2012 VoIP With Virtualization

本文探讨了使用虚拟化服务器进行实时语音通信(RTP)的共识,考虑了声音质量、并发呼叫数量、是否修改RTP流等因素。虚拟化存在风险,特别是在与高负载数据库服务器共存的情况下。文章还提供了虚拟化解决方案的建议,并对比了不同虚拟化环境(如ESX、Xen、Jails)的表现。同时,文章强调了在实际部署前进行负载测试的重要性。
Hi FS Users, 

What is the consensus about using virtualized servers for real-time voice (RTP)?  Even up until a few years ago it was hard to guarantee the CPU cycles to the voice nodes. 
Virtualizing the signalling (SIP) has always been favourable in terms of HA. 

If the entire host is under control, is it safe to say - we can virtualise voice? 
Does it depend on what else is shared in the host - say it would be silly to put a high load database server on the same hardware node... 

Thanks for your inputs 

Chris 

--------------------

basically it depends ... on MANY things..


a) how much you care about your voice quality
b) how many concurrent calls 
c) do you modify the RTP stream or just pass it through.

A)你有多么在乎你的声音质量
b)有多少个并发呼叫 
c)你修改的RTP流,或者只是把它传递

basically if your real picky or have more than a handful of concurrent calls.. then dont do it.
if your using conferences etc... dont do it.
基本上,如果你真正的挑剔或以上的并发调用了一把.. 不这样做。
如果您使用的会议等。.. 不这样做。

that being said, you can get away with a lot if you dont mind the occasional blip and chirp in your media..
这样说,你可以走了很多,如果你不介意偶尔在你的媒体的消息,并啁啾..

Thing to not though... above is for virtualization ... ( ESX, Xen, etc )

Jails like OpenVZ, BSD Jails, etc work well and provide many of the benefits found in virtualization.

Jay

-------------------

Hi, Chris

That depend of your enviroment. First you have to take a sort of scientific approach: define, trial and test again and again until the result be consistent.
Second tell us about your enviroment:
    - Real hardware: the server that will run your vm. ¿Are blades, small server, etc? VM software used (VMware, OpenVZ, XEN, Jails...)?
    - ¿How you set your VM? cpu, cpu cores, memory, disk, #nic?
    - ¿How many calls have you handle in the real world?
    - ¿Transcoding?
    - ¿Have you tested your VM?


-硬件:服务器将运行你的虚拟机。¿刀片,小型服务器等?VM使用的软件(VMware的,OpenVZ的,XEN,监狱...)     

- ¿如何设置你的虚拟机呢?CPU,CPU核心,内存,磁盘,#网卡?     

- ¿在现实世界中,你有多少呼叫处理?     

- ¿转码     

- ¿你测试你的VM虚拟化环境的VoIP的增长。

Virtualized enviroment for VoIP are growing. I am consultant for contact center services, and I have receive a lot of questions about virtualization. A solution provider can give some guidence about sizing and technical considerations but at the end the final word is:  YOU HAVE MAKE A LOAD TEST TO TEST YOUR ENVIROMENT AND IDENTIFY POSSIBLE ISSUES BEFORE GO TO PRODUCTION!!!!

Virtualization solutions will  do their homework to make VoIP a secure option. Look this paper for wmware:
http://blogs.vmware.com/performance/2012/01/voip-performance-on-vsphere-5.html
http://www.vmware.com/files/pdf/techpaper/voip-perf-vsphere5.pdf


--------------------------------------

tl;dr;  3 opportunities, none are urgent.  virtualization isn't a need-to-have but it would be nice to know what is do-able these days... i've had the best part of 4 years away from 100% voip, and spent 2 years in academia on gentoo plus haven't worked for nearly 12 months just faffing about with J2EE stuff and a bit of recent asterisk/kamailio and now freeswitch.



Hi Dario - I am creating a load test at the moment for a queue of 90 calls and comparing FS with asterisk for queuing - just reading over my old notes for SIPp.  I am very familiar with asterisk through to 1.4 but have been doing plain old web sys admin with gentoo in academia for 2 years and stayed away from voip and the industry - but times/opportunities are changing up to get back into it.   I have lurked in the FS irc channel for about 6 months whilst doing this other stuff and now is the time to pull the finger out so to speak.

