Thursday, June 15th |
Thursday, June 15th 09:00-10:00 AM | |
Thursday, June 15th 09:00-10:00 AM | B-201: Software 2 – Drivers and Managers (Software Track) | | Paper Presenters:
| Writing Universal DPDK Drivers for SmartNICs Jan Zieleźnicki, Sr Software Engineer, CodiLimeDeveloping Secure IPU/DPU Device Management Naru Sundar, Principal Engineer, Intel | Session Description:
| The programmability of SmartNICs is obviously a huge advantage, allowing them to take on a variety of tasks and be updated as needed. However, it does complicate the software. Drivers become more complex, as does device management. Software is flexible, but developing it is expensive. And there always seems to be a need for more utilities unrelated to the actual applications.
| About the Organizer/Moderator: | |
|
Thursday, June 15th 09:00-10:00 AM | C-201: Best Way to Accelerate Networks Today Panel (sponsored by Chelsio) (Panels Track) | | Panel Members:
| | Session Description:
| Network acceleration is a major application for SmartNICs. The major clouds all use them to handle their huge and ever-changing application workloads. Telcos use them to implement virtualization (NFV) and to provide easy upgrades and higher modular modularity and scalability. Enterprise networks use them to avoid overburdening CPUs and to simplify the addition of new or updated services.
| About the Organizer/Moderator: | |
|
Thursday, June 15th 02:00-3:20 PM | |
Thursday, June 15th 02:00-3:20 PM | B-202: Best Architecture for SmartNICs Today Panel (Panels Track) | | Panel Members:
| | Session Description:
| Current SmartNIC architectures offer many tradeoffs. Ones based on ASICs or FPGAs are generally faster and easier to implement. Those based on standard processors are more flexible and easier to update. Coprocessors such as GPUs or DPUs increase performance but are expensive and difficult to program. Most observers expect the software-based devices to dominate the market because of their flexibility, but performance issues could leave a substantial market for hardware-based SmartNICs. The cost and time requirements for developing typical applications for processor-based SmartNICs is as yet unknown, but could have a big effect.
| About the Organizer/Moderator: | |
|
Thursday, June 15th 02:00-3:20 PM | |
Thursday, June 15th 03:30-5:00 PM | A-203: Application Acceleration 2 - AI Applications (Application Acceleration Track) | | Paper Presenters:
| Using SmartNICs to Achieve the Promise of ChatGPT Itay Ozery, Director Product Management, NVIDIA | Session Description:
| AI is clearly a major emerging application for SmartNICs. AI applications always need more computing power, and SmartNICs are an obvious way to provide it. An emerging class of AI applications are the so-called formative systems, such as the incredibly popular ChatGPT. Their basic engine requires tremendous amounts of processing andadd-ins can increase the requirements even further.
| About the Organizer/Moderator: | |
|
Thursday, June 15th 03:30-5:00 PM | B-203: SmartNICs in 2028 and How We Got There Panel (sponsored by X-Scale) (Panels Track) | | Panel Members:
| | Session Description:
| The five-year horizon for SmartNICs is very promising. They will make up an ever increasing part of the NIC market, as their cost is more than balanced by the tremendous capabilities they bring to networked systems. More canned software will be available for them, as well as development platforms, operating systems, utilities, and other tools. Standardization will be a major issue, as most large customers will want multiple sources as well as large ecosystems and wide support for both development and test. Other issues include achieving higher throughput and lower latency, isolating executing applications from one another, security, and operating system support.
| About the Organizer/Moderator: | |
|