Search for: All records

Data centers require high-performance and efficient networking for fast and reliable communication between applications. TCP/IP-based networking still plays a dominant role in data center networking to support a wide range of Layer-4 and Layer-7 applications, such as middleboxes and cloud-based microservices. However, traditional kernel-based TCP/IP stacks face performance challenges due to overheads such as context switching, interrupts, and copying. We present Z-stack, a high-performance userspace TCP/IP stack with a zero-copy design. Utilizing DPDK's Poll Mode Driver, Z-stack bypasses the kernel and moves packets between the NIC and the protocol stack in userspace, eliminating the overhead associated with kernel-based processing. Z-stack em-ploys polling-based packet processing that improves performance under high loads, and eliminates receive livelocks compared to interrupt-driven packet processing. With its zero-copy socket design, Z-stack eliminates copies when moving data between the user application and the protocol stack, which further minimizes latency and improves throughput. In addition, Z-stack seamlessly integrates with shared memory processing within the node, eliminating duplicate protocol processing and serializationldese-rialization overheads for intra-node communication. Z-stack uses F-stack as the starting point which integrates the proven TCP/IP stack from FreeBSD, providing a versatile solution for a variety of cloud use cases and improving performance of data center networking. 

Federated Learning (FL) typically involves a large-scale, distributed system with individual user devices/servers training models locally and then aggregating their model updates on a trusted central server. Existing systems for FL often use an always-on server for model aggregation, which can be inefficient in terms of resource utilization. They also may be inelastic in their resource management. This is particularly exacerbated when aggregating model updates at scale in a highly dynamic environment with varying numbers of heterogeneous user devices/servers. We present LIFL, a lightweight and elastic serverless cloud platform with fine-grained resource management for efficient FL aggregation at scale. LIFL is enhanced by a streamlined, event-driven serverless design that eliminates the individual, heavyweight message broker and replaces inefficient container-based sidecars with lightweight eBPF-based proxies. We leverage shared memory processing to achieve high-performance communication for hierarchical aggregation, which is commonly adopted to speed up FL aggregation at scale. We further introduce the locality-aware placement in LIFL to maximize the benefits of shared memory processing. LIFL precisely scales and carefully reuses the resources for hierarchical aggregation to achieve the highest degree of parallelism, while minimizing aggregation time and resource consumption. Our preliminary experimental results show that LIFL achieves significant improvement in resource efficiency and aggregation speed for supporting FL at scale, compared to existing serverful and serverless FL systems. 

Federated Learning (FL) typically involves a large-scale, distributed system with individual user devices/servers training models locally and then aggregating their model updates on a trusted central server. Existing systems for FL often use an always-on server for model aggregation, which can be inefficient in terms of resource utilization. They also may be inelastic in their resource management. This is particularly exacerbated when aggregating model updates at scale in a highly dynamic environment with varying numbers of heterogeneous user devices/servers. We present LIFL, a lightweight and elastic serverless cloud platform with fine-grained resource management for efficient FL aggregation at scale. LIFL is enhanced by a streamlined, event-driven serverless design that eliminates the individual, heavyweight message broker and replaces inefficient container-based sidecars with lightweight eBPF-based proxies. We leverage shared memory processing to achieve high-performance communication for hierarchical aggregation, which is commonly adopted to speed up FL aggregation at scale. We further introduce the locality-aware placement in LIFL to maximize the benefits of shared memory processing. LIFL precisely scales and carefully reuses the resources for hierarchical aggregation to achieve the highest degree of parallelism, while minimizing aggregation time and resource consumption. Our preliminary experimental results show that LIFL achieves significant improvement in resource efficiency and aggregation speed for supporting FL at scale, compared to existing serverful and serverless FL systems. 

Serverless computing promises an efficient, low-cost compute capability in cloud environments. However, existing solutions, epitomized by open-source platforms such as Knative, include heavyweight components that undermine this goal of serverless computing. Additionally, such serverless platforms lack dataplane optimizations to achieve efficient, high-performance function chains that facilitate the popular microservices development paradigm. Their use of unnecessarily complex and duplicate capabilities for building function chains severely degrades performance. ‘Cold-start’ latency is another deterrent. We describe SPRIGHT, a lightweight, high-performance, responsive serverless framework. SPRIGHT exploits shared memory processing and dramatically improves the scalability of the dataplane by avoiding unnecessary protocol processing and serialization-deserialization overheads. SPRIGHT extensively leverages event-driven processing with the extended Berkeley Packet Filter (eBPF). We creatively use eBPF’s socket message mechanism to support shared memory processing, with overheads being strictly load-proportional. Compared to constantly-running, polling-based DPDK, SPRIGHT achieves the same dataplane performance with 10× less CPU usage under realistic workloads. Additionally, eBPF benefits SPRIGHT, by replacing heavyweight serverless components, allowing us to keep functions ‘warm’ with negligible penalty. Our preliminary experimental results show that SPRIGHT achieves an order of magnitude improvement in throughput and latency compared to Knative, while substantially reducing CPU usage, and obviates the need for ‘cold-start’. 

