面向中文开发者的 AWS 实战知识库 代维服务
上云咨询

SAP-C02 考试领域深度解析:迁移与现代化

导出 PDF 浅色排版、带版权水印,可在打印窗口另存为 PDF
SAP-C02 考试领域深度解析:迁移与现代化

TL;DR

  • Domain 4“加速工作负载迁移与现代化”占 SAP-C02 的 20%——约 75 道题中的 15 道——并且是整个考试中最具工具特定性的领域。
  • 该领域几乎每道题都归结为三个决策:哪种迁移工具符合给定约束(MGN vs DMS vs DataSync vs Snowball)、切换如何排序,以及暂存 VPC 能否在物理上容纳复制服务器。
  • 如果你已持有 SAA-C03,总共预算 60–80 小时;仅 Domain 4 就安排 8–10 小时的控制台实操,因为阅读复制延迟和在控制台看着它攀升不是一回事。

Need "staging VPC" maybe "暂存 VPC" or "中转 VPC". In MGN, "staging subnet" = 暂存子网. Use "暂存". "replication servers" = 复制服务器.

Next:

SAP-C02 实际面向谁,以及 SMB 工程师会在哪里卡住

SAP-C02 假设你已经设计过多账户 AWS Organizations 布局、混合连接,并至少做过一次真实迁移。没有强制前置条件,但题目面向的是那种曾与客户争论过为什么 2 TB Oracle 表要走 DMS、而 400 TB 文件共享要放到 Snowball Edge 上的人。

我看到第一次考试失败的人通常是 SAA-C03 的强持有者。他们能凭记忆构建 VPC,但从未必须证明一个切换窗口合理,或解释为什么某台复制服务器落在了错误的子网。考试考查的是在冲突约束下的判断力,而不是定义。

还要一开始就决定你是否真的需要它。SAP-C02 是一个广度认证——它不会让你成为迁移专家,也不会让你成为安全专家。如果你的工作是单账户 SMB 工作负载,SAA-C03 加上动手实践,每个学习小时覆盖的面更广。

Need "SMB" maybe keep. "attempt one" = 第一次尝试.

当前考试指南中的领域权重

AWS 为 SAP-C02 公布了四个领域。权重在 2026 考试指南中保持稳定。

Table:

领域 权重 约 75 题中的题量 考生失分点
1. 为组织复杂性设计解决方案 26% ~20 SCP 边缘情况、Control Tower 护栏、多账户网络
2. 为新解决方案设计 29% ~22 固定预算下的成本最优与韧性权衡
3. 现有解决方案的持续改进 25% ~19 合理调整规模、可观测性缺口、DR 测试节奏
4. 加速工作负载迁移与现代化 20% ~15 两个约束冲突时的工具选择

A 75-question form includes unscored pretest items, and AWS does not tell you which ones they are. Treat every question as scored. Domain 4 at 20% is not the largest block, but it has the highest density of "you either know the tool or you don't" items — which makes it the cheapest domain to convert into points.

Translate: 一份 75 题的试卷包含不计分的预测试题,AWS 不会告诉你哪些是。把每道题都当作计分题。Domain 4 占 20%,不是最大的板块,但它拥有最高密度的“你要么知道这个工具,要么不知道”的题目——这使它成为最容易转化为分数的领域。

Domain 4 深度解析:将迁移工具映射到约束

任务陈述分为两半:选择迁移策略,以及现代化现有架构。策略题通常会给你一个源环境和一项约束——停机窗口、带宽、数据库引擎、合规性——然后要求你选择工具。老实说,下表覆盖了你将看到的大部分内容。

Table:

场景约束 正确工具 陷阱答案
块级服务器迁移、最小化切换停机时间、本地 VMware 或物理机 AWS Application Migration Service (MGN) “用 DMS 重新托管”——DMS 仅用于数据库
Oracle 或 SQL Server 迁移到 Aurora PostgreSQL DMS 加 Schema Conversion Tool 假设 SCT 能干净地转换存储过程
MySQL 迁移到 Aurora MySQL DMS,不需要 SCT(同构) 添加 SCT,并在一个学习答案上浪费时间
NFS 或 SMB 共享,需要持续增量同步 DataSync 当同步是单向且有限时选 Storage Gateway
PB 级数据集,链路数周饱和 Snowball Edge Storage Optimized Snowmobile——该服务已退役,不要选
第三方 SFTP 合作伙伴 Transfer Family 用 DataSync 进行交互式合作伙伴传输

Need "Rehost" maybe "重新托管". Good.

