Lecture 40: Enhanced Networking
- 课程：[NEW] AWS Certified Advanced Networking Specialty 2026
- 视频时长：12:23
- 来源：Udemy Lecture 26073034
- Slide 数量：9
- 画面切换：00:33、02:40、04:42、06:38、07:19、07:48、08:48、09:27
扫描说明
按 Udemy Transcript 的 274 个时间点逐一定位并比较视频画面，确认 Enhanced Networking 标题、原理、SR-IOV/PCI、前提条件、实例支持、驱动验证以及三种数据路径共 9 个 Slide。同页组件逐步出现视为动画构建，截图保留最终稳定状态。
Transcript 来自 Udemy 页面自带的英文自动字幕，没有使用 ASR。下面保留页面原文，包括其中的语法或识别错误。
1. Enhanced Networking
时间范围：00:00–00:33
总结 / Summary
中文
增强联网（Enhanced Networking）是提升 EC2 带宽、PPS 和延迟表现的重要机制，也是考试重点。它建立在前面介绍的集群置放群组（Cluster Placement Group）和 EBS 优化之外。
English
Enhanced networking is a key mechanism for improving EC2 bandwidth, packets per second, and latency, and it is an important exam topic. It builds on the earlier discussion of cluster placement groups and EBS optimization.
Transcript
00:00 Hello and welcome
00:01 to this lecture, Enhanced Networking.
00:02 Earlier we saw other two options
00:04 by which you can increase the network throughput in AWS.
00:07 The first we saw, cluster replacement group
00:10 for EC2 instances,
00:11 and then we saw EBS-optimized EC2 instances.
00:14 Now, let's look at the more important concept,
00:18 that's Enhanced Networking.
00:19 And in the exam,
00:20 I would expect two to three questions on this topic.
00:23 So let's try to understand what does it mean
00:25 by Enhanced Networking,
00:26 and what are the options you have when we say you
00:30 have to configure enhanced networking in AWS.
2. What Enhanced Networking Does
时间范围：00:33–02:40
总结 / Summary
中文
增强联网（Enhanced Networking）目标是实现超过每秒百万包（Packets Per Second, PPS）的处理能力、降低实例间延迟并提高带宽。核心技术是单根 I/O 虚拟化（Single Root I/O Virtualization, SR-IOV）配合 PCI 直通（PCI Passthrough），尽量绕过虚拟机管理程序（Hypervisor）。EC2 有两种互斥实现：Intel 82599 虚拟功能驱动 ixgbevf 与弹性网络适配器（Elastic Network Adapter, ENA）。
English
Enhanced networking targets more than one million packets per second, lower instance-to-instance latency, and higher bandwidth. Its core is Single Root I/O Virtualization with PCI passthrough, which removes as much hypervisor processing as possible. EC2 offers two mutually exclusive implementations: the Intel 82599 Virtual Function driver ixgbevf and the Elastic Network Adapter.
Transcript
00:33 So first, let's understand what is Enhanced Networking.
00:36 Of course, the term says itself,
00:38 it means you have to enhance the network performance
00:41 between two systems.
00:43 Now, these two systems are typically,
00:45 we can take an example like EC2 instances,
00:47 but it can also be like from AWS to on-premise network,
00:51 from AWS to S3.
00:53 So whenever you deal with any AWS services,
00:56 we are talking about this enhanced networking,
00:58 but typically, EC2 would be involved there
01:00 because most of this enhanced networking
01:02 is applicable for EC2 instances.
01:05 Okay. So when you can process the packets,
01:09 over 1 million packet per second,
01:11 we can consider it as like enhanced networking.
01:13 And for achieving this, you need to do something special,
01:17 something extra and that's what we are going to see today.
01:20 So ultimately,
01:21 it reduces the instance-to-instance latencies
01:24 as well as instance to maybe on-premises latencies,
01:28 like that.
01:29 So the concept behind the enhanced networking
01:32 is something called SR-IOV with PCI passthrough.
01:35 And what does that mean?
01:36 I'll just talk shortly after this slide.
01:39 But essentially, what it means is you remove
01:43 as much obstacle when the packets moves
01:45 from one system to another.
