What is a VPS?
The full name of VPS is Virtual Private Server (VPS). In fact, it is the physical machine on the open virtual machine.
Many IDC service providers like to give their VPS a silly-sounding name such as Aliyun’s ECS, such as Tencent’s CVM and so on.
Directly say * * * host * * * or directly say * * * instances * * * feel very difficult to accept in their corporate culture as, must be a name.
In fact, I think the reason is that AWS, the originator of cloud services to name their machines EC2.
Some people say ECS no matter what, and that’s typical of the idiots who have only used Aliyun.
Classification of VPS virtualization
There are 4 main categories of virtualization technologies on the market today:
XEN is typically represented by the early AWS and the old VPS service provider Linode, but both have migrated to KVM.
XEN is an application of Linux, and KVM is a Linux module.
The following service providers are using the KVM in addition to Microsoft.
Microsoft’s own virtual technology, exclusive to Windows. (You think Azure’s hens are a bunch of Linux?)
Hyper-V is not a big player in this field, but Azure has a huge market share.
OpenVZ is not essentially virtualization, but rather containers. The performance loss is minimal (almost negligible) compared to XEN and KVM, and memory/CPU/hard disk scaling does not require a reboot (XEN and KVM require a reboot to properly configure).
The biggest disadvantage of OpenVZ is that the virtual isolation is very low, and the hardware level of access is low or even can not have their own independent Linux kernel.
At the same time, the biggest problem with OpenVZ is not the technical aspect, but the fact that most of the service providers use OpenVZ for the purpose of memory overselling. What is the concept of overselling? Simply put, one physical machine with 32GB of RAM can open up 64 VPS with 1G of RAM.
In addition to the above, there are some non-mainstream virtual technologies, such as VMWare commercial solutions (of course, the price …). . But the various other virtual technologies combined have less than 1% share.
There are also several network considerations, such as latency, bandwidth, packet loss and so on:
Latency is essentially the same as physical distance when the routing is normal.
If conditions allow, it is ideal to choose the nearest city, such as Chongqing, the first to consider Alibaba Cloud / Tencent Cloud Guangdong / Shenzhen, or directly choose the Tencent Cloud Chengdu.
But if you want to choose a server room outside the mainland, then it is more complicated, you need to consider the different cities of different ISPs and so on. For example, the first choice of South Telecom / Unicom is to go to Hong Kong, the first choice of the North to go to Korea and Japan, and so on.
If you want to choose North America, the first choice is Los Angeles/Seattle/San Francisco in the western part of the United States, and the delay from Guangdong or Shanghai to the western part of the United States can be less than 150.
The level of bandwidth is directly determined by the level of network development in each region, in Europe and the United States equipment rooms are common G port (Europe even has 10G port equipment rooms) in the case of domestic rentals and sales by M bandwidth. (Or choose to be billed by the traffic can be opened to 100M or 200M bandwidth, but you need to face the sky-high traffic charges)
The vast majority of foreign IDC service providers will be in the purchase of VPS to a large number of traffic quotas at the same time a sufficient amount of bandwidth (generally in the hundreds of Mbps, 100M is rarely the case), for example, Linode DigitalOcean Vultr three of the cheapest package have 1T traffic:
At the same time, the three cloud service giants (AWS, Azure, Google Cloud) is the practice of liberalization of bandwidth (in many cases, bursts of bandwidth can run to full G port), and charge traffic charges. In addition to a number of non-mainstream IDCs, the domestic words are almost all uplink 5M-10M bandwidth of the machine.
3. open lines
special domestic network environment, AWS Cloudflare, Akamai, Google, etc. because you understand the reasons can not be established in the domestic CDN nodes, directly resulting in everyone access to these nodes need to go international exports, resulting in the mainland’s international exports are almost always in a state of congestion.
But **add money to the world can reach **well … China Telecom has a second-generation backbone network, CN2, and China Unicom has a Global VIP line. The access speed and stability of these lines is much higher than the access speed and stability of the general line room, just like hanging up. It’s not a good route, but it’s good to use less people so delay and stability are guaranteed.
VPS configuration is mainly CPU memory hard disk size and read and write performance and so on.
