Lecture 38: Basics of Network performance - Bandwidth, Latency, Jitter, Throughput, PPS, MTU
- 课程：[NEW] AWS Certified Advanced Networking Specialty 2026
- 视频时长：20:07
- 来源：Udemy Lecture 26072960
- Slide 数量：12
- 画面切换：00:55、04:13、06:58、09:41、12:35、14:05、15:01、15:34、16:04、17:08、17:33
扫描说明
按 Udemy Transcript 的 313 个时间点逐一定位并比较视频画面，确认 8 个正式 Slide；中间 EC2 控制台与终端演示按目标拆为 4 个稳定阶段，共 12 小节。动画构建保留最终状态，过渡帧经最终总览复核后重新截取。
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. VPC Network Performance and Optimization
时间范围：00:00–00:55
总结 / Summary
中文
本节介绍 VPC 网络性能与优化（Network Performance and Optimization）。目标是在具体场景中，根据吞吐量、带宽、延迟和数据传输路径选择合适配置。
English
This section introduces VPC network performance and optimization. The goal is to select the right configuration for a scenario based on throughput, bandwidth, latency, and the data-transfer path.
Transcript
00:00 All right.
00:00 Now in this section let's understand the VPC network performance and optimization.
00:07 And this is also one of the important topic for your exam.
00:10 Because given a scenario you should be able to decide which will be the best option to optimize the
00:17 overall network performance.
00:19 And when we say network performance, it could mean different things.
00:23 For example, how much the network throughput should be or how much should be the bandwidth.
00:28 So by choosing the right configurations, you can optimize the network when you have to typically transfer
00:34 the data from the source to the destination.
00:37 Now, before we get into the details of all the controls that VPC provides, let's first understand
00:43 the basics of network performance.
00:46 And in that we will discuss some generic terminologies that we use to define the network performance.
00:53 So very first is bandwidth.
2. Core Network Performance Metrics
时间范围：00:55–04:13
总结 / Summary
中文
带宽（Bandwidth）是网络可提供的最大传输速率，通常以 bit/s 表示；延迟（Latency）包括传播、路由跳数和设备处理时间；抖动（Jitter）是包间延迟的变化。吞吐量（Throughput）是成功传输的实际速率，会受到带宽、延迟与丢包影响，因此通常低于理论带宽。每秒数据包数（Packets Per Second, PPS）反映主机/CPU 的包处理能力；最大传输单元（Maximum Transmission Unit, MTU）是路径可承载的最大单包大小。
English
Bandwidth is the maximum available transfer rate, usually measured in bits per second. Latency includes propagation, hop, and device-processing delays; jitter is variation in inter-packet delay. Throughput is the successful transfer rate and is reduced by latency and loss, so it need not equal bandwidth. Packets per second measures host or CPU packet-processing capacity, while the Maximum Transmission Unit is the largest packet a path can carry.
Transcript
00:55 We talk about bandwidth all the time.
00:57 So what does bandwidth mean.
00:58 It means maximum rate of transfer over the network.
01:01 So if you are getting say, 25 GB s bandwidth, typically it means 25 gigabits per second.
01:08 Understand that it's not byte, it's bits per second.
01:11 Now the another parameter that affects your network performance is the latency.
01:16 And typically latency is measured in terms of delay between the two points in a network.
01:22 And in further latency can be seen as the delay, which includes the propagation delays for the signals
01:28 to travel across the medium.
01:30 Because you know that the signal cannot travel at the speed of the light, right?
01:35 We can't achieve that speed, and this introduces additional delays further as the packet goes through
01:41 different hops.
01:43 There are networking devices like routers which may further add the network packet processing delays.
01:50 So together these all constitute to additional latency.
01:54 Then there is a term called jitter right.
01:57 So jitter means variation in the inter packet delays.
02:00 Now Jeter sometimes is not good for, you know, a seamless experience for the network.
02:07 Maybe if you are watching a video, right?
02:09 And if there is a Jeter, you will see that you are not getting a consistent, uh, rate, uh, network.
02:15 Right.
02:16 And that's where you will see the problem, right?