01:47 When I say obstacle, it's more about the multiple hops.
01:51 You have to trim down that,
01:53 which means if you are using hypervisor,
01:56 can you bypass the hypervisor when a packet flow
01:58 from say, one system to another system?
02:01 And how to do that, we will see.
02:03 And there are two options for you
02:06 to enable the enhanced networking for EC2 instances.
02:10 There is something called, Intel Virtual Function.
02:13 The name is ixgbevf.
02:17 And second option is using Elastic Network Adapter.
02:20 That is in short, ENA,
02:22 and we'll get through more details about it shortly.
02:25 So enhanced networking,
02:26 meaning you can process 1 million packets per second,
02:30 and ultimately, goal is to reduce the latencies
02:32 between instance-to-instance and increase the bandwidth.
02:36 How much data can be sent per second.
3. SR-IOV and PCI Passthrough
时间范围：02:40–04:42
总结 / Summary
中文
SR-IOV 让一块物理网卡（Physical NIC）呈现为多个虚拟功能（Virtual Functions），使不同虚拟机能直接访问网卡能力。PCI 直通（PCI Passthrough）让弹性网络接口（Elastic Network Interface, ENI）在来宾操作系统中看起来像物理连接，减少 Hypervisor 中的包处理，降低 CPU 利用率和延迟并提高 I/O 性能。
English
SR-IOV lets one physical NIC present multiple Virtual Functions so different virtual machines can access the card directly. PCI passthrough makes an Elastic Network Interface appear physically attached to the guest operating system, reducing hypervisor packet processing, CPU use, and latency while increasing I/O performance.
Transcript
02:40 So as I said, let's look
02:41 at the SR-IOV and PCI for enhanced networking.
02:45 So SR-IOV stand for Single Root Input/Output Virtualization
02:49 and basically it's a standard that allows
02:52 your single, physical NIC.
02:54 NIC is Network Interface Card
02:56 to present itself as a multiple virtual NICs
02:59 or you can call it virtual functions.
03:01 So consider a case you have physical host
03:04 and then that physical host has a network interface card.
03:07 Now, if virtual machine which is launched
03:09 on top of physical host has to get
03:11 to that card for sending and receiving the traffic,
03:15 all this traffic goes through, you know,
03:17 the hypervisor layer and which adds some latency,
03:20 some extra processing of the packets.
03:22 Now, SR-IOV and PCI,
03:24 PCI is Peripheral Component Interconnect.
03:28 This allows your VMs to directly access the network card.
03:34 So it is like, enables direct communication
03:37 between your VM and the physical NIC.
03:40 Ultimately, that is this functionalities
03:42 and to enable these features, there are multiple options
03:45 that we are going to see now.
03:47 So again, SR-IOV and PCI passthrough are methods
03:50 of device virtualization that provides
03:52 of course, higher input/output performance
03:55 and lower CPU utilization.
03:58 It allows a single physical NIC to present itself
04:00 as multiple virtual NICs
04:02 or you can call it virtual functions.
04:05 Now, PCI passthrough enables your PCI devices, such as ENI.
04:10 ENI's network interface,
04:12 which you attached to your EC2 instance.
04:14 It's Elastic Network Interface.
04:16 ENI appears as if they're physically attached
04:19 to the guest operating system.
04:21 As you know, EC2 is a virtual machine, right?
04:23 Ultimately, there is a physical server.
04:25 So this PCI passthrough enables that communication.
04:30 So ultimately in combination,
04:32 this allows the lowest latency and high data transfer rate,
04:37 up to like, 1 million packet per second.
04:39 So that is in short is SR-IOV and PCI passthrough.
4. Enhanced Networking Prerequisites
时间范围：04:42–06:38
总结 / Summary
中文
增强联网依赖实例类型（Instance Type）和 AMI/操作系统（Operating System）同时支持。Intel 82599 VF（ixgbevf）课程给出的上限约 10 Gbps，ENA 约 100 Gbps；一个实例按其家族使用其中一种，不能同时使用。若实例类型或镜像任一不支持，就无法启用增强联网。
English
Enhanced networking requires support from both the EC2 instance type and the AMI or operating system. The lecture cites about 10 Gbps for Intel 82599 VF (ixgbevf) and about 100 Gbps for ENA. An eligible instance uses the implementation supported by its family rather than both at once; lacking support on either side prevents enablement.
