Serverless pros and cons----a record of my assignment

Serverless pros and cons

I have been reading these two papers carefully and they both come from UC Berkeley, which is not strange that there have been two completely converse viewpoints from the same university. It is interesting that I noticed the scholar Joseph E. Gonzalez and Vikram Sreekanti appear in both papers. I do not know if they have different opinions in different times or they hold totally different views at the same time---pessimistically look at new techs which are viewed optimistically by themselves.

 

Back to the topic, there are some key points I summarized respectively for each paper.

One Step Forward, Two Steps Back:

1. serverless system omits the importance of efficient data processing;

2. serverless system hinders the pace of distributed system;

3. serverless system has limited lifetimes due to stateless;

4. serverless system has I/O bottlenecks;

5. serverless system has to communicate through slow storage;

6. there is no specialized hardware for serverless system.

A Berkeley View on Serverless Computing:

1. serverless computing could be better in the tomorrow, just like cloud computing in 2009;

2. serverless computing has more auto-scaling characteristics that could allow users pay for services not pay for machines.

3. serverless concepts improve the utilization of hardware through multiplex.

 

In my opinion, I am with the latter paper. I hold optimistic attitude to serverless computing. I am going to strengthen my view by following aspects:

 

Technology is market driven

It is said that Bill Gates once said, humans just need 640KB memory and that’s enough. Although the speech has been denied by the party, the speech is still been spread for a long time. The reason is that people cannot realize what the market really wants. When Amazon and Google developed different kinds of virtualization in the cloud and Google’s even better in a sense. However, market embraced Amazon’s low-level virtual machine approach to cloud computing. Then, many other companies followed this trace.

 

Robin Li, Tony Ma and Jack Ma are on the behalf of Chinese Internet market. Among them, Robin and Tony are adept at techs and Jack is totally layman. It is strange that Robin and Tony have pessimistically attitude to cloud computing initially while Jack has optimistic one. The result is Ali Cloud is the largest cloud platform in Asia and earn a lot from it. Nowadays, Baidu Cloud, Tencent Cloud are developing cloud service due to the market’s response.

 

Besides cloud services, almost all techs are driven by market. I am a loyal user of Tiktok and the App was created in 2016. At that time, the App had only a few functions like 15 second short video. Not until 2019 did Tiktok develop long video about 2 min for each because 15s is not enough for more information and the users have to cut a whole one into pieces. In other words, market demands drove Tiktok to develop long video techs and now the company permits us to do live broadcast. At first, the company may consider the server limits and funding limits which do not allow us to upload long videos. Later, the demand of market inflated and long videos and live broadcast came.

 

Cons are easy to overcome with flourishing perspective

Generally speaking, if we do not consider the cost of a new tech, the new tech will develop better. The best example is the US-Soviet struggle for hegemony in the last century, especially in Aerospace. I’m not going to talk about why they could ignore the cost but the experience is we can heap the hardware to solve any problems.

 

In our topic, paper One Step Forward, Two Steps Back stated several limitations as stated in summary section. Among them, 4,5,6 could be solved with the flourishing perspective.

 

1. serverless system has I/O bottlenecks: This limitation was put forward in 2018, while 5G tech was Commercialized in 2019. It is foreseeable that 5G could improve the speed of network. One of the key features of serverless computing is distributed storage, which is not any more a problem in data communication with 5G tech. Moreover, the number of calls supported by the Lambda function in a certain period of time will be more than before.
2. serverless system has to communicate through slow storage: Lambda functions have to get addresses which they want to communicate with from memory and this impede the speed of functions. However, in the future, there will be more high-speed cache or memory that applied to serverless computing. If the number of users is great, the profit will be larger which means we can use more caches to solve the problem.
3. there is no specialized hardware for serverless system: Similarly, there is no specialized hardware because there is no specialized survey to support or instruct how to design them. When serverless is applied in large scale, surveys and experiment will be carried out and hardware will be designed. All problems are no longer problems with funds injection.

 

 

Computing latent is profoundly needed

The arrival of 5G era makes many things possible. Autonomous driving technology, for example, need a great computing power which ensures the safety of customer. 5G overcomes the obstacles in signal transmission, serverless computing ensure the computing power (using BaaS). Autonomous driving is the potential application of serverless computing because the autoscaling ability of serverless computing will expand or compress the scale of servers which is a big thrift to the auto manufactories.

 

In a word, serverless computing is absolutely a thrift way for developers. In my undergraduate university, we are often distressed during course selection, because the server always crashes due to a large number of visits. If the school rents more servers, it will alleviate the crash problem. However, it’s unworthy for certain period of a semester. Obviously, serverless computing could solve the problem.

 

However, there indeed exists some reasons that limit the scope of serverless computing. In the future, information security will be a great challenge if this tech is applied in a big scope. Besides, some operations around kernel are dangerous if virtualization or containerization is not perfect and if the machine crashes, other users on this machine will be affected.