Introduction to the infrastructure of large Web site architecture

The AJAX-based attacks described in the previous article have many questions, such as not being able to cross domains and lighten the burden. Ajax is delivered through simple get and post, using Httpdebugger, fetching data, and then using the following scenario, By the way, write an example of the attack code. More than the traditional webform, we are more likely to construct some, in fact, for WebForm and Ajax processing and contract process is the same, Ajax data volume relatively small, faster.

Combining Sharppcap and HttpWebRequest we construct a reasonable normal IP packet past, the code is very long, we use pseudo code simple expression.

Request. CreateURL (Ajax processing page);

Request. Method= "getOrPost";

Request.refere= "Web page source";

Sharppcap.setlinkconnection (fake IP address);

String content = Request. GetResponseStream () If you are a bulk contract for each other's web as a multithreaded application (if it is http://www.aliyun.com/zixun/aggregation/16485.html "> DEDECMS), enough to bring down the DEDECMS database.

Text to the point:

For the last book mentioned to solve the problem A, we first explain the telecom company ADSL layout scheme. There is no Visio installed on the machine, so I simply describe the process in words.

ADSL User aè Enter username password è remote connection to account database (in Tianjin) è account database connection billing database and return form è if successful, connect PPPoE server and further connect billing database è authentication service and connect. There is no special place, but and QQ communication service is the same, is the use of a unified user authentication server, while the user authentication of the information database is read-only, we can think of anything from it?

The above is a simple example, the following start to talk about the specific architectural strategy, first of all mentioned in the previous issue a, we first take the user database as an example to explain and request.

The first is to do the user estimate demand, if we do the academic community, then this user is not very large, maybe we do not need to consider this, for the level of users, we temporarily set the user level to three, millions (M) and tens of millions of domain (S), as well as billions of levels (Q), and consider the user login verification and query common operations, the M and s to expand and understand.

As we all know, in this case, the load on user data is not feasible but not feasible, but how to maximize the guarantee of query and update and the data synchronization between the servers. Here we no longer explain how to optimize the index, only the initial framework of the scheme, the following scenarios described in the case of a full table query, you can use the scheme of partitioned view, you can specifically search for relevant information.

For the M-level, the existing DBMS is perfectly well laid out to meet the requirements. We need to solve the problem is actually dealing with IO problems, the processing scheme is relatively simple, the database file files are divided into disk storage (not partitions, is a different hard disk), according to the size of the load, we can properly control the number of hard disk and the number of file partitions.

For the S-level, the last processing scheme can not fully meet the requirements, this time we need to register and warehousing process to make a simple change, the solution is the concept of data hashing and partitioned views, specific concepts to Google, I do not elaborate.

There are three of our common scenarios. The first is the equal volume expansion method, on the basis of user Registration control, ensure that the user capacity of each library is not more than 5 million, after the second library, and so on, this scheme can ensure the effective expansion of the system, but can not guarantee that the data is effectively indexed. The second is the common area index scheme, in fact, and the first scheme has the same kind of say but the first scheme was optimized, according to the user name for storage. For example, we can build 26 of the database, according to the user name index to control the user data into which library. If the username is Crazycoder, then the user name of the data stored in the user table C, in the data storage can be very convenient according to the user name for the corresponding data query, Scheme II can effectively solve the data indexing problem. Scenario three is a more modeled scheme, combining scenario one and scenario two, to encode user IDs, not indentify Cloumn, we use a serialization scheme to store user names in encoded form, such as the username is Crazycoder, Our coding program is digitized by the algorithm, the Crazycoder in accordance with the C,r,a,.... storage as a digital index, and then partition storage, digital type of data in the database can be more efficient to be queried and updated and shared, combined with programme one and programme two this is programme three.