Transcript
04:42 Now, what you need to remember
04:44 is it's not like you can take any EC2 instance
04:46 and then have the SR-IOV enabled on that.
04:50 There are certain prerequisites and you need to understand
04:54 that because there could be some questions around that,
04:57 so first, it depends on your instance type.
05:01 It's not like on any EC2 machine you can enable
05:03 this SR-IOV or enhanced networking.
05:06 So what you can do,
05:08 you can enable this enhanced networking
05:10 on specific instance type.
05:12 And as I said, there are two ways
05:13 in which you can enable this enhanced networking.
05:16 Option one, you have to have a special driver.
05:21 We call it Intel 82599 virtual function driver,
05:25 also it is called ixgbevf driver,
05:29 and it can provide up to 10 Gbps network bandwidth.
05:34 Now option two, is you can use Elastic Network Adapter,
05:37 ENA, which can provide much higher bandwidth
05:41 up to hundred Gbps.
05:42 Now, these options are like mutually exclusive.
05:45 It's not like you can use both at the same time.
05:48 Depending on your instance type
05:50 and the support for that instance,
05:53 you can use either of these two.
05:55 So the eligible EC2 instance families
05:57 can support either of these two.
05:59 And remember that for enhanced networking,
06:01 it requires support from both EC2 operating system,
06:04 that means EMI and also the instance type.
06:07 And that type should be flagged for enhanced networking.
06:11 Again, not every EC2 instance can be enabled
06:16 with enhanced networking.
06:17 You have to look at the description
06:19 of the EC2 instance type as well as the machine image
06:23 that you use for that EC2 instance.
06:25 Both should support the enhanced networking
06:27 with either of these two options, right?
06:30 So that means there are certain EC2 instance,
06:32 which does not support any of these two.
06:34 That means you can't enable the enhanced networking
06:36 for those instances.
5. Supported Instance Families
时间范围：06:38–07:19
总结 / Summary
中文
不同实例家族使用不同驱动。支持 ENA 的现代实例家族可获得更高带宽；较旧的一组实例使用 Intel 82599 虚拟功能（Intel VF）并通常受 10 Gbps 限制。具体家族列表会随 AWS 更新，因此应查实例类型文档，而不是死记列表。
English
Different instance families use different drivers. Modern ENA-capable families provide higher bandwidth, while an older set uses Intel 82599 Virtual Function and is commonly limited to 10 Gbps. The exact family list changes over time, so consult the instance-type documentation rather than memorizing it.
Transcript
06:38 Okay, so just a quick view on which EC2 instances
06:43 supports enhanced networking,
06:44 and this is not a limited number of EC2 instances.
06:48 There will be many more and this list will grow of course.
06:51 So you don't really need to remember this,
06:54 but this is just for information purpose.
06:58 There are instances which supports
07:00 virtual function, intel virtual function,
07:03 that means they can support up to 10 Gbps network bandwidth.
07:10 And then there are several instances,
07:11 which supports ENA, Elastic Network Adapter, sorry.
07:15 So with that, let's look at how do you check
6. Verify the Enhanced Networking Driver
时间范围：07:19–07:48
总结 / Summary
中文
在 Linux EC2 上可运行 ethtool -i eth0 检查接口驱动。输出为 ixgbevf 表示 Intel VF，输出为 ena 表示弹性网络适配器（Elastic Network Adapter）。这适合实操验证，但课程说明通常不是直接考试命令题。
English
On Linux EC2, run ethtool -i eth0 to inspect the interface driver. ixgbevf identifies Intel VF, while ena identifies the Elastic Network Adapter. This is useful for operational verification, although the lecture notes that the exact command is not usually tested.
Transcript
07:19 whether your EC2 instance
07:20 supports Intel Virtual Functions or ENA.
07:24 What you can do is you can just log into EC2 instance
07:26 if it's a Linux machine and just run this command,
07:29 and it will tell you what driver it is using, right?
