With the maturity of cloud computing, more and more applications are developing rapidly with the help of the cloud. Today, let's talk about the rendering industry. Cloud rendering is generally divided into CPU rendering and GPU rendering.
1. Trend analysis
1. Cloud-based development of rendering services
The reduction of rendering costs and the improvement of rendering efficiency are the most important competitiveness of the industry. Offline self-built has obvious disadvantages compared to cloud rendering.
2. Large computing power requirements and unreasonable resource utilization
The rendering nodes render one frame at a time. The larger the cluster, the faster the rendering speed, and the overall resource utilization rate is very low. However, the peak period resources are not enough to meet the flexible business form of the enterprise.
3. The scale of the cluster is heavy and the management and maintenance complexity is high
Rendering clusters, a large number of physical machine monitoring and alarm management are complicated, and consume a lot of manpower. Enterprises often invest a lot of manpower and material resources in maintenance.
4. High construction and maintenance costs, long on-line cycle
Building a private rendering farm requires the purchase of a large amount of IT resources, which is costly. The procurement of IT resources has a long deployment cycle, which cannot meet the needs of rapid business launch.
2. Business challenges
1. Difficulty in resource forecasting and high self-construction cost
The rendering business demand is a fluctuating curve. When the business is idle, resources are wasted; when the business is busy, resources are insufficient; building a private rendering farm requires the purchase of a large amount of IT resources, which is costly.
2. Low rendering efficiency
Based on general server technology, a single rendering task requires tens of hours; when the cluster is overloaded, the task delay fluctuates too much.
3. Various storage specifications
The rendering cluster accesses shared storage at the same time, which requires high bandwidth for shared storage; low-cost storage is required for mass data storage.
4. Low delivery efficiency and complicated operation and maintenance
The self-built environment has a long cycle, high installation and deployment complexity, and difficulty in later operation and maintenance expansion; a large number of physical machine monitoring and alarm management are complicated, consume a lot of manpower, and cannot focus on their own business.
3. Solution architecture
Business description
Rendering business is one of the important application areas of high-performance computing. For different business scenarios, rendering requirements are different, and the required computing requirements are also different. Huawei high-performance computing solutions can provide enterprises with end-to-end computing services.
Flexible and flexible high-performance cloud server
There are dozens of specifications such as ECS computing instances of various specifications and fat nodes based on bare metal (BMS) computing instances to choose from, matching the calculation and memory requirements of different HPC applications.
Excellent performance bare metal server
Provides excellent computing performance according to the dedicated physical server mode, without virtualization loss, and has excellent performance advantages. Users can realize the automatic provisioning of bare metal servers through the management console to meet the demand for flexibility and flexibility of HPC services.
Distributed and scalable block storage service
Provides elastically scalable block storage services based on a distributed architecture. Each elastic cloud server can mount cloud hard drives on demand to provide storage space for HPC computing nodes and storage nodes. A single disk supports a maximum of 32TB capacity, 30K IOPS, and 1TBps throughput.
Flexible and flexible services help improve efficiency and reduce costs
Provide API / SDK adaptation, flexible management of cloud computing, storage, and network resources; data express service supports fast import and export between the cloud and the cloud; flexible billing mode, purchase on demand, reduce costs and enhance competitiveness.
Fourth, the solution architecture
Rendering solution architecture
Provide high-performance, highly reliable, simple and safe computing, storage, and network services to meet the performance requirements of various rendering scenarios, shorten the rendering cycle, and improve corporate benefits
Architecture advantage.
Excellent performance
Industry-leading performance, providing 100G IB computing network, local 3.2T enterprise-level SSD disk, the largest local cache disk in cloud services; providing 96-core 2T memory fat node computing instance.
Rich examples
Supports virtual machine and bare metal deployment modes to meet enterprise deployment requirements for different applications.
Open architecture
Based on the open OpenStack architecture.
Three-dimensional protection
Seven-layer comprehensive and three-dimensional security protection measures (access, transmission, basic, network, virtualization, data, management), three-layer Anti-DDoS traffic cleaning, securely isolated private network VPC technology, and access control IAM technologies.
Five, program advantages
1. Full specifications and multiple types of high-performance on-demand instances
CPU/GPU instances with different configuration specifications meet customers' low-cost or high-performance needs. Industry-leading performance solves the problem of jitter in rendering time and ensures the stability of image rendering time.
2. OBS single bucket 1wTPS, single stream bandwidth 300MBps
OBS single bucket TPS can reach 1W, delay 10ms, single stream bandwidth 300MBps, and support HDFS and POSIX multi-protocol. Based on the elastic file service SFS, the maximum file transfer can reach 20GB/s.
3. Docking Batch to provide management job scheduling on the cloud
Docking with Batch services provides resource pool management and job scheduling capabilities on the cloud. Large-scale virtual machines can be quickly deployed in batches and used on demand, reducing operation waiting time and increasing efficiency.
4. Second-level billing on demand, multi-dimensional payment mode
Optimize scheduling efficiency, supply resources on demand, bill by second, and greatly reduce costs. Provide multiple payment modes including annual subscription, payment on demand, and bidding.