02:19 Then comes the throughput.
02:20 So throughput is the rate of successful data transfer.
02:22 And it is also measured in bits per second.
02:26 Now if you talk about the throughput uh, bandwidth latency and packet loss directly affects your throughput.
02:33 So even though you have bandwidth of say, one Gbps, it does not mean that you will get a throughput
02:39 of one Gbps, because there would be latency and there will be a packet loss due to which you have to
02:44 resend the packets.
02:46 Now all that ultimately results into the throughput.
02:49 So don't consider that you have bandwidth of one Gbps.
02:52 That means you will get throughput also of one Gbps.
02:55 It's a rate of successful data transfer bits from one, uh, endpoint to another endpoint.
03:02 then packet per second.
03:03 This is more with respect to, you know, processing of your packets by your machine or a CPU.
03:09 How many packet it can process per second.
03:12 Right.
03:12 Now it's possible that you have, say, a 25 gigabit per second of bandwidth of the network.
03:18 But is your CPU of your machine is capable of processing those packets at it arrives into your machine
03:25 or leaves your machine.
03:26 So you would need to have corresponding compute power as well.
03:30 Okay.
03:31 Which means that the computing power that you have also affects the network performance.
03:38 And now one more important thing with respect to the network performance is the MTU or maximum transmission
03:44 unit.
03:45 And it means the largest packet that can be sent over the network.
03:50 And it is not always the same depending on your network devices like routers and gateways.
03:50 This MTU may change and as you can imagine higher the M2.
04:01 That means you can send more data in a single package, and lower MTU means you can send less data in
04:08 a single packet.
04:09 So higher the MTU, more the throughput.
04:13 Okay.
3. MTU and Jumbo Frames
时间范围：04:13–06:58
总结 / Summary
中文
互联网常见 MTU 为 1500 字节；超过 1500 字节的帧称为巨型帧（Jumbo Frame），AWS 最高通常为 9001 字节。更大 MTU 让相同数据使用更少的数据包、确认和 CPU 处理，从而提高吞吐量并降低 PPS 压力；但端到端路径的每一跳都必须支持该大小，否则会分片或丢包。
English
A common Internet MTU is 1500 bytes. Frames larger than 1500 bytes are jumbo frames, and AWS commonly supports up to 9001 bytes. A larger MTU moves the same data with fewer packets, acknowledgments, and CPU operations, improving throughput and reducing PPS pressure. Every hop must support the size, however, or fragmentation or packet loss can occur.
Transcript
04:13 So let's talk about MTU right.
04:16 So consider there are two hosts host A and host B and host A wants to send some packet to host B.
04:22 It could be an SSH.
04:23 It could be a ping anything.
04:24 Right.
04:24 So host A initiates that connection and it sends a packet.
04:28 Now typically that packet is uh 1500 bytes.
04:33 That's a standard over the internet.
04:35 Uh you will have the packet of size of 1500 bytes.
04:39 But uh, if you increase this MTU, that means what if you want to send a larger packet right now,
04:46 that would depend on, um, all the network hops you have or routers you have in between.
04:53 Do they support the packet transmission?
04:56 Uh, for the frame of size, more than 1500 bytes?
05:00 If not, then maximum that M2 can have between two hosts would be 1500.
05:07 But in certain cases.
05:09 And we will talk more about that.
05:11 You can also send the packets in the form of jumbo frames.
05:14 Jumbo frames are basically a packet which are greater than 1500 bytes.
05:19 And then we will see how in AWS you can use jumbo frames to improve the network performance.
05:26 Because you can see now, right, if, say your jumbo frame is of size 9000, that means almost you
05:33 are sending the six times the data.
05:35 You would send in one packet of 1500 bytes.
05:38 That means in fewer number of transmissions, you would have those many bytes delivered to the target
05:46 host.
05:47 So most of the network supports the M2 of 1500 bytes.
05:52 That what.
05:52 That's what we just talked about.
05:54 But the jumbo frames are the packets which are larger than 1500 bytes, and it goes up to 9000 bytes.
