Average Flow Completion Time Research Articles

A distributed load balancing architecture based on in‐band network telemetry

AbstractA data center (DC) is supposed to efficiently distribute the bandwidth of the network to provide high‐quality traffic transmission. However, the load imbalance issue can easily occur due to the complex topology and traffic features. Equal‐Cost Multi‐Path(ECMP) distributes traffic on different paths but doesn't consider network congestion. Although HULA solved some of ECMP's problems, it can easily congest the best path. RPS randomly distributes packets across multiple paths, potentially causing packet reordering in certain scenarios. This paper presents DHLB, a distributed hop‐by‐hop load balancing architecture based on in‐band network telemetry. With active In‐band network telemetry, DHLB gathers essential load information and systematically records it in a load information table. DHLB distributes traffic proportionally on different paths based on their load degree. We build a fat tree topology on mininet to verify the performance of our design. Experimental results indicate that DHLB outperforms other schemes regarding average Flow Completion Time (FCT). It also performs better on additional overhead than another probe‐based scheme.

Towards fine-grained load balancing with dynamical flowlet timeout in datacenter networks

In modern datacenter networks (DCNs), load balancing mechanisms are widely deployed to enhance link utilization and alleviate congestion. Recently, a large number of load balancing algorithms have been proposed to spread traffic among the multiple parallel paths. The existing solutions make rerouting decisions for all flows once they experience congestion on a path. They are unable to distinguish between the flows that really need to be rerouted and the flows that potentially have negative effects due to rerouting, resulting in frequently ineffective rerouting. Fine-grained rerouting will also cause severe packet reordering, especially in asymmetric topology scenarios. To address the above issues, we present a fine-grained traffic-differentiated load balancing (TDLB) mechanism, which aims to distinguish flows that are necessarily to be rerouted and reroute traffic in fine-grained without packet reodering. Specifically, TDLB distinguishes the traffic that must be rerouted through the host pair information in the packet header, and selects an optimal path for rerouting. To prevent severe packet reodering caused by excessive path delay differences, TDLB dynamically adjusts the flowlet timeout to segment the traffic and select the optimal path for rerouting. The NS-2 simulation results show that TDLB effectively reduces tail latency and average flow completion time (FCT) for short flows by up to 49% and 46%, respectively, compared to the state-of-the-art load balancing schemes.

A receiver-driven transport protocol using differentiated algorithms for differential congestion in datacenters

With the continuous development of applications, data centers run application traffic from various service providers, and the requirements for data center networks are also increasing. Congestion control has always been a hot topic of discussion. Prevalent sender-driven transport protocols require adjusting their sending rate after network congestion occurs, and the long feedback delay makes them susceptible to a buffer overflow. Therefore, receiver-driven transport protocols are proposed, and the receiver adopts credits to allocate downlink bandwidth to overcome network congestion. This type of scheme solves last-hop congestion and reduces congestion feedback latency. However, the existing receiver-driven transport protocol is insensitive to congestion location and thus reacts inaccurately to congestion. To address this problem, this paper proposes a congestion control algorithm with congestion location awareness. In this scheme, 1) we first design an ECN-based congestion detection method to achieve lightweight congestion types awareness, i.e., instantaneous congestion and continuous congestion. 2) Based on this, we further design a differentiated transmission control strategy. For instantaneous congestion, we adopt adaptive backoff and delay control algorithms. For continuous congestion, we employ a rate control algorithm to reduce in-network congestion. This does not require complex modifications to the switch. In our evaluation, the overall average flow completion time (FCT) of DCC is up to 21%, 89%, 1.5×, and 3.4× better than Homa, ExpressPass, Timely, and pHost, respectively.

