大普微(DapuStor)
eTLC SSDs in Baidu Cloud Hosting Background

As BCH's demand grows, Baidu AI Cloud seeks better storage solutions for its cloud hosts. Traditional cloud host solutions typically involve: 1) deploying various NVMe SSDs directly for stable performance but at high costs and presenting complex capacity scaling; or 2) creating virtual disks on NVMe devices through a hypervisor, which offers lower performance and higher CPU consumption. To address these issues, DapuStor has proposed an NVMe SSD SR-IOV solution tailored to enhance performance and efficiency.



The Challenge
Ensure high performance while reducing deployment costs.
01
Address deployment challenges caused by varying user capacity requirements.
02
Reduce server CPU consumption.
03
The Results

Directly assigned multiple VFs to virtual machines, achieving performance close to that of physical disks and reducing host CPU load.

Decreased deployment costs of virtualization platforms.

Provided flexible storage capacity options.

The Solution

DapuStor’s SR-IOV solution uses PCIe Gen4 SSD J5100 3.84TB/7.68TB supporting up to 16 VFs and 240GB/480GB namespaces. It enhances virtual machine operations, reduces virtual platform deployment costs, and delivers performance close to physical disks. 


The solution offers two modes: automatic flow control based on capacity or manual custom control, allowing separate read/write flow limits for each VF/NS to ensure QoS. It also requires only a fixed configuration on the server, enabling flexible storage capacity options and improving the user experience.


官网客户案例_画板 1 副本 2.png

The Products