Current opportunities are 2 potential network providers and a start up i'm the tech guy for:
A) This 90 call queue scenario will not be virtualised as it probably would not suit queuing which I understand to be a bit hungry for resources - it will be a failover set up for an Avaya where the avaya is on-site and this hosted-queue will be at the DC.

B) I am working on another startup that does purely conferencing and sms (hence my persistence with mod_gsmopen over the last couple of days)

C) Am getting some information on a hosted pabx environment where the current set up is 2 asterisk boxes with 700 registrations and about 60 concurrent calls and where to go with it.  the people that run that show haven't told me too much  - i.e. we want to give access to "IT Guys" to administer their own customers, we just want to bill the minutes  VS  they have their own hosted apps and want to keep everything 100% theirs.

A)这90个呼叫队列的情况下将不被虚拟化,因为它可能不适合排队,我理解是有点饿了的资源 - 这将是成立的Avaya Avaya是现场和本托管故障转移队列将在DC。
B)我的工作再次启动,纯粹会议和短信(因此我的持久性与mod_gsmopen在过去的几天)
C)我得到的一些信息在托管程控交换机环境下,目前的设置是2个星号箱700注册和大约60个并发呼叫和去用它。运行该节目的人,都没有告诉我太多 - 也就是我们想给获得“IT人”来管理自己的客户,我们只是想账单分钟与他们有自己的托管应用程序,并希望把一切都他们的100%。

The great thing with (C) is that I will be doing system engineering and no support.

I will read through the links you sent.  I am unfamiliar with virtualisation in terms of real world usage / performance etc. but have dabbled in a convenience-for-development  xen and virtualbox.
Using openvz in a "you can manage your own customer handsets" is appealing - i just had a quick read of it all.

But yes -- it's a load testing thing at the end of the day, and there can be no blips, or other voiping artifacts (i.e. digitized voice and the like)

In general:

1.  No transcoding (ever) - it will all be alaw all the way.
2.  All servers have Xeon CPUs.  Whether they are recent or from the last 5 years is another story depending on the availability of servers and/or cash.  I think there are more servers than cash to spend - rackspace is not a problem.  That being said, I just priced up some dells and supermicros to have some kind of figures ready.
3.  Currently, there are 60 concurrent calls across about 700 registrations.

1。无转码(永远) - 它都将是alaw的方式。
2。所有服务器都Xeon处理器。无论他们是最近从过去5年则是另一回事,这取决于服务器的可用性和/或现金。我想有更多的服务器比现金支出 - Rackspace公司是没有问题的。话虽这么说,我只是售价高达一些峡谷和supermicros的有某种数据准备的。
3。目前,有60个并发通话的约700注册。

Switches and routers are none of my concern, but they use a range of some cisco switches, a redback router, extremenetworks switch and some juniper firewalls will be in before end of year.

There is no rush on any of the above, it's all early days, plus I still have some of this non-computer project work I've been doing - manual labour upgrades...fun fun fun!!!

Thanks for your links and comments - I look forward to hearing more.

----------------------------------------------

I currently run a testing setup on rackspace cloud, doesn't seem to be too bad, but then again the call volume is low.

----------------------------------------------
For the setup in question:
 
The hardware:
IBM SystemX 3650.
8 core Xeons, E5*.
48GB RAM
Virtualization is done by VMWare ESXi, mostly 4 or 5.
Network is all Cisco.
 
3 VM Boxes that reach an average of 400 calls during working hours.
FreesSWITCH there is very old(1.0.6).
4 CPU Cores
2GB
64 bit Centos 5.2
local dialplan XML
no transcoding
xml_cdr
very stable.
 
Servers have several virtual machines with a lot of different things running, FSs have priority on hardware resources...
 
We have some other VM boxes(FS Sep 2011) with less load(270 calls avg) that do trancoding(25 calls avg g729<->g711), no quality issues but the FS clock drifts a lot(up to 10 minutes a day), VMware's fault ->  http://www.vmware.com/files/pdf/Timekeeping-In-VirtualMachines.pdf
fsctl sync_clock solves the problem but might ruin CDRs so BEWARE!
 
There are several other VMs in quite a few hardware setups, but none of them has impressive call volumes...
 
P.S.: Please don't shout at the age of the installs... they have served us well! :D.

