Item | Standard | Log Enhancement Edition | AI Search Enhancement Edition |
Key Features | Suitable for various scenarios, with faster updates for the Community Edition. | Higher write performance, lower cost | Enhanced AI support and higher stability assurance |
Applicable Scenario | Logging, search, analytics, AI enhancement, and other scenarios | Log analysis: Web logs, App logs, database logs, and so on. Metric analysis: server metrics, network metrics, storage metrics, and so on. Application performance tracing: Python, NodeJS, and so on. | RAG: Build retrieval-augmented intelligent Q&A applications by collaborating with large models, and so on. Vector search: Serves as a vector database for building various AI applications. Classic search: such as e-commerce search, full-text search, and so on |
Core strengths | More Elasticsearch versions of Community Edition are supported, with faster updates. It also incorporates some Tencent Cloud ES self-developed technologies. | Cost: The storage-compute separation architecture based on COS offers lower costs. Write: The proprietary index write acceleration technology improves performance by 3 - 5 times. | AI enhancement: Supports invoking Tencent's self-developed vectorization and semantic re-ranking models, Availability: Based on the self-developed kernel, it comprehensively optimizes query latency, concurrency, and sorting performance. Under multi-replica + multi-AZ conditions, it provides 99.95% availability guarantee. Vector pruning: Supports removing vector row storage and original high-dimensional vectors, significantly saving disk space. |
Architecture Patterns | A classic ES architecture and hot-warm architecture applicable to various scenarios such as logging/search/analytics | A storage-compute separation architecture and hot-warm architecture specifically designed for logging scenarios | A classic ES architecture for search scenarios |
Supported Version | 9.1.3, 8.16.1, 7.14.2, and so on | 7.14.2 | 9.1.3,8.16.1, 7.14.2 |
Advanced Features | Platinum Edition, Basic Edition, Open Source Edition | Platinum Edition | Platinum Edition |
Billing | Billed based on cluster node specifications, storage space, and number of nodes | Billing is based on cluster node specifications, storage space, and the number of nodes. Adopting a storage-compute separation architecture incurs distributed shared storage fees. (Lower cost in logging scenarios) | Billed based on cluster node specifications, storage space, and number of nodes |
Category | Features | Description | Standard | Log Enhancement Edition | AI Search Enhancement Edition | |
Supported Elasticsearch Versions | | | 9.1,8.16,7.14 | 7.14 | 7.14 | 9.1,8.16 |
Stability | SLA | Service Level Agreement (SLA), which commits to service quality, availability, and responsibility. | Up to 99.9% | Up to 99.9% | Up to 99.95% | Up to 99.95% |
| Full-link circuit breaking and rate limiting | Integrating elastic funnel and full-link memory circuit breaking, comprehensively considering CPU, IO, and memory conflicts, to support hybrid scenarios of high concurrency and large queries. | √ | √ | √ | - |
| Spread Placement Groups | Placing cluster nodes in different underlying hardware to enhance disaster recovery and high availability capabilities. | - | - | √ | √ |
AI Enhancement | Text vectorization model | Supporting the invocation of Tencent's self-developed text vectorization model. | - | - | - | √ |
| Semantic reranking model | Supporting the invocation of Tencent's self-developed semantic reranking model. | - | - | - | √ |
| Vector storage pruning | Supporting the removal of vector row storage and original high-dimensional vectors, which can significantly save disk space. | - | - | - | √ |
Storage cost | Storage-compute separation | Proprietary hybrid storage architecture, cache layering, intelligent Uninstallation, integrated hot and cold search, saving 50% - 80% in storage cost. | - | √ | - | - |
| FST Off-Heap | Proprietary FST off-heap storage, reducing heap memory usage and significantly improving single-node disk management capability. | √ | √ | √ | - |
| Deep data compression | Performing systematic encoding optimization for row storage, column storage, and inverted index dictionary files, and supporting fine-grained switch control, resulting in a storage cost reduction of 20%+. | √ | √ | √ | - |
Write performance | Write acceleration | Proprietary shared-nothing, self-contained read-write separation architecture, isolating resources while improving write performance by 5 - 20 times. | - | √ | - | - |
| Physical replication | Primary-secondary replica physical replication, eliminating replica redundant computation and improving write performance by 50%. | - | √ | - | - |
| Directed routing | Optimizing shard-directed routing to address high fan-out and long-tail issues in write operations, improving write performance by 30%+. | √ | √ | - | - |
Query performance | Query/IO parallelization | Proprietary multi-level parallel query framework, supporting all query scenarios and improving query performance by 3 - 5 times. | - | √ | - | - |
| Query interception and rewriting | Proprietary keyword query DSL interception and rewriting optimization, helping to reject unreasonable large queries and ensuring cluster stability. | - | - | √ | - |
| Adaptive bypass Merge policy | Merging and consolidating Segments to reduce random query I/O, thereby improving overall query performance by more than 2 times. | - | √ | √ | - |
| Query cache optimization | Significantly improving query concurrency capability, with QPS increased by more than 50%. | √ | √ | √ | √ |
| Query pruning | Optimizing node time series index query pruning by changing time range pruning to boundary-specific pruning, improving high-cardinality time series search performance by over 10 times+. | √ | √ | √ | - |
| Sort acceleration | Skipping unnecessary data comparisons when inverted lists are traversed based on the bkd storage architecture improves sorting performance by 5-10 times. | √ | √ | √ | - |
| Authentication acceleration | Optimizing X-Pack authentication performance, eliminating authentication hotspot overhead, and improving query performance by over 30%+. | √ | √ | √ | - |
Ops efficiency | Autonomous indexing | Industry-unique, supporting automatic shard tuning, intelligent rolling, query pruning, and fault self-healing to achieve fully managed indexing. | √ | √ | - | - |
| Online reindexing | Supporting second-level completion of number of shards multiple adjustments. | - | - | √ | - |
フィードバック