Fine-grained load balancing with proactive prediction and adaptive rerouting in data center

Though the existing load balancing designs successfully make full use of available parallel paths and attain high bisection network bandwidth, they reroute flows regardless of their dissimilar performance requirements. But traffic in modern data center networks exhibits short bursts characteristic, which can easily lead to network congestion. The short flows suffer from the problems of large queuing delay and packet reordering, while the long flows fail to obtain high throughput due to low link utilization and packet reordering. In order to solve these inefficiency, we designed a fine-grained load balancing method (FLB), which uses an active monitoring mechanism to split traffic, and flexibly transfers flowlet to non-congested path, effectively reducing the negative impact of burst flow on network performance. Besides, to avoid packet reordering, FLB leverages the probe packets to estimate the end-to-end delay, thus excluding paths that potentially cause packet reordering. The test results of NS2 simulation show that FLB significantly reduces the average and tail flow completion time of flows by up to 59% and 56% compared to the state-of-the-art multi-path transmission scheme with less computational overhead, as well as increases the throughput of long flow.

PDLB: Path Diversity-aware Load Balancing with adaptive granularity in data center networks

Modern data center topologies often take the form of a multi-rooted tree with rich parallel paths to provide high bandwidth. However, various path diversities caused by traffic dynamics, link failures and heterogeneous switching equipment widely exist in the production datacenter network. Therefore, the multi-path load balancer in data center should be robust to these diversities. Although prior fine-grained schemes such as RPS and Presto make the best use of available paths, However, they are prone to experiencing packet reordering problem under the asymmetric topology. The coarse-grained solutions such as ECMP and LetFlow effectively avoid packet reordering, but easily lead to under-utilization of multiple paths. To cope with these inefficiencies, we propose a load balancing mechanism called PDLB, which adaptively adjusts flowcell granularity according to path diversity. PDLB increases flowcell granularity to alleviate packet reordering under large degrees of topology asymmetry, while reducing flowcell granularity to obtain high link utilization under small degrees of topology asymmetry. PDLB is only deployed on the sender without any modification on switch. We evaluate PDLB through large-scale NS2 simulations. The experimental results show that PDLB reduces the average flow completion time by up to ∼\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\sim$$\\end{document}8-53% over the state-of-the-art load balancing schemes.

A reinforcement learning method for scheduling service function chains with multi-resource constraints

Traditional networks are usually equipped with many dedicated middleboxes to provide various network services. Though these hardware-based devices certainly improve network performance, they are usually expensive and difficult to upgrade. To overcome this shortcoming, network function virtualization (NFV), which accomplishes network services in the form of virtual network functions (VNF) has been presented. Compared to middleboxes, the VNFs are easy to deploy and migrate. Usually, multiple VNFs are chained in a specified order as a service function chain (SFC) to serve a given flow. There are many works to schedule SFCs to minimize the average flow completion time. However, they only consider single resource limitation. In this paper, we are committed to addressing the problem of multi-resource SFC scheduling (MR-SFCS) and minimizing the average flow completion time. We formulate this problem with an Integer Linear Programming (ILP) model and prove its NP-hardness. To well tackle this problem, we propose an approach based on deep reinforcement learning (DRL), which has specific reward design and state representations. Besides, we extend the offline approach to online SFC scheduling. The experiment results demonstrate that our DRL method can significantly reduce the average flow completion time and achieves a cost saving of 69.07% against the benchmark method.

DiffTREAT: Differentiated Traffic Scheduling Based on RNN in Data Centers

Transmission schemes in data centers are supposed to accurately distinguish flow types for different scheduling. However, prior efforts failed to meet the needs at all levels in a cost-effectively way. Nor the existing schemes proved applicable to all the diverse scenarios or dynamic traffic patterns. Therefore, we proposed <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Diff</u> erentiated <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Tr</u> affic sch <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">E</u> duling in d <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A</u> ta cen <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">T</u> ers (DiffTREAT) based the Recurrent Neural Network (RNN), aiming to simplify the transmission in the dynamic and diverse network scenarios. First, DiffTREAT utilizes deep learning methods for traffic classification and flow size prediction. Second, according to the classified results of flows, DiffTREAT adopts multilevel priority queues to ensure the preferential transmission of latency-sensitive flows while optimizing the overall average flow completion time (FCT). Third, DiffTREAT employs the network cache to increase the capacity of data center networks (DCN), which effectively fights against the traffic burst and improves the throughput of latency-insensitive flows. DiffTREAT has been tested in different topologies in the contexts of diverse network loads and real-world workloads. Experiment results showed that compared with state-of-the-art schemes, DiffTREAT yielded both the lower average flow completion time for latency-sensitive flows and the higher throughput for latency-insensitive flow.

