[Serusers] Hardware and Linux for SIP Express Router

['Jan Janak']

Hello,

On 26-03 14:00, Valery Shampal wrote:
> Hi, Jan
> 
> Thank you very much indeed.
> So quick answer :-) 
> 
> As from the page http://www.iptel.org/ser/ under SIP Express Router (ser)
> topic
> 
> ========================================================================
> Technical Information: 
> 
> C-Written. Ported to Linux (PC, IPAQ), BSD (PC) and Solaris (Sun).
> Throughput thousands of calls per second (CPS) on a dual-CPU PC (capacity
> needed to cover Bay Area) and hundreds of CPS on Compaq IPAQ. Support for
> both IPv4 and IPv6. Small footprint size: 300k core, all common modules
> (optional) up to 630k. 
> ========================================================================
> 
> A dual CPU Pc is mentioned. This was a "trigger" to ask the questions.

  Yes, we use a dual Athlon CPU for performance measurements. On this HW
  ser with simple configuration is able to do ~ 5000 CPS. The 5k CPS are
  stateful, stateless ser could do more.

> We will use it within some Test-Demo Lab along with Hammer, Hammer ST and
> PacketSphere
> products from Empirix (http://www.empirix.com). We are their value added
> distributors here in Israel.

  Unfortunately I don't know the products.

> So it might be thousands calls over IP in this Lab. As I understood from
> your answer, there are no
> firm limitations on PC hardware. In other words, one with 450MGhz CPU and
> 512MB 
> physical memory might be enough. Am I right?
> 

  No, there is no HW manufacturer limitation. HW configuration really
  depends a lot on the test scenarios. If you are going to have many
  concurrent transactions, you will need at least 4 kB of memory per
  transaction. So if you know duration of the transactions you can
  easily calculate how fast you will run out of memory.

  If you are going to use user location and will have many users
  registered simultaneously, you will need some additional memory for
  user location records (~2kB per record).

  Our tests showed that usually memory is the bottleneck. The more
  memory you have the longer your tests can run.

> So what about a dual CPU above?

   That's a machine we use for testing.

  Could you, please, provide us with more information regarding the
  testing ? We are also interested in such testing (especially if
  you can generate really high number of CPS or messages per
  second). Maybe we could provide you with some ser optimizations so
  both sides could benefit from it.

    Jan.