Internet system operators in Africa - Infrastructure

 

March 2018, the first edition of R & D center design system PAYGo plan. Nigeria deployed in Africa.

 

The system is self-built private cloud deployment architecture platform + self-built SMS cat, which a private cloud platform without geographical restrictions; however the SMS modem device must be deployed locally, otherwise considered international SMS.

 

He was found a few short message service providers, prices are high, but self-built several times cheaper SMS cat / bar, technical support on the downside program /.

 

The pre-research: hire a local computer room, self-built room, near the National rental server three deployment plan;

 

Hire a local room: Nigeria's infrastructure is very underdeveloped, pull the cable cost is very high, the general wireless networks are walking (traffic card installed router), we talked about several local service providers to provide room rental, prices are high, and the service is stable and security unknown, the lower-priority programs;

 

Self-built room: was intentionally self boss, after all servers are in their own controllable range, there was choking, the company leased office also reserved locations of the room.

 

Domestically made three servers in the past, I thought about packing again tight, after being transported to all broke, hard drives are falling out, and appalling.

 

Considering the very unstable local power, power most of the day, so that the application program is only deployed SMS cat equipment, servers into a low power industrial equipment (capable of supporting the battery power after about 8 hours); in high-frequency devices / power down, run fairly stable at high temperatures.

 

SMS cat equipment from domestic procurement, and then bring their own past, you can use 16 cards at the same time. Said to be industrial, but the reach and stability is not Zeyang, many manufacturers can not solve the problem, only their own research and testing, toss enough to choke.

 

Nearby National rental server (current plan): operation and maintenance sector in Europe (Frankfurt) rented 3 units high with physical servers (virtualized 15 servers), set up enterprise private cloud platform, has deployed OTT complete system, the company and subsidiaries official website, Mall, PAYGo (CRM + WMS) and so on. Taking into account the subsequent other company systems business, set aside 70% of the resources (three light machine, network and IP spend a month 8000RMB).

 

By 2018, a full year of iterative development, PAYGo systems business, cloud platform stability, operation and maintenance response, messaging reconstruction program, coverage and payment solutions sales model to explore and so on are mature;

 

直至2019年初,公司结构大调整,运维部、测试部和研发部合并。

 

我接管了公司所有线上系统,并重新对服务器资源做了规划,运维人员租用了价格较低的服务器,成本从8000RMB/月降到了约2000RMB/月。

 

低廉服务器的服务商非常不专业,1个月内经常随意重启服务器,导致数据库文件损坏、硬盘坏道、网络被禁、被随意断网做测试等一系列问题,导致我们大量时间耗在折腾服务器上面,果断更换了服务商并重新规划部署方案。

 

当前部署方案:

-关键服务部署到阿里云ecs服务器;

-数据库换成阿里云实例主备高可用;

-采用阿里云+七牛云双备份数据;

-短信对接了阿里云、华为云、创蓝、云片,自建短信猫做应急备选;

-子系统全部Docker化,采用自建容器云管理系统环境安装、升级,实施了全套DevOps理念。

 

从0到1的过程虽然很艰辛,但是只要跨过这道坎后,一切都变得非常简单。虽然折腾了,但团队成长了,以当前的研发能力和第三方业务对接能力,随时可多国方案快速复制。

 

从法国到尼日利亚访问延迟大概200毫秒左右,整体体验还算过得去。今年“华为云”在南非布了点,据说到肯尼亚延迟在50毫秒内,纳入下阶段测试计划。

 

总结:

1、如果是新市场,基建就用业界已有的稳定产品,后面再替换成自己的,把精力放在开阔市场和产品优化改进上,前期真心没必要折腾。

 

2、用熟悉或大品牌服务器服务商,稳定、靠谱很重要,贵不了几个钱;

 

3、永远要有Plan B;

 

4、不要以为一下子就能做成一个大而全的系统,一个实用系统都是要靠不断的设计、体验实践、持续优化迭代出来的;

 

5、不要重复造轮子,在非洲,抓住合作机会,抱团发展才有出路。

 

- end -

 

Guess you like

Origin www.cnblogs.com/huige-you/p/11762788.html