Load balancing for heterogeneous traffic in datacenter networks

In modern datacenter networks (DCNs), the overwhelming heterogeneous flows have various stringent demands, ranging from delay-sensitive short flows, throughput-sensitive long flows to best-effort flows without deadline. Recently, many load balancing schemes are proposed to deliver good performance for datacenter applications. However, the existing solutions cannot meet all the above requirements simultaneously. Especially, the short flows experience head-of-line blocking due to queued behind the long and best-effort flows. The long flows often suffer from throughput degradation due to bursty congestion. To solve these issues, we present LBT, a traffic-differentiated load balancer. The key design point of LBT is to adaptively adjust the switching granularity of long flows and best-effort flows according to the two calculated switching thresholds based on the traffic strength. Specifically, under heavy load, the switching granularity is increased to guarantee required bandwidth capacity for delay-sensitive short flows to finish quickly. In contrary, the switching granularity is reduced to enable the long flows to make full of parallel equal-cost paths. Moreover, we adopt different routing strategies for the three different categories flows. We conduct NS-2 simulations to evaluate the effectiveness of LBT. The experimental results show that LBT significantly reduces the average flow completion time of short flows by up to 55.9% ∼ 65.4% compared to the state-of-the-art solutions and achieves high throughput for long flows concurrently.

A Receiver-Driven Transport Protocol With High Link Utilization Using Anti-ECN Marking in Data Center Networks

Existing reactive or proactive congestion control protocols are hard to simultaneously achieve ultra-low latency and high link utilization across all workloads ranging from delay-sensitive flows to bandwidth-hungry ones in datacenter networks. We present an Anti-ECN (Explicit Congestion Notification) Marking Receiver-driven Transport protocol called AMRT, which achieves both near-zero queueing delay and full link utilization by reasonably increasing sending rate in the case of under-utilization. Specifically, switches mark the ECN bit of data packets once detecting spare bandwidth. When receiving the anti-ECN marked packet, the receiver generates the corresponding marked grant to trigger more data packets. The testbed and simulation experiments show that AMRT effectively reduces the average flow completion time (AFCT) by up to 42% and improves the link utilization by up to 38% over the state-of-the-art receiver-driven transmission schemes.

RaceCC: A rapidly converging explicit congestion control for datacenter networks

Congestion control (CC) in datacenter networks has three primary goals: high link utilization, low queuing delay, and rapid convergence to fairness. Most of the host-driven CC schemes perform well in achieving the first two goals, while it is difficult to converge to fairness rapidly. Switch-driven CC schemes can be a compelling alternative to achieve fairness since switches explicitly provide feedback to hosts. However, we found that existing switch-driven CC schemes converge slowly or sometimes could not guarantee low queues.Based on these observations, in this paper, we propose RaceCC, a RApidly Converging Explicit CC to achieve the three primary goals simultaneously. As a switch-driven CC scheme, RaceCC enables flows to reach a fair rate after the first RTT and guarantees high link utilization and low queue length. To converge rapidly, RaceCC adjusts flow rates through an intuitive MIMD method with additive-decrease for short queues and precise update for increase. Meanwhile, RaceCC can be implemented with several simple operations. We theoretically analyze the stability of RaceCC and evaluate its performance through micro-benchmark and large-scale simulations. The results show that RaceCC reduces the overall average and tail flow completion time (FCT) by 20% ∼ 57% and 15% ∼ 63%, compared to DCQCN, TIMELY, HPCC, PowerTCP, RCP and RoCC.

