Login | Register | 中文

Qihoo 360 Eliminates Response Time Bottleneck with PBlaze4

2016-06-23


    

The offline testing search response time is shortened by 20% The online business search response time is shortened by 10%
TCO and the total operation and maintenance costs of the system are reduced
Industry: Internet
Use Case: database of search engine
Qihoo 360 Technology Co., Ltd. (“Qihoo 360”) is one of China's largest Internet security software providers. It has set up a world large-scale cloud security system with cutting-edge technologies, to remove multiple security threats on the Internet, such as Trojans, viruses, malware and phishing scam web pages.
In August 2012, Qihoo 360 launched search engine service, in a bid to build a secure, clean and effectively competitive Internet search marketplace. To achieve this objective, the company has built a core search technical team comprising hundreds of engineers and owns more than ten thousand servers. The company’s huge web crawler (spider) system can grad as many as 1 billion web pages each day, and the number of high-quality web pages indexed by its search engine has reached tens of billions. CNZZ report shows that by the end of 2014, Qihoo's web search traffic acquired around 30% shares in the market.
In recent years, all kinds of search engines have been competing for the search market. Besides, users have higher demands on the search applications, and the most direct experience among users is the search response time.





The Challenges


Disk is a hardware most closely related to the search response time, and the performance and quality of the disk plays a decisive role for the response time of the upper-level search applications.


While constantly optimizing the algorithm of search engine, 360’s R&D team of search engine has also come to realize some serious problems in the traditional storage architecture (using single SATA SSDs as storage or using SATA SSDs to build RAID):


lAt present, the response time of search engine is accurate to microsecond. The traditional storage with SATA SSD has high latency, and traditional soft RAID will further increase the latency, thus storage infrastructure has become a bottleneck for the search response time;


In the traditional storage method, each server needs several SATA SSDs to configure a soft RAID0, thus more SATA SSDs are needed for each single equipment, increasing the failure rate of each device. Besides, RAID needs to be rebuilt in case of failures, resulting in a huge amount of workload for operation and maintenance;


SATA SSD throughput and Input Output Per Second (IOPS) have low performance, and multiple SATA SSDs in one server will lead the overall performance to be less stable, and even inconsistency can occur in some cases. This may further cause inconsistent performance of upper-level applications.



The Solution


In 360's search application architecture, the data storage is divided into two classes. The first class uses open source levelDB database to store primary search data, which requires the search response time to be within 10ms. The second class uses the database independently developed by 360, which requires the search response time to be within 30ms.


In the original hardware architecture, the underlying storage solution for levelDB uses 8 independent SATA SSDs to store data, while the underlying storage solution for database developed by 360 uses 5 SATA SSDs to build RAID0.


In view of the inherent problems in the hardware architecture and the challenge of search response time, 360 has replaced the 8 independent SATA SSDs in the primary search data with 1 PBlaze4, and replaced the RAID0 made up of 5 SATA SSDs with 1 PBlaze4 in the new storage solution.





Business Benefits



Improve the performance of business system


By virtue of PBlaze4's advantages of high performance, low latency and low performance inconsistency, the speed of database query has been greatly increased. The most direct benefit is that the search response time is reduced, the offline testing search response time is shortened by 20%, and the online business search response time is shortened by 10%.



Reduce workloads of operation and maintenance


The storage method using RAID0, which is made up of multiple SATA SSDs, increases the hardware failure rate of data center, and the RAID needs to be rebuilt in case of failures, this will lead to heavy workloads of operation and maintenance, and overall performance inconsistency will occur to the data center in the wake of offline operation and maintenance of servers. After deploying PBlaze4, only one module is needed for each server, and a single server has a lower failure rate, thus the single device and search cluster can become more stable, so as to largely reduce the workloads of operation and maintenance, and shorten the search response time.



Reduce TCO of the data center


With equal purchase costs, the PBlaze4 PCIe SSD solution can reduce the energy consumption as well as operation and maintenance costs in the data center compared with SATA SSD solution.


Qihoo 360 started to develop applications on flash memory solution very early, and the infrastructural architecture built with SATA SSD in early period is rather complicated and reorganization of RAID will result in performance inconsistency in the system. PBlaze4 has significant strengths in IOPS and response latency. After being configured with PBlaze4, not only the system performance has been substantially improved, its operation and maintenance has also become much easier. The maintenance costs and energy consumption of the data center have also been lowered. -- Operation and Maintenance Engineer at Qihoo 360



中文

Technical Support: support@memblaze.com

Sales Email: contact@memblaze.com