Tair (Enterprise Edition) provides various data modules and enterprise-class capabilities. Shihuo is an e-commerce player that uses Tair to build an online business system to support various large-scale promotions and navigate millions of queries per second (QPS).
Customer remarks
"Tair (Enterprise Edition) helps Shihuo build a high-concurrency and low-latency system architecture that can handle frequent access to hot data and traffic spikes during major e-commerce promotions such as Double 11 and the 618 Shopping Festival. This facilitates Shihuo in providing timely and exciting shopping information for users who are interested in categories such as sports, fashion, and lifestyle." - Director of O&M, Shihuo
Customer profile
Shihuo is a website that specializes in shopping guides for discounted goods and is committed to providing professional online shopping decision guidance to a wide range of users. It provides product information across popular shopping malls at home and abroad.
Challenges
Ultra-high concurrency
As the business of Shihuo grows and its user base explodes, it is important to have a system that can support high concurrency at low latency. This ensures that Shihuo consistently delivers stable and optimal user experiences.
Traffic spikes on large-scale flash sales
Flash sales are a common marketing technique that is used by major e-commerce platforms. However, these events often result in a traffic surge. In that case, the demand for smooth user experience in a condition of intensive requests for popular products poses a severe challenge on the stability of the system.
Exponential growth of data volume
As competition in the industry becomes more intense, the cost of acquiring new customers increases. In this context, Shihuo is committed to analyzing existing users to better serve them. Based on user behavior data, Shihuo implements a personalized marketing plan for each user, which involves storing and analyzing massive amounts of data. This comes with an increasingly prominent cost issue.
Large keys and hotkeys
When you use Redis, large keys and hotkeys can be problematic due to insufficient business planning, accumulation of invalid data, or traffic spikes. Such problems may cause service jitter, degraded performance, poor user experience, and even widespread failures. In this case, it is crucial to identify large keys and hotkeys quickly and provide optimization solutions.
Solutions
Use Tair as the caching layer for a high-concurrency and low-latency system architecture
Tair provides high-speed caching technology to effectively improve system performance and availability. Tair is a cloud-native ready-to-use database service that can meet various business requirements such as high throughput, low latency, and elastic scaling. You can change the number of data shards in a cluster instance within the range of 2 to 256 and change the specifications of a data shard to easily support tens of millions of concurrent access requests.
Use the proxy query cache feature to support hot data access requests
In e-commerce flash sales, hot-selling products are natural hot data. Tair identifies hotkeys in an instance by using efficient sorting and statistical algorithms. Proxy nodes cache hotkey requests and responses based on the specified rules. If a proxy node receives a duplicate request within the validity period of the cached data, the proxy node directly returns the response of the request to the client without the need to interact with backend data shards, and therefore reduces network transmission overhead. This improves the read speed, reduces the negative effects of hotkeys on the performance of data shards, and prevents skewed requests. This way, Tair helps Shihuo navigate high-concurrency, read-dominant scenarios in which hot data is frequently accessed.
Select cost-efficient persistent memory-optimized instances
Alibaba Cloud has developed an in-house engine for data persistence based on Persistent Memory (PMEM) and launched the persistent memory-optimized instance series of Tair based on ECS Bare Metal Instance. Shihuo uses persistent memory-optimized instances of Tair to handle massive data generated by the business system. This instance series delivers nearly the same performance as Redis Open-Source Edition at 70% costs, achieving a better balance between performance and cost.
Use various features to identify large keys and hotkeys
Tair provides comprehensive, diverse, and powerful observability capabilities. The instance diagnostics feature of Tair combines performance metrics, slow query logs, and key analysis capabilities to provide a one-stop, end-to-end solution. It evaluates the health of Tair instances from various aspects, including performance levels, skewed requests, and slow query logs, and offers optimization suggestions. This greatly enhances the automated O&M capabilities of Tair instances, reduces usage costs, and enables users to quickly identify and resolve issues, ensuring system stability and reliability.
Benefits
As a caching layer, Tair supports high concurrency access in the range of tens of millions, helping Shihuo deliver stable and ultimate user experiences.
With the help of Tair, Shihuo can navigate traffic surges during major e-commerce promotions such as Double 11 and the 618 Shopping Festival.
Persistent memory-optimized instances of Tair offer a cost reduction of 30%, helping Shihuo achieve efficiency improvements and cost savings.
Tair provides a one-stop solution to address issues and challenges that may arise during its usage. This significantly reduces O&M costs.
DRAM-based instances of Tair deliver three times the read/write performance of open source Redis or Redis Open-Source Edition instances of the same specifications. This helps Shihuo overcome performance constraints caused by high-frequency read/write requests for hot data.