07:32 So command it, ethtool minus I,
07:34 and then the network interface.
07:36 So it will show the required output
07:40 and then you can make that out.
07:42 This won't be typically asked in the exam,
07:43 but just, if you want to try that out,
07:45 you can just use these commands.
7. Default EC2 Networking Path
时间范围：07:48–08:48
总结 / Summary
中文
默认 EC2 网络路径从来宾操作系统（Guest OS）进入虚拟化层（Virtualization Layer），再到物理网卡；接收端反向重复该过程。额外的 Hypervisor 处理增加延迟和 CPU 开销，课程给出的普通实例间带宽基准约为 5 Gbps。
English
The default EC2 networking path goes from the guest operating system through the virtualization layer to the physical NIC, and reverses the process at the receiver. Extra hypervisor processing adds latency and CPU overhead; the lecture uses about 5 Gbps as the baseline between ordinary EC2 instances.
Transcript
07:48 Now, let's understand what really happens
07:51 when you use the enhanced networking.
07:54 Now, this diagram shows the typical network flow,
07:58 how it happens.
07:59 For example, there is system A here and system B here,
08:02 and system A wants to send some network packet to system B.
08:05 Of course, you are sitting on a physical host.
08:08 There is a guest operating system,
08:10 that means from your machine,
08:13 the traffic will flow to the virtualization layer.
08:16 From virtualization layer,
08:18 it will go to hardware physical NIC,
08:21 and then this two physical NIC would exchange the data.
08:25 And that's where the, again, at receiving end,
08:29 again, it goes through the virtualization layer,
08:31 and it reaches your virtual machine.
08:33 Now, this is a typical flow and by default,
08:37 EC2 would support up to 5 Gbps of the network bandwidth
08:42 when you use a normal two EC2 machines.
08:45 Now, what happens when you use enhanced networking?
8. Enhanced Networking with Intel VF
时间范围：08:48–09:27
总结 / Summary
中文
使用 Intel 82599 VF 驱动（ixgbevf）时，虚拟机通过旁路虚拟化层（Bypass Virtualization Layer）直接访问物理网卡的虚拟功能。两端均启用后，课程给出的实例间带宽可达约 10 Gbps，关键收益来自减少软件转发与 Hypervisor 跳数。
English
With the Intel 82599 VF driver (ixgbevf), the virtual machine bypasses the virtualization layer and directly accesses a Virtual Function on the physical NIC. When enabled at both endpoints, the lecture cites about 10 Gbps between instances, with the gain coming from fewer software and hypervisor hops.
Transcript
08:48 Basically this virtualization layer
08:49 is taken out of the flow.
08:52 That means you completely bypass virtualization layer.
08:55 And by using these drivers,
08:58 like our Intel Virtual Function from your VM,
09:02 it appears like you are directly talking
09:04 to the hardware network interface card.
09:07 And that's where you can get up to 10 Gbps bandwidth
09:10 between this two system.
09:12 Both are enabled using the Intel ixgbevf driver.
09:18 So remember, you are bypassing the virtualization layer.
09:21 So as you remember, the first method is
09:23 to use Intel VF, and the second method is to use ENA.
9. Enhanced Networking with ENA and Flow Limits
时间范围：09:27–12:23
总结 / Summary
中文
ENA 同样绕过虚拟化层，但课程给出的聚合带宽（Aggregate Bandwidth）可达约 100 Gbps。该数字不是单流带宽：一个网络流（Network Flow）由源/目的 IP、源/目的端口和协议组成；同一集群置放群组（Cluster Placement Group）中单流最高约 10 Gbps，非同组约 5 Gbps。要接近 100 Gbps 需要多个并行流、支持 ENA 的实例类型与操作系统，并且实例本身的网络规格足够。后续 DPDK 与 EFA 面向更专业的 HPC 优化。
English
ENA also bypasses the virtualization layer, but the lecture cites aggregate bandwidth up to about 100 Gbps. This is not a per-flow figure: a flow is defined by source and destination IPs, source and destination ports, and protocol. A single flow reaches about 10 Gbps inside the same cluster placement group or about 5 Gbps otherwise. Approaching 100 Gbps requires multiple parallel flows, ENA-capable instance and operating-system support, and sufficient instance network capacity. DPDK and EFA provide additional HPC-focused optimization.