06:00 In AWS it is 9001 and I'll talk more about that.
06:05 Now, what are the benefits of using jumbo frames?
06:07 Of course you have to send less packets.
06:09 That means you will get more throughput, because sending every packet also means you have to send it,
06:15 acknowledge it.
06:16 If it's a TCP connection and there is a lot of processing out there.
06:20 And then increasing the MTU increases the throughput.
06:23 Because suppose you want to send a one GB of data across two hosts, and if size is 1500, you can do
06:31 your maths and understand how many packets needs to be sent.
06:34 And then those packets needs to be processed at the receiving end or even at the sending end as well.
06:43 So that's called packet per second.
06:45 Now if you would have jumbo frames that means there are less less packets to process.
06:49 So even with lower PS you can send the same amount of data across the host.
06:55 So a jumbo frames gives you that benefit.
4. Path MTU Discovery
时间范围：06:58–09:41
总结 / Summary
中文
路径 MTU 发现（Path MTU Discovery, PMTUD）发送设置“不分片”（Don't Fragment, DF）的探测包。若某路由器只支持更小 MTU，它不能分片该包，而会返回 ICMP “Packet Too Big/Fragmentation Needed” 响应，源主机据此降低包大小。示例路径从 1500 降到 1000；因此 PMTUD 依赖端点之间允许相关 ICMP。
English
Path MTU Discovery sends probes with the Don't Fragment flag set. If a router supports only a smaller MTU, it cannot fragment the packet and returns an ICMP Packet Too Big or Fragmentation Needed response. The source then lowers its packet size; the example path falls from 1500 to 1000 bytes. PMTUD therefore depends on the necessary ICMP being allowed.
Transcript
06:58 So uh, now let's understand how the jumbo frame works frameworks, right?
07:03 So how this M2 is decided when you send the packet.
07:06 So for example there are two hosts.
07:08 And from this host you want to send some packets to this host.
07:12 Now you send first a packet of size say 1500.
07:18 Now every packet has some flags associated with that.
07:22 And those flags has certain meaning out of this.
07:25 We'll be interested in one flag which is called DF or don't fragment.
07:29 So when you send a packet with the df equal to one, that means don't fragment.
07:35 Don't further divide this packet into a smaller packet.
07:39 Right.
07:39 Then router will check at its end.
07:41 Uh, can I support the packet of size 1500?
07:45 It is.
07:45 It says okay.
07:46 Yes I can process that.
07:47 And it will take that packet and route to the next, um, uh, hop basically.
07:53 Now it will reach to the next router in the path.
07:55 And then again, the same question.
07:57 Can I process the, uh, m2 of 1500?
08:01 Now, for some reason, Maybe this router does not support the packet of size 1500.
08:06 It can maximum processes 1000.
08:08 Now in this case we have set the DF equal to one.
08:11 That means this is.
08:13 This router is not allowed to fragment this packet into two packets.
08:18 Now like one maybe of size of 1000 and other.
08:21 Maybe a size of 500.
08:22 That's we are saying don't do that with this fragment equal to one.
08:26 Now in this case you will uh, it will say it's not okay with me to process 1500.
08:32 And even you have set the DF equal to one.
08:34 It will send the response back to the originating server.
08:38 And this is typically an ICMP response.
08:41 Uh, and it will say change MTU to 1000 because I can't process that.
08:45 So in this case now this host would understand that my network only supports the M2 of 1000.
08:53 It will then send the smaller packet and it will say packet with size m2 1000.
08:58 And then ultimately it will reach to the destination.
09:01 Now in this case it means, this host A would have to send two packets instead of one packet.
09:07 So you can also do this MTU path discovery using some Linux commands.
09:12 And I'll show you how to do that.
09:13 So if you see this we are talking about MTU of 1000.
09:18 This network supports and this is just an example okay.
09:22 So for MTU path discovery as I said the response goes over ICMP.
09:26 So ICMP must be allowed between these hosts okay.
09:30 Just remember that.
09:31 All right.
09:31 So until now we talked about the M2 and jumbo frames in general.
09:36 So now let's talk about the jumbo frame in the context of AWS.
