps - offset2;
y = 0;
// for (i = fir->taps - 1; i >= offset1; i--)
// y += fir->coeffs[i]*fir->history[i - offset1];


i=fir->taps - 1;
while ( i >= offset1 )
{
position=i-offset1;
y += fir->coeffs[i]*fir->history[position];
y += fir->coeffs[i-1]*fir->history[position-1];
y += fir->coeffs[i-2]*fir->history[position-2];
y += fir->coeffs[i-3]*fir->history[position-3];
y += fir->coeffs[i-4]*fir->history[position-4];
y += fir->coeffs[i-5]*fir->history[position-5];
y += fir->coeffs[i-6]*fir->history[position-6];
y += fir->coeffs[i-7]*fir->history[position-7];

y += fir->coeffs[i-8]*fir->history[position-8];
y += fir->coeffs[i-9]*fir->history[position-9];
y += fir->coeffs[i-10]*fir->history[position-10];
y += fir->coeffs[i-11]*fir->history[position-11];
y += fir->coeffs[i-12]*fir->history[position-12];
y += fir->coeffs[i-13]*fir->history[position-13];
y += fir->coeffs[i-14]*fir->history[position-14];
y += fir->coeffs[i-15]*fir->history[position-15];

i=i-16;
}

and guess what,after recompiling there was 10% speed increase in OSLEC execution.
Then i moved to the second part that OSLEC spends a lot of time, the coefficient calculation.
Unrolling this loop gave roughly another 10% speed increase.
Roughly 20% speed increase in total, that's not bad :)

To verify that the increase were not an x86 "fluke" i added the patches to the oslec code in E-IPBX and recompiled the WARP version.
(WARP is a powerpc based PBX appliance from PIKA that we are working on)

I kept the original speedtest program and build 2 more adding each patch.
Bellow are the results first the unmodified speetest,and then with the first patch and the last with both patches.

root@e-ipbx:/$ speedtest

Testing OSLEC with 128 taps (16 ms tail)
CPU executes 527.63 MIPS
-------------------------

Method 1: gettimeofday() at start and end
601 ms for 10s of speech
31.71 MIPS
16.64 instances possible at 100% CPU load
Method 2: samples clock cycles at start and end
31.71 MIPS
16.64 instances possible at 100% CPU load
Method 3: samples clock cycles for each call, IIR average
cycles_worst 140815 cycles_last 3535 cycles_av: 4036
32.29 MIPS
16.34 instances possible at 100% CPU load

root@e-ipbx:/$ ./speedtest2

Testing OSLEC with 128 taps (16 ms tail)
CPU executes 528.91 MIPS
-------------------------

Method 1: gettimeofday() at start and end
497 ms for 10s of speech
26.29 MIPS
20.12 instances possible at 100% CPU load
Method 2: samples clock cycles at start and end
26.29 MIPS
20.12 instances possible at 100% CPU load
Method 3: samples clock cycles for each call, IIR average
cycles_worst 100679 cycles_last 3046 cycles_av: 3933
31.46 MIPS
16.81 instances possible at 100% CPU load
root@e-ipbx:/$ ./speedtest3

Testing OSLEC with 128 taps (16 ms tail)
CPU executes 527.67 MIPS
-------------------------

Method 1: gettimeofday() at start and end
485 ms for 10s of speech
25.59 MIPS
20.62 instances possible at 100% CPU load
Method 2: samples clock cycles at start and end
25.59 MIPS
20.62 instances possible at 100% CPU load
Method 3: samples clock cycles for each call, IIR average
cycles_worst 142386 cycles_last 2824 cycles_av: 2781
22.25 MIPS
23.72 instances possible at 100% CPU load

Not bad for a few lines of code :)

A super-computer in your desktop

Today i decided to spend sometime and get NVIDIA CUDA working on my desktop machine ( A quad Intel Q6600@2.40GHz. with 8 GB of Ram running in 64bit mode)
I had worked with CUDA about a year ago, when it was at the early beta stage but things have moved on a lot since then.
Setting my GeForce 8800 GT on Ubuntu 8.04 took a bit longer than "usual" due to the fact that 8.04 is not supported "out of the box"
Luckily there are some very detailed pages that helped and in less than an hour, CUDA example programs were happily running.
The GeForce 8800 GT is rated at 350+ GFLOPS which is pretty impressive (consider that Cray-2 could do a "meager" 1.9 GFLOPS)
The only problem (as with a lot of the hardware these days) is that there is very little software to take advantage of it.
My reason of getting CUDA to work is to explore the possibilities of accelerating various pieces of code/applications i am interested in.
Echo canceling (OSLEC for example) and transcoding for asterisk, are high on the list and based on the experience i had of porting asterisk to an FPGA there could be some very interesting results.

Let's start blogging

Up till now i had resisted the idea of starting a blog.
The main reason i had as my excuse, was time or to better put it, the lack of it. Trying to run a small business and a family with a new born is enough to make you wish for more hours in a day.
But the real reason was that i though of blogging as just "hype" that will run its course and sooner or later disappear.

Then i attended the OpenCoffee XV meeting in Athens and heard Jason Calacanis presentation.
At one point he mentioned something like "the more you give to the community the more you will get back" and that really got me thinking.
My company, Digital-OPSiS, is developing embedded Open Source solutions for sometime now, so you can say that we own our existence to the mentality of giving to the community.
If the people who started the Open Source movement haven't done so, probably many of the wonderful things that we now take for granted in the IT/Internet would not exist.
I call it the "avalanche effect" and seen it many times with open source projects.
Someone writes a few lines of code, posts the code online,and boom, developers from all over the world start contributing and before you know it you have a killer app in your hands (Asterisk anyone :)

So the same could be true about ideas or thoughts
If Jason had not said that phrase, probably i wouldn't have started this blog,so just think how many people could start something new,solve a problem or just feel better by things you and i take for granted or just don't mention.
That alone i think is a pretty good reason to start blogging