Commercial Bank Big Data Road Trilogy 1: Go to IOE
Commercial Bank Big Data Road Trilogy 2: Data Governance
Commercial Bank Big Data Road Trilogy 3: Intelligence
----------------------------------------------------------------------------------
Even if the technical advantages of BAT are lacking, traditional commercial banks can still make a lot of money when they go to IOE.
----------------------------------------------------------------------------------
Cost of IOE
For banks, the "I" of IOE refers to IBM's mainframe, which is about several million dollars, with an annual maintenance fee of 800,000 and a daily salary of 20,000. "O" is the Oracle database, which is charged according to the CPU, each 150,000. A good database server with 7 or 8 CPUs will consume nearly a million license fees. "E" is EMC's storage, which is charged by T (1024G). With a set of plans, it is about 50,000/T.
I participated in a technical discussion some time ago. At the meeting, I heard the person in charge of the testing department of a Chinese-prefixed bank talk about their structure and cost. They are based on the structure of IOE, with tens of millions of accounts, an average of 1 IBM mainframe, supporting Oracle database, supporting EMC storage, the total cost is nearly 100 million yuan, but it can only support 500,000 accounts. That is to say, for a single account, the cost of hardware is nearly 200 yuan.
Maybe everyone has doubts about this number. For 500,000 accounts, even if they trade 10 times within 1 hour, the TPS (5,000,000/3,600 transactions per second) is only about 1,400. Ordinary servers can do it, no mainframe at all. In fact, this is a lack of understanding of the business of the banking system.
Every transaction of a bank is generally understood in three layers: information flow, capital flow, and accounting flow (accounting entry), involving deposit systems, payment systems, etc. But in fact, for banks, the process is far from over. Every transaction must be supervised by the CBRC system and the People's Bank of China system, to handle credit reporting, to carry out risk assessment, to calculate financial indicators, to carry out Liquidity and position calculation, capital cost, profit analysis, cost allocation, money laundering, filing processing, customer information management, anti-fraud, tax processing, etc., and thousands more Thousands of reports. Therefore, the amount of computation and data storage after the transaction is huge, but essential. Lack of any one of these can have serious consequences, such as HSBC fined $1.92 billion in 2012 for failing to identify obvious money laundering. Therefore, from the perspective of the whole process, a mainframe running 500,000 accounts, the cost of 200 yuan per account is more credible.
If the bank can get rid of its dependence on IOE and reduce the cost to 1/10 to 20 yuan, the average cost of every 10 million accounts can be saved by 180 yuan * 10 million = 1.8 billion.
Difficulty going to IOE
The wish is good, but the reality is cruel. I have been shouting the slogan of going to IOE for many years, and the only ones who have really gone to IOE are the two Internet banks supported by Tencent and Ali, both of which are based on "distributed IT architecture". Based on this, the banking system was constructed.
However, traditional banks, due to years of IT construction based on IOE, have formed a huge dependence. Whether it is the accumulation of technology, the reserve of talents, or the ability to take risks, it is not enough to support this arduous task of going to IOE. In addition, they also have a historical burden that Internet banks do not have. The 100 million-level accounts need to be smoothly switched to the de-IOE structure while maintaining stability. Any error risk in the middle is enough to make one vote unsafe.
In 2015, the China Banking Regulatory Commission issued Document No. 39, urging banks to go to IOE, and stipulated clear quantitative indicators: in 2019, the proportion of IOE to go to IOE should be no less than 75%, and the annual budget should be no less than 5%. But in general, the thunder is loud and the rain is small. Not only is there a gap in technical reserves, but also in terms of willingness and difficulty.
So for banks, is it really impossible to go to IOE? the answer is negative.
Actionable plan to go to IOE
The twenty-eight principle
The author once worked for "I" for 4 years, "O" for 4 years, and then for 3 years at the bank "Go to IOE". I have more or less participated in the implementation of the IOE architecture of many systems, and deeply felt the "28 principles" in the process of IOE removal.
The crux of the IOE is to reduce the cost of the bank's IT system under controllable risks and technical capabilities. In fact, some systems put 80% of the energy, 80% of the difficulty, and 80% of the risk, but only reduce the cost by 20%. In other systems, 20% of the energy, 20% of the difficulty, and 20% of the risk can indeed reduce the cost by 80%.
Therefore, the way out for commercial banks to go to IOE at various disadvantages is to implement the system with 20% of the input and 80% of the output as much as possible .
System classification and comparison
From the perspective of de-IOE, the bank's system can be divided into two categories, online system and data system.
The main representatives of the online system are: deposit system, payment link system, loan system, capital transaction system, etc. Any system with a system response at the second or millisecond level can be included in the category of online systems.
The main representatives of the data system are: credit information system, fund transfer pricing system, asset and liability system, cost allocation system, profit analysis system, etc. Any system that can be recalculated every other day (commonly known as T+1 running batch) can be included in the scope of the data system.
So what is the difference between the two types of systems in terms of performance to IOE?
1. From a technical point
of view , removing IOE from an online system has very high technical requirements. It not only requires a distributed IT architecture, but also has very high requirements in terms of stability, fault tolerance and scalability. In the past practice, there were also many departments, nearly hundreds of people cooperated, stepped on countless pits, and it took several years of hard work to basically take shape. Therefore, for traditional banks, if it is not a diamond, don't do porcelain work.
2. From the perspective of risk
, the online system has a very high risk of going to IOE. A little accidental shutdown for a few seconds, or a calculation error, can lead to very large losses. After all, the amount in the bank account is wrong, and it is unacceptable for anyone to change it.
3.
The TPS of the online bank system is not as good as the Internet, and the current maximum is around 5,000. The business logic is not particularly complicated. After the information flow is processed, the capital flow and the accounting flow are basically over, so the amount of calculation occupied is not particularly large.
4. Data volume perspective
The data of the data system is generally copied daily from the online system, and then processed separately according to the requirements of different downstream systems. After this process, the data will expand rapidly, and the accumulated data volume of the data system is thousands of times that of the online system.
6. System selection
Due to historical reasons, basically traditional banks, these two types of banking systems follow the same IOE structure. So this gives us actionable space to go to IOE. Separating the data system from the IOE is a low-difficulty, low-risk, and high-reward thing. The big data platform is the real home of the data system.
Therefore, according to the principle of 28, our feasible plan is: The online system will continue to maintain the IOE structure, and the data system will be migrated to the big data platform. The benefits of doing so are obvious, at a minimal risk cost, while ensuring the stability of the system, it greatly reduces costs. Let's do the same calculation: For
a 100 million IOE mainframe solution, if only online transactions are processed, the average capacity will be increased by at least 10 times. For 5 million accounts, the cost of a single account is 20 yuan.
For the big data platform, with 5 million accounts, in actual production, only 200 servers of around 50,000 are needed to build a cluster to support all data system needs. The cost of a single account is 2 yuan = 10 million yuan / 5 million accounts.
Therefore, for the IOE+big data platform solution, the total cost of a single account is 22 yuan. Compared with the pure IOE solution with a single account of 200 yuan, the cost has been greatly reduced. From the point of view of cost saving, going to IOE can be considered very successful. .
Summarize
For commercial banks to go to IOE, the most economical, lowest risk, and least difficult solution is to separate the data system in the banking system from the IOE and migrate it to the self-built big data platform according to the 28 principle, reducing 80% of the investment by 20%. %the cost of. Since then, the big data road of commercial banks has really started.