5. AWS Jumbo Frame Support and Caveats
时间范围：09:41–12:35
总结 / Summary
中文
AWS VPC 内默认可支持 9001 字节巨型帧（Jumbo Frames），但互联网网关（Internet Gateway）和 NAT 网关（NAT Gateway）路径限制为 1500。区域内 VPC 对等连接（Intra-Region VPC Peering）支持 9001，跨区域对等连接自 2025 年 3 月起支持 8500。Direct Connect 也可支持巨型帧，具体取决于虚拟接口（Virtual Interface, VIF）。集群置放群组（Cluster Placement Group）可进一步减少跳数，但任何不支持大帧的中间设备都可能造成丢包。
English
AWS supports 9001-byte jumbo frames inside a VPC by default, but Internet Gateway and NAT Gateway paths are limited to 1500 bytes. Intra-Region VPC peering supports 9001 bytes, while inter-Region peering supports 8500 bytes as of the March 2025 update. Direct Connect can also support jumbo frames depending on the VIF type. Cluster placement groups can reduce hops, but any intermediary that lacks large-frame support can still cause loss.
Transcript
09:41 So first thing AWS supports the jumbo frames up to size 9001 bytes.
09:47 And it is not like for all the scenarios AWS supports this M2.
09:52 And in the rest of this lecture we will understand different cases in which different Mtus are supported.
09:58 All right.
09:59 So this means that AWS supports jumbo frames.
10:02 And with respect to the VPC, it is enabled by default.
10:06 Now to be honest, it is like EC2 instances which will initiate the traffic.
10:10 So you have to really configure that on the instance level when they transmit the traffic.
10:16 But VPC doesn't restrict that.
10:18 So AWS supports the jumbo frame within a VPC of say 9001 bytes.
10:23 But if the traffic is going over the internet, through the internet gateway or the Nat gateway, then
10:29 the packet size is limited to 1500 m2 only.
10:33 Which means for the internet bound traffic, the jumbo frame is not supported.
10:38 Now further, you know that you can also peer the VPCs, and these VPCs can be in the same AWS region
10:44 or across the regions.
10:46 Now depending on that, if the VPC peering is within the same region, then the jumbo frame of 9001
10:52 byte is supported.
10:53 But if the VPC peering is across the regions, then the jumbo frame size is 8500 bytes.
11:00 Now remember that until March 2025, Five.
11:03 This scenario wasn't covered under jumbo frame, and you were limited to 1500 bytes for inter region
11:09 VPC peering, but now it has been increased to 8500 bytes.
11:15 Further jumbo frames are also supported when the traffic goes over the direct connect between the VPC
11:21 and the on premises network.
11:23 And the size of jumbo frames really depends on how you are connecting over the direct connect.
11:28 For example, if you are using the private VIF, public wave or transit wave and we will shortly discuss
11:35 that next.
11:36 You can also maximize the performance using the jumbo frames.
11:40 If you are having your EC2 instances in a same cluster placement group, and which means that the EC2
11:47 instances are launched together, typically on the same physical hardware, so that there is not much
11:53 network hop and you can maximize the network throughput.
11:58 So while using the jumbo frames, if you launch the EC2 instances in a cluster placement group.
12:04 It gives you maximum network performance.
12:07 And finally, remember that it's not like you just enable the jumbo frame while sending the traffic
12:13 from EC2 instance, and it will increase your performance because you know that if jumbo frames are
12:19 not supported on the intermediary network devices, then you may experience the packet loss because
12:25 if the packet of size over 1500 bytes are sent and the don't fragment flag is set, then that packet
12:31 may be dropped.
12:33 So just remember this.
12:35 Okay.
6. Defining MTU on EC2 Network Interfaces
时间范围：12:35–14:05
总结 / Summary
中文
EC2 是否支持巨型帧取决于实例类型（Instance Type），现代实例通常支持。MTU 配置在弹性网络接口（Elastic Network Interface, ENI）层面；可用 tracepath 发现路径 MTU，用 ip link show <interface> 查看接口值，并以 ip link set dev <interface> mtu <value> 修改 Linux 接口 MTU。
English
Jumbo-frame support depends on the EC2 instance type, with current-generation instances generally supporting it. MTU is configured per Elastic Network Interface. Use tracepath to discover the path MTU, ip link show <interface> to inspect an interface, and ip link set dev <interface> mtu <value> to change it on Linux.