MGN is the piece people under-prepare. Each source server gets its own replication server in a designated staging subnet, plus staging EBS volumes, plus a launch template that defines the test and cutover instances. That is three design decisions the exam can probe — subnet sizing, EBS volume type and throughput for the staging area, and whether the launch template subnet has a route to your target workload.

Translate: MGN 是人们准备不足的部分。每台源服务器都会在指定的暂存子网中获得自己的复制服务器,外加暂存 EBS 卷,以及一个定义测试实例和切换实例的启动模板。这是考试可以探查的三个设计决策——子网大小、暂存区域的 EBS 卷类型和吞吐量,以及启动模板所在子网是否有通往目标工作负载的路由。

On modernization half… translate:
在现代化这一半,问题围绕运维开销。Strangler-fig 重构、容器化到 ECS 或 EKS、将计划任务迁移到 EventBridge 加 Lambda,以及将自管理 MySQL 重新平台化到 Aurora,都是常见模式。当某个答案选项说“保留单体,但把它放到更大的实例上”时,它几乎总是错的。

Need "replatforming" = 重新平台化.

那个在 02:40 响起的告警,比文档教会我更多:MGN 暂存子网耗尽

We were moving 12 source servers from a colo into us-east-1 in September 2026. The VPC had been hand-built, and the staging subnet got a /28 because the reasoning at design time was "it's just replication traffic."

Translate: 2026 年 9 月,我们正将 12 台源服务器从托管机房迁入 us-east-1。VPC 是手工搭建的,暂存子网拿到了一个 /28,因为设计时的理由是“这只是复制流量”。

At 02:40 on a Tuesday, a CloudWatch alarm on replication lag fired. In the MGN console, colo-sql-03 showed Data replication status: STALLED and Lag duration: 3 hours 12 minutes. My first instinct was bandwidth, so I checked the Direct Connect — 18% utilization, nowhere near saturated. That killed the easy theory.

Translate: 周二 02:40,一个针对复制延迟的 CloudWatch 告警触发了。在 MGN 控制台中,colo-sql-03 显示 Data replication status: STALLED 和 Lag duration: 3 hours 12 minutes。我的第一反应是带宽,所以我检查了 Direct Connect——利用率 18%,远未饱和。这排除了那个容易的解释。

Then I opened the replication server launch log… Translate: 然后我打开了该源服务器的复制服务器启动日志。上面写着:Replication server launch failed: InsufficientFreeAddressesInSubnet。根本不是网络问题。

Verified with: 用以下命令验证:

Code block unchanged:

aws ec2 describe-subnets 
  --filters "Name=tag:Name,Values=mgn-staging-use1-az1" 
  --query 'Subnets[*].[SubnetId,CidrBlock,AvailableIpAddressCount]' 
  --region us-east-1 --output table

Add close.

The table came back with `AvailableIpAddressCount` at `0` on a /28 — 11 usable addresses, consumed by ten replication servers plus the launch-template artifacts. I created a /24 staging subnet, repointed the MGN replication settings at it, and re-seeded the two affected servers from scratch. That re-seed took 5 hours 40 minutes for a 1.8 TB volume.

Translate: 返回的表格显示,在一个 /28 上 `AvailableIpAddressCount` 为 `0`——11 个可用地址,被十台复制服务器加上启动模板产物消耗殆尽。我创建了一个 /24 暂存子网,将 MGN 复制设置重新指向它,并从头重新播种了两台受影响的服务器。对一个 1.8 TB 的卷,这次重新播种花了 5 小时 40 分钟。

"The exam lesson: when a scenario says "migrate dozens of servers with MGN," staging subnet capacity is a real design constraint. The ops lesson: `STALLED` is rarely bandwidth first — read the launch log before you touch the network."

