OSS provides multiple solutions for accelerating data access and controlling bandwidth usage. Use the table below to find the right solution for your scenario, then follow the links to detailed configuration guides.
Quick selection guide:
Need to accelerate static file delivery or on-demand streaming to Internet users? Use CDN acceleration.
Need low-latency cross-border or cross-region transfers? Use Global Accelerator.
Need to upload or download large files (GBs to TBs) over long distances? Use Transfer acceleration.
Need millisecond-level latency for AI or big data workloads on internal networks? Use OSS accelerator or EFC.
Need to prevent any single client from consuming too much bandwidth? Use Single-connection bandwidth throttling.
Need to guarantee performance for critical services in a multi-tenant environment? Use Resource pool QoS.
Performance acceleration
| Solution | Best for | How it works |
|---|---|---|
| CDN acceleration | Delivering static hot spot files and on-demand video or audio to Internet users accessing small and medium-sized files on websites or applications | CDN caches static data from OSS on globally distributed edge nodes. Clients retrieve data directly from the nearest edge node. |
| Global Accelerator | Cross-border data sharing, global media delivery, software updates, and real-time interactions such as video conferencing, online education, and real-time data synchronization | A globally deployed smart routing system directs each request to the nearest access point. The Alibaba Cloud Global Accelerator network then provides a low-latency, high-bandwidth path from the user to the OSS bucket. |
| Transfer acceleration | Long-distance uploads and downloads of large files (GBs to TBs), and fast delivery of dynamic or non-hot spot data | Globally distributed cloud data centers and smart routing direct requests to the nearest access point. Optimized network and protocols provide end-to-end acceleration over the Internet. |
| OSS accelerator | AI reasoning model downloads, hot data queries in data warehouses, big data analytics, and low-latency data sharing on internal networks | Hot spot files are cached on high-performance NVMe SSD storage, delivering millisecond-level latency and high throughput. Supports three prefetch policies: read-time prefetch, synchronous prefetch, and asynchronous prefetch. |
| Elastic File Client (EFC) | AI training with large sample datasets, AI inference with shared model files, and hot data queries for engines such as Spark and Presto | EFC mounts an OSS bucket as a local file system and builds a distributed cache using the memory and disks of compute nodes. Data is served from the local cache at millisecond-level latency. On a cache miss, data is pulled from OSS and cached. Multiple nodes share cached data over a peer-to-peer (P2P) network to avoid repeated downloads. A single machine reaches up to 15 GB/s throughput and 200,000 input/output operations per second (IOPS). Performance scales linearly with the number of nodes. Active prefetching loads data into the cache before a task starts. |
Performance management
| Solution | Best for | How it works |
|---|---|---|
| Single-connection bandwidth throttling | Preventing any single client from consuming too much bandwidth, and enforcing Quality of Service (QoS) in multi-user or public-access environments | Add the x-oss-traffic-limit parameter to a request and set a speed limit value between 819,200 and 838,860,800 bit/s. Supported operations: PutObject, AppendObject, PostObject, CopyObject, UploadPart, UploadPartCopy, and GetObject. |
| Resource pool QoS | Guaranteeing performance for critical services, differentiated QoS management, and large-scale multi-tenant environments where multiple services share OSS resources | Create resource pools and assign different resource quotas to various services for fine-grained management of OSS access. Supports bandwidth limits at the bucket, bucket requester, BucketGroup, and resource pool requester levels, with corresponding service quality guarantees. |