A[formula omitted]DCT: A cubic acceleration TCP for data center networks

In recent years, online data-intensive (OLDI) applications have become particularly common in data center networks (DCNs), such as web search, advertisement systems, and distributed machine learning. OLDI applications have strict latency requirements for short flows and must operate under soft-real-time constraints (e.g., 300 ms latency). On the other hand, there exist long flows that are unaware of flow latency but are sensitive to throughput. Thus, a special transmission protocol that can satisfy different demands for both long flows and short flows is immediately needed. L2DCT is one of the most representative differentiated transmission protocols, which aim at the reduction in completion time for short flows. However, the performance of L2DCT becomes less significant in the scenario where short flows account for the majority (e.g., data mining). For the purposes of minimizing the flow completion time (FCT) of short flows, achieving the minimum average flow completion time (AFCT), and guaranteeing the deadline constraints upon flow transmission times, we propose a cubic acceleration TCP for DCNS, which is referred to as A3DCT.A3DCT employs the Shortest Remaining Process Time (SRPT) scheduling policy to adjust its congestion window according to the remaining bytes of the flow. Moreover, to improve the priority of short flows without affecting the throughput of long flows, A3DCT leverages a cubic function to indicate the urgency of short flows. Finally, A3DCT accelerates the recovery speed, in a flexible manner, depending on the urgency of different flows in a non-congestion state to request the bandwidth to finish transmission. We perform simulations of different scales by using NS-3 simulator to evaluate the performance of A3DCT. The evaluation results justify that A3DCT can not only achieve a low flow completion time, especially for short flows, but also guarantee the high throughput of long flows.

A Multipath Transmission System for Information-Centric Networking Based on Standalone Name Resolution

Information-Centric Networking (ICN) essentially supports multipath transmission. However, current multipath schemes in ICN either necessitate major network infrastructure updates or necessitate specific network settings for terminal devices. To solve these problems, we propose MPTS-ICN, a multipath transmission system for ICN that realizes end-to-end multipath transmission. Taking the ICN architecture based on the standalone name resolution approach as a basis, MPTS-ICN is easier to implement and deploy than other ICN multipath schemes. Moreover, we have extended the original network layer protocol to support multipath data transmission in ICN. To set up concurrent transmission multipath efficiently, we propose a heuristic algorithm for the selection of multipath service nodes. Extensive experimental comparisons with existing data transmission methods show that in bandwidth-constrained scenarios, MPTS-ICN outperforms the best-route method by 83.6% and the ECMP method by 79.7% in average flow completion time.

Load balancing inside programmable data planes based on network modeling prediction using a GNN with network behaviors

In data center networks, existing control plane- and end host-based load-balancing methods are encumbered by excessively large decision delays during rapid reactions to microbursts. However, existing programmable data plane-based load balancing methods require large overheads involved in active probing. Accurate network modeling can optimize load balancing. However, existing modeling methods suffer from low generalization and high overhead. In this study, we propose a network modeling method based on a graph neural network (GNN) with basic network behavior (hereinafter called GNN-Behavior). This is derived from two inherent correlations observed: the correlation between global network behavior and basic network behavior and the correlation among basic network behaviors. We employed the GNN with an improved message-passing neural network to learn such two inherent correlations. Particularly, we considered modeling end-to-end delay as a use case to validate GNN-Behavior. Furthermore, we propose a packet-level load-balancing scheme inside programmable data planes (PDPs) based on the accurate prediction of end-to-end delay from the GNN-Behavior model(LBPP). LBPP is a control plane-PDP collaborative method that integrates a global view from a controller and quick response from switches. Experimental results demonstrate the feasibility and effectiveness of the GNN-Behavior and LBPP. Compared with queuing theory (QT), RouteNet, and GNN-based scheme, GNN-Behavior increases goodness of fit (R2) by 73.1%, 11.1%, and 3.74%, respectively. Under an unknown traffic control strategy, the generalization ability of GNN-Behavior is considerably better than that of QT and RouteNet. Compared with flow-level ECMP, flowlet-level LetFlow, and packet-level DRILL, LBPP can reduce average flow completion time by up to 43.9%, 37.4%, and 17.2%, respectively.