While 5G offers fast access networks and a high-performance data plane, the control plane in 5G core (5GC) still presents challenges due to inefficiencies in handling control plane operations (including session establishment, handovers and idle-to-active state-transitions) of 5G User Equipment (UE). The Service-based Interface (SBI) used for communication between 5G control plane functions introduces substantial overheads that impact latency. Typical 5GCs are supported in the cloud on containers, to support the disaggregated Control and User Plane Separation (CUPS) framework of 3GPP. L25GC is a state-of-the-art 5G control plane design utilizing shared memory processing to reduce the control plane latency. However, L25GC has limitations in supporting multiple user sessions and has programming language incompatibilities with 5GC implementations, e.g., free5GC, using modern languages such as GoLang. To address these challenges, we develop L25GC+, a significant enhancement to L25GC. L25GC+ re-designs the shared-memory-based networking stack to support synchronous I/O between control plane functions. L25GC+ distinguishes different user sessions and maintains strict 3GPP compliance. L25GC+ also offers seamless integration with existing 5GC microservice implementations through equivalent SBI APIs, reducing code refactoring and porting efforts. By leveraging shared memory I/O and overcoming L25GC’s limitations, L25GC+ provides an improved solution to optimize the 5G control plane, enhancing latency, scalability, and overall user experience. We demonstrate the improved performance of L25GC+ on a 5G testbed with commercial basestations and multiple UEs. 

Cloud-native microservice applications use different communication paradigms to network microservices, including both synchronous and asynchronous I/O for exchanging data. Existing solutions depend on kernel-based networking, incurring significant overheads. The interdependence between microservices for these applications involves considerable communication, including contention between multiple concurrent flows or user sessions. In this paper, we design X-IO, a high-performance unified I/O interface that is built on top of shared memory processing with lock-free producer/consumer rings, eliminating kernel networking overheads and contention. X-IO offers a feature-rich interface. X-IO’s zero-copy interface supports building provides truly zero-copy data transfers between microservices, achieving high performance. X-IO also provides a POSIX-like socket interface using HTTP/REST API to achieve seamless porting of microservices to X-IO, without any change to the application code. X-IO supports concurrent connections for microservices that require distinct user sessions operating in parallel. Our preliminary experimental results show that X-IO’s zero-copy interfaces achieve 2.8x-4.1x performance improvement compared to kernel-based interfaces. Its socket interfaces outperform kernel TCP sockets and achieve performance close to UNIX-domain sockets. The HTTP/REST APIs in X-IO perform 1.4 x-2.3 x better than kernel-based alternatives with concurrent connections. 

With the commercialization and deployment of 5G, efforts are beginning to explore the design of the next generation of cellular networks, called 6G. New and constantly evolving use cases continue to place performance demands, especially for low latency communications, as these are still challenges for the 3GPP-specified 5G design, and will have to be met by the 6G design. Therefore, it is helpful to re-examine several aspects of the current cellular network’s design and implementation.Based on our understanding of the 5G cellular network specifications, we explore different implementation options for a dis-aggregated 5G core and their performance implications. To improve the data plane performance, we consider advanced packet classification mechanisms to support fast packet processing in the User Plane Function (UPF), to improve the poor performance and scalability of the current design based on linked lists. Importantly, we implement the UPF function on a SmartNIC for forwarding and tunneling. The SmartNIC provides the fastpath for device traffic, while more complex functions of buffering and processing flows that suffer a miss on the SmartNIC P4 tables are processed by the host-based UPF. Compared to an efficient DPDK-based host UPF, the SmartNIC UPF increases the throughput for 64 Byte packets by almost 2×. Furthermore, we lower the packet forwarding latency by 3.75× by using the SmartNIC. In addition, we propose a novel context-level QoS mechanism that dynamically updates the Packet Detection Rule priority and resource allocation of a flow based on the user context. By combining our innovations, we can achieve low latency and high throughput that will help us evolve to the next generation 6G cellular networks. 

Traditional network resident functions (e.g., firewalls, network address translation) and middleboxes (caches, load balancers) have moved from purpose-built appliances to software-based components. However, L2/L3 network functions (NFs) are being implemented on Network Function Virtualization (NFV) platforms that extensively exploit kernel-bypass technology. They often use DPDK for zero-copy delivery and high performance. On the other hand, L4/L7 middleboxes, which usually require full network protocol stack support, take advantage of a full-fledged kernel-based system with a greater emphasis on functionality. Thus, L2/L3 NFs and middleboxes continue to be handled by distinct platforms on different nodes.This paper proposes MiddleNet that seeks to overcome this dichotomy by developing a unified network resident function framework that supports L2/L3 NFs and L4/L7 middleboxes. MiddleNet supports function chains that are essential in both NFV and middlebox environments. MiddleNet uses DPDK for zero-copy packet delivery without interrupt-based processing, to enable the ‘bump-in-the-wire’ L2/L3 processing performance required of NFV. To support L4/L7 middlebox functionality, MiddleNet utilizes a consolidated, kernel-based protocol stack processing, avoiding a dedicated protocol stack for each function. MiddleNet fully exploits the event-driven capabilities provided by the extended Berkeley Packet Filter (eBPF) and seamlessly integrates it with shared memory for high-performance communication in L4/L7 middlebox function chains. The overheads for MiddleNet are strictly load-proportional, without needing the dedicated CPU cores of DPDK-based approaches. MiddleNet supports flow-dependent packet processing by leveraging Single Root I/O Virtualization (SR-IOV) to dynamically select packet processing needed (Layer 2 to Layer 7). Our experimental results show that MiddleNet can achieve high performance in such a unified environment.