Transcript
12:35 With that let's understand how can you define the M2 on your EC2 instances inside a VPC.
12:42 Right.
12:42 So M2 also depends on instance type.
12:44 Certain instance type supports jumbo frame.
12:47 Certain may not support jumbo frames.
12:49 And there is a list of EC2 instances.
12:51 All typically current generation instances do support the jumbo frames.
12:55 Jumbo frames or M2 is defined at any level, so if EC2 has multiple Enis, you can control or you can
13:02 configure uh, how much, What should be the, uh, should it support the jumbo frame or not?
13:08 Or what should be the size of the packet you want to send, uh, from this EC2 machine?
13:14 Now, you can also check the path discovery MTU path discovery between your machine and the target endpoint
13:21 using the command something like Trace path.
13:23 And I will talk more about that shortly.
13:25 Or we'll just see that in a console.
13:28 And you can also set or check your MTU on your network interfaces for EC2 using a simple command like
13:35 IP link show uh, and then the network interface ID, uh.
13:39 And finally you can also set your MTU, as I said, uh, using this command IP link set dev and then
13:46 your interface ID and then you set your MTU values there.
13:50 So important to understand here is that, uh, you can uh, the, the uh, sorry, MTU are defined at
13:59 any level.
14:00 And you can also set that, uh, using a Linux commands.
14:05 Okay.
7. EC2 MTU Demo Plan
时间范围：14:05–15:01
总结 / Summary
中文
演示在同一 VPC 中使用两台公有子网 EC2 实例：先以另一实例的公有 IP 执行 tracepath，验证互联网路径 MTU；再使用私有 IP 验证 VPC 内路径；最后检查 EC2 网络接口的 MTU。预期公有路径为 1500，私有路径为 9001。
English
The demo uses two EC2 instances in public subnets of the same VPC. It runs tracepath to the other instance's public IP, repeats the test with its private IP, and then checks the interface MTU. The expected public-path MTU is 1500, while the private VPC path should support 9001.
Transcript
14:05 With that, uh, let's, uh, do a quick demo.
14:08 In this demo, what we are going to do is, um, we are going to launch two EC2 instances.
14:13 Uh, and then we'll check the M2 using EC2 public IP.
14:17 That means from one EC2 we will reach to another EC2 using public IP.
14:21 That means the traffic goes over the internet, even though they are part of maybe a same subnet does
14:25 not matter.
14:26 And then we'll observe the M2 in that case.
14:29 And then we will also, um, check the M2 when we are reaching over private IP.
14:34 That means the communication is within the VPC.
14:36 So we are um, assuming that the jumbo frames would be used with M2 9001.
14:42 And in the this case probably we are expecting jumbo jumbo frames are not used and the packet size is
14:48 1500 bytes only.
14:50 And uh, we will just check the M2 on that EC2 interface as well.
14:55 What values it shows.
14:56 So let me just quickly go to my console.
14:59 I have already two.
8. Prepare the EC2 Tracepath Test
时间范围：15:01–15:34
总结 / Summary
中文
控制台确认两台实例位于公有子网，并连接到第一台实例。随后获取第二台实例的公有 IP，作为第一个 tracepath 目标，以强制流量经过互联网路径。
English
The console confirms two instances in public subnets and connects to the first instance. It then retrieves the second instance's public IP for the first tracepath test, forcing traffic onto the Internet path.
Transcript
15:01 I have two instances here.
15:03 Nothing fancy about it.
15:04 Both are in public subnet.
15:07 And what I'm going to do is just to connect to the first instance.
15:10 So this is a shortcut to connect to my instance.
15:13 And what I'm going to do is just grab the maybe a public IP address of the other instance so that I
15:20 can check the MTU.
15:22 So this is my second instance.