Transcript
09:27 So with ENA does the same thing,
09:29 it bypasses the virtualization layer,
09:31 but then you can get up to hundred Gbps bandwidth
09:34 between EC2 machines,
09:35 which are using these ENA adapters.
09:38 This is because ENA has much more capacity
09:41 to support that much of the network bandwidth.
09:45 Now, one thing to understand
09:46 that when we say hundred Gbps bandwidth
09:48 between two machines, it is not the bandwidth per flow.
09:51 A flow means a point-to-point connection
09:53 between two machines.
09:55 And it is typically a 5-tuple,
09:57 which consists of source IP, destination IP,
10:00 source port, destination port, and a protocol.
10:03 So if you really want to use full hundred Gbps bandwidth,
10:07 then you would have to use multiple search flow.
10:09 That means multiple parallel connection
10:12 between EC2 machines, right?
10:14 A single flow can get you up to maximum
10:17 of 10 Gbps bandwidth.
10:19 And that too, when those EC2 machines are part
10:22 of a same cluster placement group.
10:25 If they're not part of the same cluster placement group,
10:28 then the maximum bandwidth per flow will be 5 Gbps only.
10:33 So in order to get full hundred Gbps bandwidth,
10:36 you need to use thin multiple flows, right?
10:38 I hope that is clear.
10:40 So with that, these are just the numbers
10:43 with respect to how much bandwidth you can get,
10:46 and that's an aggregated bandwidth.
10:48 And these numbers can go up with advanced hardware,
10:51 but for exam perspective,
10:53 you just need to know this concept.
10:55 Like you can enable the enhanced networking
10:58 with either Intel virtual functions
11:00 or elastic network adapters,
11:03 and their bandwidths are different of course.
11:06 And as of today, you can get hundred Gbps bandwidth
11:09 with ENA and 10 Gbps with Intel virtual functions.
11:13 Now, there are few other mechanisms
11:15 to enhance the performance
11:17 of network throughput using DPDK and EFA as well,
11:22 which is specially designed for HPC workloads.
11:24 And we'll touch upon that in the next lecture.
11:27 But for enhanced networking,
11:28 you need to understand what all things are required.
11:31 So ultimately, you need to have
11:33 the right EC2 instance types,
11:34 which supports enhanced networking.
11:36 You need to have the correct operating system,
11:39 which supports enhanced networking.
11:41 And then you need to have either Intel virtual functions
11:44 or ENA attached to your EC2 machine
11:47 to leverage the enhanced networking functionality.
11:50 And with respect to the overall bandwidth,
11:52 how much EC2 instance can have,
11:54 it also depends on EC2 instance type.
11:57 So you can go to AWS page
11:59 and see how much the maximum bandwidth,
12:01 and particular EC2 instance can provide.
12:04 It is not that you can just pick any EC2 instance
12:07 and you can get up to a hundred Gbps bandwidth
12:09 between these two.
12:10 It also depends on the capacity of EC2 machine.
12:13 So these are all the things you need to remember
12:15 for enhanced networking,
12:17 and then we will see more features in the next lecture.
12:21 So thanks for watching.
整讲总结 / Overall Summary
中文
增强联网（Enhanced Networking）通过 SR-IOV 和 PCI 直通绕过 Hypervisor 网络路径。Intel 82599 VF（ixgbevf）和 ENA 是两种互斥实现，必须由实例类型和 AMI/操作系统共同支持。ENA 的高带宽是多流聚合结果，单流仍受 5/10 Gbps 路径条件限制；选型时要同时看驱动、实例网络规格和集群置放群组。
English
Enhanced networking uses SR-IOV and PCI passthrough to bypass the hypervisor data path. Intel 82599 VF (ixgbevf) and ENA are mutually exclusive implementations that require support from both the instance type and AMI or operating system. ENA's high bandwidth is an aggregate, multi-flow result; a single flow remains subject to roughly 5 or 10 Gbps path conditions. Driver, instance network specification, and cluster placement should all be considered together.