Translate: 考试教训:当场景说“用 MGN 迁移数十台服务器”时,暂存子网容量是一个真实的设计约束。运维教训:`STALLED` 很少首先是因为带宽——在碰网络之前先读启动日志。

## 报名、费用和形式

Table:
| 项目 | 详情 |
| 费用 | 美国为 300 美元;其他地区按当地货币及当地费率计费 |
| 形式 | 75 道题,180 分钟,单选题和多选题 |
| 通过分数 | 在 100–1000 的缩放分数中为 750 |
| 交付方式 | Pearson VUE 考试中心或 OnVUE 在线监考 |
| 有效期 | 自通过之日起 3 年 |
| 重考 | 两次尝试之间适用等待期;预约前查看当前政策 |
| 前置条件 | 无强制要求;SAA-C03 是合理的前置认证 |

Book the online proctored slot only if you have a quiet room and a wired connection — a dropped session mid-exam is a genuinely bad afternoon. Otherwise a test center removes an entire category of risk.

Translate: 只有当你有一个安静的房间和有线连接时,才预约在线监考考位——考试中途断线会让那个下午变得非常糟糕。否则,考试中心会消除一整类风险。

## 真正能围绕工作安排的学习时间线

Eight weeks at 8–10 hours per week is realistic for someone with SAA-C03 and daily AWS exposure. Compress it to five weeks only if you already run migrations.

Translate: 对于持有 SAA-C03 且每天接触 AWS 的人来说,每周 8–10 小时、持续八周是现实的。只有当你已经在运行迁移时,才把它压缩到五周。

Bullets:
- Weeks 1–2: Domains 1 and 2. Organizations, SCPs, Transit Gateway, Direct Connect, DR patterns. Read the whitepapers, not just the docs.
Translate: - 第 1–2 周:Domain 1 和 2。Organizations、SCP、Transit Gateway、Direct Connect、DR 模式。阅读白皮书,而不只是文档。
- Weeks 3–4: Domain 4, hands-on. Launch a t3.medium in us-east-1, install the MGN agent on a test VM, watch a replication run complete, then do a test cutover.
Translate: - 第 3–4 周:Domain 4,动手实践。在 us-east-1 启动一台 t3.medium,在测试 VM 上安装 MGN 代理,观察一次复制运行完成,然后做一次测试切换。
- Weeks 5–6: Domain 3. Right-sizing, cost tools, observability gaps, patching strategy.
Translate: - 第 5–6 周:Domain 3。合理调整规模、成本工具、可观测性缺口、补丁策略。
- Weeks 7–8: Practice exams. Anything under 75% on a full-length set means one more week.
Translate: - 第 7–8 周:模拟考试。任何一套完整长度的模拟题低于 75%,就意味着再加一周。

The single highest-value lab is a DMS task from a small MySQL RDS instance into Aurora MySQL with CDC enabled. You will see how `MaxFullLoadSubTasks` interacts with a small replication instance class, and that intuition transfers straight into exam questions about throughput tuning.

Translate: 单项价值最高的实验是一个 DMS 任务:从一个小型 MySQL RDS 实例迁移到启用 CDC 的 Aurora MySQL。你会看到 `MaxFullLoadSubTasks` 如何与小型复制实例类交互,而这种直觉会直接迁移到关于吞吐量调优的考试题中。

## 值得记住的高产出陷阱

A few facts come up repeatedly and cost people easy points.

Translate: 有几个事实反复出现,并让人丢掉容易拿的分数。

- MGN is free for 90 days per server; after that it is billed per hour, so long-running replications have a cost tail.
Translate: - MGN 每台服务器 90 天内免费;之后按小时计费,因此长时间运行的复制会有一段成本尾巴。
- Raising `MaxFullLoadSubTasks` above the default of 8 on an undersized replication instance makes the load slower, not faster.
Translate: - 在规格不足的复制实例上,将 `MaxFullLoadSubTasks` 提高到默认值 8 以上,会让加载变慢,而不是变快。
- Migration Hub tracks migrations — it does not perform them. Answer options that use it as a migration engine are wrong.
Translate: - Migration Hub 跟踪迁移——它不执行迁移。把它当作迁移引擎的答案选项是错误的。
- DataSync requires an agent on-premises for NFS/SMB sources; a 10 Gbps agent host is the usual sizing target for large backfills.
Translate: - 对于 NFS/SMB 源,DataSync 需要本地代理;对于大型回填,通常以 10 Gbps 的代理主机作为规格目标。
- Snowball Edge comes in Storage Optimized and Compute Optimized variants, and picking the wrong one is a classic distractor.
Translate: - Snowball Edge 有 Storage Optimized 和 Compute Optimized 两种变体,选错是经典的干扰项。