Packet rank-aware active queue management for programmable flow scheduling

Flow scheduling is crucial in improving network Quality of Service (QoS). According to different optimization goals, many scheduling algorithms (e.g., STFQ, pFabric, SRPT, WFQ) have emerged in network. However, these scheduling algorithms are bound with hardware, which are only applicable to fixed network scenarios. Push-In First-Out (PIFO) queues are proposed as a queue abstraction to support programmable packet scheduling, but implementing them in hardware is difficult. We present a general framework named PR-AQM, which uses limited number of FIFO queues to approximate PIFO behavior. The core idea of PR-AQM is to perform Active Queue Management (AQM) on programmable data plane by capturing packet rank distribution and queue congestion level in real time. Specifically, PR-AQM can dynamically adjust the mapping between packet ranks and queues according to the delay status of different priority queues, thus efficiently achieving priority scheduling of packets. We carry out extensive simulations and implement PR-AQM prototype in Tofino switches to evaluate the performance of PR-AQM. Experimental results show that compared with state-of-the-art algorithms, PR-AQM can adapt well to different scheduling algorithms, and achieve better performance by reducing the average flow completion time for short flows up to 45%.

A Novel Multipath Transmission Scheme for Information-Centric Networking

Due to the overload of IP semantics, the traditional TCP/IP network has a number of problems in scalability, mobility, and security. In this context, information-centric networking (ICN) is proposed to solve these problems. To reduce the cost of deployment and smoothly evolve, the ICN architecture needs to be compatible with existing IP infrastructure. However, the rigid underlying IP routing regulation limits the data transmission efficiency of ICN. In this paper, we propose a novel multipath transmission scheme by utilizing the characteristics and functions of ICN to enhance data transmission. The process of multipath transmission can be regarded as a service, and a multipath transmission service ID (MPSID) is assigned. By using the ICN routers bound to the MPSID as relay nodes, multiple parallel paths between the data source and the receiver are constructed. Moreover, we design a path management mechanism, including path selection and path switching. It can determine the initial path based on historical transmission information and switch to other optimal paths according to the congestion degree during transmission. The experimental results show that our proposed method can improve the average throughput and reduce the average flow completion time and the average chunk completion time.

RCC: Enabling Receiver-Driven RDMA Congestion Control With Congestion Divide-and-Conquer in Datacenter Networks

The development of datacenter applications leads to the need for end-to-end communication with microsecond latency. As a result, RDMA is becoming prevalent in datacenter networks to mitigate the latency caused by the slow processing speed of the traditional software network stack. However, existing RDMA congestion control mechanisms are either far from optimal in simultaneously achieving high throughput and low latency or in need of additional in-network function support. In this paper, by leveraging the observation that most congestion occurs at the last hop in datacenter networks, we propose RCC, a receiver-driven rapid congestion control mechanism for RDMA networks that combines explicit assignment and iterative window adjustment. Firstly, we propose a network congestion distinguish method to classify congestions into two types, last-hop congestion and in-network congestion. Then, an Explicit Window Assignment mechanism is proposed to solve the last-hop congestion, which enables senders to converge to a proper sending rate in one-RTT. For in-network congestion, a PID-based iterative delay-based window adjustment scheme is proposed to achieve fast convergence and near-zero queuing latency. RCC does not need additional in-network support and is friendly to hardware implementation. In our evaluation, the overall average FCT (Flow Completion Time) of RCC is <inline-formula> <tex-math notation="LaTeX">$4{\sim}79\%$</tex-math> </inline-formula> better than Homa, ExpressPass, DCQCN, TIMELY, and HPCC.

Swing: Providing Long-Range Lossless RDMA via PFC-Relay

Remote Direct Memory Access (RDMA) has been widely deployed in datacenters for its high performance. Large-scale high performance cloud services built on geographically distributed datacenters require long-range RDMA for performance requirements. However, existing RDMA solutions can hardly satisfy the stringent requirements of the emerging large-scale high-performance cloud services built on geo-distributed datacenters in terms of throughput and delay. On the one hand, lossless RDMA suffers from a deep buffer and potential suboptimal throughput for inter-datacenter traffic due to delayed response to Priority Flow Control (PFC) messages. On the other hand, lossy RDMA with selective retransmissions suffers from poor performance when multiple flows with different round-trip times (RTTs) coexist in cross-datacenter scenarios. This article proposes <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Swing</small> , which expands the high-performance lossless RDMA to long-distance links through PFC-Relay. <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Swing</small> ensures the throughput of long-distance links while minimizing the buffer requirement for long-range RDMA. It enables long-range RDMA without making any modifications to existing in-datacenter networks. The evaluation shows that <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Swing</small> can reduce the average flow completion time (FCT) by 14%-66% in a variety of traffic scenarios.

