I'm not so sure that we'll be running the rtp proxy piece. I'm thinking it'll all be public addresses and the endpoints we have already do nat transversal.
-----Original Message----- From: Maxim Sobolev [mailto:sobomax@portaone.com] Sent: Wednesday, December 10, 2003 11:31 AM To: Andrei Pelinescu-Onciul Cc: Darren Sessions; 'serusers@lists.iptel.org ' Subject: Re: [Serusers] Scalability Question
Some time ago at a request from one of prospective clients we had performed peak performance estimate for our rtpproxy. According to results, it should be able to handle up to 2,000 simulateneous G.723.1 sessions on a decent machine (P4 2.5-3.0 GHz). Please note that fine-tuning of OS network stack parameters can be necessary to get such high numbers, since RTP traffic consists of big number of very short UDP frames (up to 30 frames/sec for one session), so that network stack should be prepared to handle huge number of short packets.
-Maxim
Andrei Pelinescu-Onciul wrote:
On Dec 10, 2003 at 06:56, Darren Sessions dsessions@ionosphere.net
wrote:
Here is a little more insight into what I envision to be the core network. Hopefully this will help in the scalability question I'm desperate to get
a
better understanding of.
I'm just not sure if I've under-engineered this, or over-engineered it -
and
what I can get rid of to save money or what more I need to spend.
I don't know about sems, but ser, if tuned for performance can do 4500-4800 cps on a dual athlon mp 2000+. On your configuration (dual Xeon 3Ghz 1Mb cache) it should do much better (but I think it is very improbable to have 10000 calls in a second from 100000 users, at least not sustained more then a few seconds). You might consider adding more memory (4Gb so you can use 3-3.5Gb for ser). This only if you want to be prepared for the worse (DOS attacks that try to initiate as many calls possible to non-responding destination, creating a lot of transaction that will be deleted only after the final response timer hits (default 120s); ser uses about 5k/transaction ).
Andrei
Thanks again for all the help!
- Darren
Architecture = Intel i386
Bandwidth = DS3 to OC12
Population Size = Initially, a 100k user base at 10-1 ratio (10,000 simultaneous calls) and a 100-1 ratio for voicemail (1,000 simultaneous calls).
Scenario Implemented:
2x centralized database servers (For SER/SEMS) Running Redhat Advanced Server 2.1 - Quad 2.5Ghz 1MB Cache Xeon CPUs - 8GB Ram - 60GB Raid Storage on host - Multi-Threaded MySQL - Multiple Gigabit Network Interfaces
2x centralized storage servers (For VM, Web Content, etc) Running Redhat Advanced Server 2.1 - 2GB Ram - 140GB Storage on Host - 2.04TB Raid
Storage
- Multiple Gigabit Network Interfaces
4x SER servers Running RedHat AS 2.1 or RedHat 9 - Dual 3Ghz Xeon CPUs per server - 2GB RAM - NFS to Storage Server via Deticated Gigabit Network - Seperate Network card for Internet Connectivity - 18GB Raid Storage on
Host
4x SEMS servers Running RedHat AS 2.1 or RedHat 9 - Dual 3Ghz Xeon CPUs
per
server - 2GB RAM - NFS to Storage Server via Deticated Gigabit Network - Seperate Network card for Internet Connectivity - 18GB Raid Storage on
Host
6x WEB Servers 3x DNS Servers
-----Original Message----- From: Jan Janak To: Darren Sessions Cc: serusers@lists.iptel.org Sent: 12/10/2003 5:55 AM Subject: Re: [Serusers] Scalability Question
No, it depends on many factors like, architecture used, bandwitdth, population size, scenario implemented and so on.
Jan.
On 09-12 12:35, Darren Sessions wrote:
Is there a scalability guide anywhere I can use in helping me decide
how
much hardware to buy for SER + SER/SEMS application?
Thanks,
- Darren
Serusers mailing list serusers@lists.iptel.org http://lists.iptel.org/mailman/listinfo/serusers
Serusers mailing list serusers@lists.iptel.org http://lists.iptel.org/mailman/listinfo/serusers
Serusers mailing list serusers@lists.iptel.org http://lists.iptel.org/mailman/listinfo/serusers
-
Darren Sessions