15:24 I'm just going to grab its public IP.
15:27 And what I'm going to do is just trace the path and the IP address.
9. Public-IP Path MTU Is 1500
时间范围：15:34–16:04
总结 / Summary
中文
对第二台实例公有 IP 的 tracepath 从本机 9001 开始，但路径最终收敛到 1500，因为流量经互联网网关（Internet Gateway），不支持 VPC 内的 9001 字节巨型帧。这验证了端到端 MTU 由最小路径能力决定。
English
tracepath to the second instance's public IP starts from the local 9001-byte setting but converges on a path MTU of 1500 because the traffic crosses the Internet Gateway. The result confirms that end-to-end MTU is determined by the smallest capability along the path.
Transcript
15:34 Now you can see that uh, initially it started with 9001, but then it defaulted to 1500 because traffic
15:41 is going over to the internet.
15:43 Now it will try 30 times and it will see all possible path to the target destination.
15:48 And then it will show you ultimately uh, path MTU for uh, this communication.
15:53 So I'm just watching this video for a couple of seconds.
15:57 Okay.
15:57 So this has been completed.
15:59 And you can see the final, uh, MTU is 1500.
16:03 And that was expected.
16:04 Now, what?
10. Private-IP Path MTU Is 9001
时间范围：16:04–17:08
总结 / Summary
中文
改用第二台实例私有 IP 后，tracepath 报告 MTU 9001，说明同一 VPC 内通信使用巨型帧（Jumbo Frames）。ip link show 也显示 EC2 接口默认 MTU 为 9001；接口能力和实际路径能力在此一致。
English
Repeating tracepath with the second instance's private IP reports an MTU of 9001, showing that same-VPC traffic uses jumbo frames. ip link show also reports a default EC2 interface MTU of 9001, so the interface and actual path capabilities align.
Transcript
16:04 I'm, I'm trying to do is now just do the same thing.
16:08 But this time using the private IP address of my second EC2 instance.
16:12 And let me see if I have the private IP.
16:15 No I don't, so I'll just grab it here.
16:17 So this is a private IP address.
16:19 I'm just copying it and just putting it here.
16:22 And now let's see what happens.
16:24 So again I'm going to just pause this video for a couple of seconds.
16:26 And then I'll show you this.
16:28 Okay.
16:29 As you can see here my trace has completed.
16:31 And in this case the M2 is 9001.
16:33 That means it's using the jumbo frames.
16:35 So this implies that within VPC, jumbo frames are enabled.
16:39 And they are used for all within VPC communication.
16:42 Right.
16:43 Now one more thing I just wanted to show you how to see the jumbo, uh, how to see the MTU at interface
16:49 level.
16:49 So you can just say IP link show and then your interface ID.
16:56 And here you can see the MTU by default is set to 9001.
17:01 So that's it.
17:01 What we wanted to see here and just going back here, we just did this demo.
11. EC2 MTU Demo Recap
时间范围：17:08–17:33
总结 / Summary
中文
实操结果与预期一致：EC2 公有 IP 路径为 1500，私有 IP 的 VPC 内路径为 9001，接口本身也配置为 9001。判断性能时必须区分 ENI 的本地 MTU 与完整路径 MTU（Path MTU）。
English
The demo matches the prediction: the EC2 public-IP path is 1500, the private same-VPC path is 9001, and the interface itself is configured for 9001. Performance analysis must distinguish the local ENI MTU from the complete path MTU.
Transcript
17:08 You can try yourself as well.
17:09 Uh, there are commands already given here.
17:12 Now, finally, what we want to understand is that, uh, what are the different MTU values when, uh,
17:18 there is a communication between VPC and other entities?
17:22 Like it could be like over the internet, it could be within VPC, it could be over private network.
17:28 Uh, like a direct connect to your on premises.
17:30 So let's uh, see those values.