An Adaptable and Agnostic Flow Scheduling Approach for Data Center Networks

Cloud applications have reshaped the model of services and infrastructure of the Internet. Search engines, social networks, content delivery and retail and e-commerce sites belong to this group of applications. An important element in the architecture of data centers where these applications run is the communication infrastructure, commonly known as data center networks (DCNs). A critical challenge DCNs have to address is the processing of the traffic of cloud applications, which due to its properties is essentially different to the traffic of other Internet applications. In order to improve the responsiveness and throughput of applications, DCNs should be able to prioritize short flows (a few kB) over long flows (several MB). However, given the time and space variations the traffic presents, the information about flow sizes is not available in advance in order to plan the flow scheduling. There has been a wealth of solutions developed in this space, and prior work includes flow scheduling mechanisms optimizing for a specific workload but fall short when workloads are not known in advance, or comprise a collection of applications changing dynamically. In this paper, we present an adaptable mechanism called Adaptable Workload-Agnostic Flow Scheduling (AWAFS). It is an adaptable approach that can adjust in an agnostic way the scheduling configuration of DCN forwarding devices. This agnostic adjustment contributes to reduce the Flow Completion Time (FCT) of those short flows, representing around 85% of the traffic handled by cloud applications. AWAFS operates by observing the traffic and detecting statistical properties that provide a hint to adapt the scheduling parameters. Our evaluation results based on simulation show that AWAFS can reduce the average FCT of short flows between 16.9% and 45.2% when compared to the best existing agnostic non-adaptable solution, without inducing starvation on long flows. Indeed, it can provide improvements as high as 39% for long flows. Additionally, AWAFS can improve the FCT for short flows in scenarios with high heterogeneity in the traffic present in the network, with a reduction up to 5% for the average FCT and 15% for the tail FCT.

Achieving Per-Flow Fairness and High Utilization With Limited Priority Queues in Data Center

Modern data centers often host multiple applications with diverse network demands. To provide fair bandwidth allocation to several thousand traversing flows, Approximate Fair Queueing (AFQ) utilizes multiple priority queues in switch to approximate ideal fair queueing. However, due to limited number of queues in programmable switches, AFQ easily experiences high packet loss and low link utilization. In this paper, we propose Elastic Fair Queueing (EFQ), which leverages limited priority queues to flexibly achieve both high network utilization and fair bandwidth allocation. EFQ dynamically assigns the free buffer space in priority queues for each packet to obtain high utilization without sacrificing flow-level fairness. The results of simulation experiments and real implementations show that EFQ reduces the average flow completion time by up to 82% over the state-of-the-art fair bandwidth allocation mechanisms.

FINT: Flexible In-band Network Telemetry method for data center network

By embedding the state data maintained by the programmable data plane into user packets, the In-band Network Telemetry (INT) can greatly improve the freshness of measured data, such as capturing the micro-bursts in data center networks. However, because the existing INT schemes use fixed telemetry tasks and metadata adding strategies, INT cannot be adjusted in time according to the network state, thus lacking flexibility. To solve the above problems, we propose a Flexible In-band Network Telemetry method FINT. With the designed triple bitmap mechanism, telemetry tasks and parameters can be set dynamically at runtime, which providing flexibility without redeployment to improve telemetry efficiency and reduce the impact of telemetry on network performance. With the designed flexible INT field structure, the types of telemetry metadata carried by user packets can be flexibly selected without significantly adding the indication field. Based on this, to meet the limitation of packet length and reduces the impact on the flow completion time, a greedy telemetry metadata selection algorithm MSG is designed to preferentially select the telemetry metadata with maximum bit-width among the remaining telemetry metadata. The experimental results show that FINT is flexible in adjusting telemetry tasks and add telemetry metadata, and effectively reduce the average flow completion time of the mice flow carrying INT data in the network without significantly increasing the bandwidth consumption.