Abstract: The software part of Ali Cloud WebP picture solution is provided by CTAccel, and then integrated with ali Cloud’s own FaaS (FPGA as a Service) elastic computing platform, forming a complete Ali cloud WebP picture solution.

Click here to see the original article

At present, pictures still occupy a large part of the traffic, especially for mobile terminals. Therefore, how to reduce the size while ensuring the visual effect is very important to solve the problem of high bandwidth consumption and fast battery consumption of mobile phones. However, we are familiar with JPEG, PNG, GIF and other commonly used image optimization has reached the extreme.

Therefore, In 2010, Google launched a new image compression format, WebP, which aims to achieve the same image quality as JEPG while reducing the file size, hoping to reduce the transmission time of images across the network.

WebP provides lossy and lossless compression capabilities for images. According to official experiments, lossless WebP reduces the size of PNG by about 26%, and lossless WebP reduces the size of JPEG by 25% ~ 34% under the condition of ensuring structural similarity. Lossy WebP is typically about 34% of the size of the corresponding PNG.

Limited to JPG, PNG and other formats of image transcoding slow impact, so powerful technology, but only a small number of domestic companies in use. Ali cloud launched a new WebP picture solution to help more enterprises, able to use WebP format, enjoy the convenience brought by WebP format.

What did Ali Cloud do for WebP coding? The software part of Ali Cloud WebP picture solution is provided by CTAccel, and then integrated with ali Cloud’s own FaaS (FPGA as a Service) elastic computing platform, forming a complete Ali cloud WebP picture solution. Based on the reliability, security and convenience of ECS products, Ali Cloud WebP picture solution further gives full play to the ability of FPGA in parallel data processing to meet the computing needs of users and optimize the operation efficiency of programs.

Strapdown computing technology has been committed to the research and development of FPGA image processing and analysis accelerated computing technology, the core technology has obtained the United States patent. The image processing capability provided by the core product redefines the image processing computing mode of data center and provides the most efficient solution for Internet image computing.

Users can use JPEG decoding and WebP coding system on this platform, and can realize batch conversion of JPEG images into WebP images of different resolutions. Compared with the traditional CPU processing scheme, WebP image processing solution based on FPGA acceleration provides several times of performance improvement for image processing.

Control group 1 The control experiment was based on FaaS F1 ECS.F1-C28F1.7Xlarge instance (28 CPU Cores, 1 Intel A10 FPGA Card)

Test picture group parameters: 10000 pictures, average size: 830K, resolution: 4096×2160, total 8.3GB

The test results

Interpretation of test results

We can see that the FPGA + CPU group (blue) has a greater performance improvement than the pure 28-core CPU group at all resolutions due to the involvement of FPGA computing acceleration. The acceleration ratio can reach 3.24 times on average! Also, we can see that the larger the image size, the greater the acceleration effect. For users, the ability to process more images in less time brings cost reductions and performance improvements.

Control group 2 The control experiment was based on FaaS F1 ECS.F1-C28F1.7Xlarge instance (28 CPU Cores, 1pcs Intel A10 FPGA Card)

Test picture group parameters: 10000 pictures, average size: 130K, resolution: 1024×768, total 1.3GB

The test results

Interpretation of test results

When the input image is switched to 1K, it can be seen that, limited by the size of the input image, the acceleration effect of FPGA decreases slightly compared with the control group 1, but the processing speed of FPGA + 28-core CPU group is still 2.54 times higher than that of the simple 28-core CPU group.

Control group 3 The control experiment was based on FaaS F1 ECS.F1-C8F1.2Xlarge instance (8 CPU Cores, 1pcs Intel A10 FPGA Card)

Test picture group parameters: 10000 pictures, average size: 830K, resolution: 4096×2160, total 8.3GB

The test results

Interpretation of test results

In this group evaluation, the acceleration effect of FPGA is greatly amplified compared with the results of the first two groups of tests due to the decrease of the number of cpus in the instance. FPGA + 8-core CPU group compared with the pure 8-core CPU group improved 6.5 times!

Control group 4 The control experiment was based on FaaS F1 ECS.F1-C8F1.2Xlarge instance (8 CPU Cores, 1pcs Intel A10 FPGA Card)

Test picture group parameters: 10000 pictures, average size: 130K, resolution: 1024×768, total 1.3GB

The test results

Interpretation of test results

In this group evaluation, although the number of CPU is reduced and the acceleration performance of FPGA is compared and magnified, the performance is only improved by 3.75 times due to the size of the input image.

Comprehensive evaluation Results The following conclusions can be drawn from the above four evaluation results:

Under the same conditions, the performance of FPGA will be greatly improved, the average improvement is more than 2.5 times, and the optimal performance is 6.5 times. The larger the picture, the better the FPGA acceleration. This fits our scenario, the bigger the image, the stronger our need to compress. With less CPU, the relative acceleration effect of FPGA is more obvious. Under the condition of ensuring the processing speed, we can choose to use FPGA machine processing with less CPU to achieve higher cost performance. In the past, FPGA is only the Internet players can play, play affordable, but Ali Cloud hopes to let more enterprises and individuals enjoy the benefits of improving computing power.

Ali Cloud encapsulates FPGA into cloud products, and on the original basis of productization, modularization, mirroring, all-round optimization of product delivery process, for any company that wants to use FPGA or can provide FPGA technical guidance and development services for other companies, can find themselves in the whole process. FPGA cloud delivery, so that more enterprises can never use the cool!

Even if no one in your organization is good at FPGas, you can still go to the cloud without coding and with a unified delivery process by purchasing development services from the cloud marketplace.

For the incalculable value, the power of PRatt & Whitney FPGA, let WebP rise from here!