Different service providers to provide hardware performance differences is likely to be very large, it is possible that a 4-core machine compiles the same set of code is not as fast as others single-core machine compiled.
Hard drive then read and write performance is also very important, especially to run this kind of database. I have encountered 1000M/s read and write performance on good hard drives, and 5M/s on the worst ones.
The memory difference is not like the CPU and hard disk, the general performance difference is not too big,** but the non-ECC memory server do not buy .
If the purpose of buying a server is to do the station performance is something to consider, find a free time to run a **unixbench is a good way to determine the performance.
Branding of VPS
The brand of this thing is quite important, some of the old service providers can rest assured that you can buy but the new service providers, then it is not sure, the machine or network instability and even run away from time to time.
1. Established Service Providers
Represented by Linode,DigitalOcean,Vultr,Ramnode,Bandwagonhost.
- Sufficient guarantees for stable machine lines.
- :: Affordable and decreasing in price over time
- are pure SSD hard disk will not be like the domestic service providers as the default to give you stone disk, want SSD also have to add money.
- :: A certain degree of resistance to attack
- Too well known and used by too many people, especially internationals.
2. Domestic cloud providers
Represented by Aliyun Tencent Cloud
- Very desirable domestic room latency
- Domestic domain name needs to be filed
- :: Multiple machine lift configurations
- The most important thing to do is to make sure that you’re getting the most out of your money.
- Cannot be billed by the hour or if hourly billing will end up costing several times more than monthly billing.
- Either accept the small pipe bandwidth or pay per traffic.
- Domestic companies have technical limitations or * deliberately restrict certain features for profit*. For example, Azure’s free intranet high speed channel costs tens of thousands of dollars a month. For example, if someone only needs to click a button to lift the allocation, Aliyun will ask you to renew the fee for a month and so on. Even Alibaba’s failure to troubleshoot must be replaced after the IP address you dare to believe?
3. International cloud providers
Azure, Google Cloud and others.
AWS for specific reasons, after being focused on the quality of the network is very general.
As for IBM, either its Softlayer (for a while, the telecom went CN2, but could not resist the Chinese people and the presence of such rogues as RuiShi can only switch the line) or container cloud routing is quite unfriendly to the mainland, not in the scope of discussion.
- Specific server room is excellent for mainland network
- :: Hourly billing
- :: Sufficient bandwidth
- :: True clouds
- :: Complicated billing rules, but no rules.
- Problems may be encountered with domestic purchases
By the way, from the market share statistics on these four cloud services company is behind Aliyun, and now the mainland Internet start-ups to buy a server too much, hard to support a fourth.
4. National small company
National small companies are more complicated, some are really companies, while others are set up by one or more people (commonly known as oneman).
Oneman does not mean that you are only one person or that you are a studio. Maybe you have a few people or even set up a company, but your technical support your volume simply can’t operate like a company, hire a few part-time as customer service or even can’t afford to hire customer service a work order after a few days can’t respond to you how can you have the courage to say you are not oneman?
Hostigation for a long time only the boss himself a person, Ramnode is in fact a husband and wife, but there are people who dare to say others are oneman?
Oneman this kind of well, do not encounter problems, once you encounter problems light service suspension for a few days or service providers directly run away is possible. (For example, two days ago, my cloud.)
Not a few are even planning to cut leeks in the first place
This is why small national companies are not recommended to buy annual payments.
However, the advantages of a small national company are obvious: the location of the machines and lines are optimized to meet the specific needs of the national population.
In addition, this type of service provider is also very easy to attack, and is almost certainly attacking each other to steal customers, so not too many small companies were defeated.
I’ve been using it for 4 years now, and I don’t have any drawbacks except that the CPU and bandwidth are too expensive. The network quality is excellent, but it’s been deteriorating since Google cleverly extended the $300 trial from 3 months to 12 months over a year ago, and I can only say that Google doesn’t understand Chinese people.
Microsoft’s stuff is very good if you can afford it, but if you can’t then…. I’ll tell you what, the preceding example of writing a file at 5m/s is the cheapest version of Azure.
Internet quality is good, but nothing else.