None of these are hard concepts. They are recall items dressed up in scenario prose, and they are the reason domain 4 is worth drilling even though it is only 20%.

Translate: 这些都不是难概念。它们是披着场景叙述外衣的记忆项,也正是尽管 Domain 4 只占 20% 仍值得反复练习的原因。

## 考试技巧:每题 2.4 分钟,不能更多

180 minutes across 75 questions is 2.4 minutes each, and the scenario items will eat four if you let them. Answer in one pass, flag anything you cannot resolve in 90 seconds, and come back. Multiple-response items tell you how many to select — if it says "choose three," exactly three are correct and a fourth selection is a wrong answer.

Translate: 180 分钟做 75 道题,每题 2.4 分钟,而场景题如果放任不管会吃掉四分钟。第一遍就作答,把任何 90 秒内无法解决的问题标记出来,然后回头处理。多选题会告诉你需要选几个——如果它说“选择三项”,那么恰好三个正确,选第四个就是错误答案。

Watch the qualifier in the last sentence. "Most cost-effective," "least operational overhead," and "highest availability" produce different correct answers on identical scenarios. I have watched people pick the multi-region active-active design on a question that asked for the cheapest option and then wonder why the score came back at 712.

Translate: 注意最后一句话中的限定词。“最具成本效益”“最低运维开销”和“最高可用性”在相同场景下会产生不同的正确答案。我见过有人在要求最便宜选项的题上选择了多区域主动-主动设计,然后纳闷为什么分数回来是 712。

## 什么时候不要考 SAP-C02

Skip it if you do not hold SAA-C03 yet, or if you have under a year of hands-on AWS work — the scenarios assume context you cannot cram. Skip it if your actual goal is depth in one area; the Security and Data Analytics specialty exams cover narrower ground better. And skip it if your employer just needs someone competent at lift-and-shift. Two weeks of real MGN cutovers will teach you more about migration than this exam will.

Translate: 如果你还没有 SAA-C03,或者你的 AWS 动手工作不足一年,就跳过它——这些场景假设了你无法靠突击获得的背景。如果你的实际目标是在某一领域深入,就跳过它;Security 和 Data Analytics 专业考试更好覆盖更窄的领域。如果你的雇主只是需要一个能胜任 lift-and-shift 的人,也跳过它。两周真实的 MGN 切换会比这场考试教会你更多迁移知识。

If you need a structured study plan and someone to sanity-check your lab designs, the folks at awsgood.com run exactly that kind of hands-on prep alongside managed AWS operations work.

Translate: 如果你需要结构化学习计划,并需要有人帮你 sanity-check 实验设计,awsgood.com 的人正好在托管 AWS 运维工作之外提供这类动手备考。

Need "sanity-check" maybe "合理性检查". Good.

## 来源

- [AWS Certified Solutions Architect – Professional 考试指南 (SAP-C02)](...)
- [AWS Certification – Solutions Architect Professional](...)
- [什么是 AWS Application Migration Service?](...)
- [AWS Database Migration Service 用户指南](...)

Need keep URLs unchanged. Titles can translate, but first two maybe keep English? We translated. Good.

参考来源

常见问题

How much of the SAP-C02 exam is migration and modernization?

Domain 4... Translate:

Is SAP-C02 harder than SAA-C03?

Yes, meaningfully. SAA-C03 asks you to design a solution; SAP-C02 asks you to choose between two defensible designs under a conflicting constraint such as cost versus recovery time. Candidates with strong SAA-C03 scores often land in the low 700s Translate:

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注