12. AWS MTU Values by Network Path
时间范围：17:33–20:07
总结 / Summary
中文
路径矩阵：VPC 内为 9001；VPC 端点（VPC Endpoint）为 8500；互联网网关为 1500；区域内/跨区域 VPC 对等分别为 9001/8500。使用虚拟私有网关（Virtual Private Gateway）或 Transit Gateway 的站点到站点 VPN（Site-to-Site VPN）均为 1500。Direct Connect 私有 VIF（Private VIF）支持 9001，Transit VIF 支持 8500，Public VIF 为 1500。考试中应按完整流量路径选择最小 MTU。
English
The path matrix is: 9001 inside a VPC; 8500 through VPC endpoints; 1500 through an Internet Gateway; and 9001/8500 for intra-Region/inter-Region VPC peering. Site-to-Site VPN through either a Virtual Private Gateway or Transit Gateway is 1500. For Direct Connect, a Private VIF supports 9001, a Transit VIF 8500, and a Public VIF 1500. On the exam, use the smallest MTU across the complete traffic path.
Transcript
17:33 So within the AWS if the traffic is within the VPC then jumbo frames are supported.
17:38 So uh, imagine that there are EC2 instances and then these instances communicating with each other.
17:44 They are part of VPC.
17:45 So by default jumbo frames are enabled.
17:48 Now if the traffic is flowing over through the VPC endpoint, that means if you are trying to reach
17:53 to the other AWS services, for example, to the S3 or DynamoDB or, um, SQS kinesis over the VPC endpoints.
18:02 Now in that case, the M2 of size 8500 bytes is supported.
18:07 That means again, jumbo frames are supported.
18:09 Now what are VPC endpoints?
18:11 We will learn that in the next lecture.
18:14 Now if the traffic is going out to the internet via internet gateway, then again the jumbo frames are
18:19 not supported.
18:19 So you are limited to 1500 bytes only.
18:24 Further, with respect to the VPC peering connection, the intra region VPC peering supports the jumbo
18:29 frame of size 9001 bytes, but if it's an inter region VPC peering, then the jumbo frame of size 8500
18:37 bytes is supported.
18:39 And remember, this is relatively new update.
18:44 Now further, if you're connecting your VPC to the on premises network, then there are different scenarios.
18:49 For example, you may connect over the VPN using the Virtual Private Gateway.
18:54 Now in that case, jumbo frame is not supported and MTU is limited to 1500 bytes.
19:00 And same is the case if you are using Transit Gateway to connect over the VPN.
19:06 So in both the cases, jumbo frames are not supported.
19:09 However, if you are using the Direct Connect, then depending on which virtual interface that you use,
19:16 the MTU differs.
19:17 So if you are using the private VIF using which you can connect to the VPCs directly, then the jumbo
19:23 frame of size 9001 byte is supported.
19:26 And otherwise, if you are using the transit wave, which ultimately uses the Direct Connect Gateway,
19:33 which connects to the Transit Gateway.
19:35 In that case, jumbo frame of size 8500 byte is supported.
19:41 And otherwise, if you are using the public wave which connects to the AWS public endpoints, then jumbo
19:48 frame is not supported and the M2 is limited to 1500 bytes.
19:53 So this is all the MTU values that you should typically know while using the jumbo frames.
20:00 So that's it about the jumbo frames and the M2.
20:03 And with that now let's go to the next lecture.
整讲总结 / Overall Summary
中文
本讲从带宽、延迟、抖动、吞吐量和 PPS 建立性能基础，再用 MTU 与巨型帧（Jumbo Frames）说明“更大包可减少处理开销，但路径最小能力决定结果”。Path MTU Discovery、EC2 tracepath 实操和 AWS 路径矩阵共同说明：VPC 内可达 9001，互联网/VPN 常为 1500，VPC 端点、跨区域对等和 Transit VIF 常为 8500。
English
This lecture builds the performance foundation from bandwidth, latency, jitter, throughput, and PPS, then uses MTU and jumbo frames to show that larger packets reduce processing overhead but the weakest path segment controls the result. Path MTU Discovery, the EC2 tracepath demo, and the AWS path matrix establish the practical values: 9001 inside a VPC, commonly 1500 across Internet or VPN paths, and often 8500 for VPC endpoints, inter-Region peering, and Transit VIFs.