硬件:

IBM SystemX 3650。
8核心Xeon处理器,E5 *。
48GB的RAM
虚拟化是VMware ESXi的,多为4或5。
网络是所有的思科。
 
3 VM盒,在工作时间内达到平均400电话。
FreesSWITCH是很老的(1.0.6)。
4个CPU核心
2GB
64位元的CentOS 5.2
本地拨号方案的XML
不转码
xml_cdr
非常稳定的。
 
服务器有多个虚拟机运行了很多不同的东西,FSS对硬件资源的优先级...
 
与负载(270呼叫平均),trancoding(25个电话平均G729 < - > G711),无质量问题,但FS时钟漂移了很多(最多10分钟的时间,我们有一些其他的虚拟机盒(FS 2011年9月) ),VMware的故障- >  http://www.vmware.com/files/pdf/Timekeeping-In-VirtualMachines.pdf
FSCTL sync_clock解决了这个问题,但可能会毁了的CDR所以要小心!
 
还有几个其他的虚拟机在相当多的硬件设置,但他们没有令人印象深刻的通话量...

As said before, it depends heavily in your setup... We do virtualize FreeSWITCH and aside from a few glitches it has been good... We have approx 400 calls per VM. Later when I hit the office I can describe my setup in detail.

正如前面所说的,它在很大程度上取决于你的设置...  我们做虚拟化FreeSWITCH的,除了一些小问题,它已经好了... 我们每一个虚拟机有大约400电话。 后来,当我打的办公室,我可以描述我的详细设置。

http://freeswitch-users.2379917.n2.nabble.com/2012-VoIP-With-Virtualization-td7559541i20.html



需求响应动态冰蓄冷系统与需求响应策略的优化研究(Matlab代码实现)内容概要:本文围绕“需求响应动态冰蓄冷系统与需求响应策略的优化研究”展开,基于Matlab代码实现,重点探讨了冰蓄冷系统在电力需求响应背景下的动态建模与优化调度策略。研究结合实际电力负荷与电价信号,构建系统能耗模型,利用优化算法对冰蓄冷系统的运行策略进行求解,旨在降低用电成本、平衡电网负荷,并提升能源利用效率。文中还提及该研究为博士论文复现,涉及系统建模、优化算法应用与仿真验证等关键技术环节,配套提供了完整的Matlab代码资源。; 适合人群:具备一定电力系统、能源管理或优化算法基础,从事科研或工程应用的研究生、高校教师及企业研发人员,尤其适合开展需求响应、综合能源系统优化等相关课题研究的人员。; 使用场景及目标:①复现博士论文中的冰蓄冷系统需求响应优化模型;②学习Matlab在能源系统建模与优化中的具体实现方法;③掌握需求响应策略的设计思路与仿真验证流程,服务于科研项目、论文写作或实际工程方案设计。; 阅读建议:建议结合提供的Matlab代码逐模块分析,重点关注系统建模逻辑与优化算法的实现细节,按文档目录顺序系统学习,并尝试调整参数进行仿真对比,以深入理解不同需求响应策略的效果差异。
综合能源系统零碳优化调度研究(Matlab代码实现)内容概要:本文围绕“综合能源系统零碳优化调度研究”,提供了基于Matlab代码实现的完整解决方案,重点探讨了在高比例可再生能源接入背景下,如何通过优化调度实现零碳排放目标。文中涉及多种先进优化算法(如改进遗传算法、粒子群优化、ADMM等)在综合能源系统中的应用,涵盖风光场景生成、储能配置、需求响应、微电网协同调度等多个关键技术环节,并结合具体案例(如压缩空气储能、光热电站、P2G技术等)进行建模与仿真分析,展示了从问题建模、算法设计到结果验证的全流程实现过程。; 适合人群:具备一定电力系统、能源系统或优化理论基础,熟悉Matlab/Simulink编程,从事新能源、智能电网、综合能源系统等相关领域研究的研究生、科研人员及工程技术人员。; 使用场景及目标:①开展综合能源系统低碳/零碳调度的科研建模与算法开发;②复现高水平期刊(如SCI/EI)论文中的优化模型与仿真结果;③学习如何将智能优化算法(如遗传算法、灰狼优化、ADMM等)应用于实际能源系统调度问题;④掌握Matlab在能源系统仿真与优化中的典型应用方法。; 阅读建议:建议结合文中提供的Matlab代码与网盘资源,边学习理论模型边动手调试程序,重点关注不同优化算法在调度模型中的实现细节与参数设置,同时可扩展应用于自身研究课题中,提升科研效率与模型精度。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值