GCP中的网络负载均衡器和HTTP负载均衡器有什么不同

Question

GCP提供了两个负载均衡器，即网络和HTTP(s)，前者工作在第四层上，后者工作在第七层上。

还有一个文档指出，即使是HTTP流量也可以通过网络负载均衡器进行负载均衡。这稍微混淆了在GCP中为web应用程序选择哪个负载均衡器。在为项目选择一个之前，最好先了解其中的差异。

根据工作流、设置、基于区域/区域、会话亲和性的选项以及其他设置，它们之间有何区别？

Answer 1

网络负载均衡器与HTTP负载均衡器

+---------------------+------------------------------------------+------------------------------------------------------+
|       Category      |       Network Load Balancing (NLB)       |             HTTP(S) Load Balancing (HLB)             |
+---------------------+------------------------------------------+------------------------------------------------------+
|     1. Region /     | NLB supports only within a region.       | HLB supports both within cross-region                |
|     Cross-Region    | Does not support cross-region            | load balancing.                                      |
|                     | load balancing                           |                                                      |
+---------------------+------------------------------------------+------------------------------------------------------+
|  2. Load balancing  | NLB is based on IP address, port         | HLB is based only on HTTP and HTTPS                  |
|       based on      | and protocol type. Any TCP/UDP           | protocols.                                           |
|                     | traffic, even SMTP can be                |                                                      |
|                     | load balanced.                           |                                                      |
+---------------------+------------------------------------------+------------------------------------------------------+
|      3. Packet      | Packet inspection is possible and        | HLB cannot inspect packets.                          |
|      inspection     | load balance based on packets            |                                                      |
+---------------------+------------------------------------------+------------------------------------------------------+
|     4. Instance     | No need of creating instance group.      | Managed / UnManaged Instance group                   |
|         Group       | Target pools need to be created.         | is necessary for creating HTTP / HTTPS               |
|                     | Instance can be just tagged to the pool. | load balancer.                                       |
|                     | Ideal for unmanaged instance group       |                                                      |
|                     | where instances are non homogeneous.     |                                                      |
+---------------------+------------------------------------------+------------------------------------------------------+
|     5. Workflow     | Forwarding rules is the starting point.  | This is quite complex in HTTP(s) load balancer.      |
|                     | It directs the request to the            | Global forwarding rulesroutes direct the request     |
|                     | target pools from which compute          | to target HTTP proxy, which in turn checks the       |
|                     | engines will pick the request.           | URL map to determine appropriate backend             |
|                     |                                          | services.  These services in turn direct the request |
|                     | Forwarding rules -> target pool          | to the instance group.                               |
|                     |  -> instances                            |                                                      |
|                     |                                          |                                                      |
|                     |                                          | Global forwarding rules -> Target HTTP proxy ->      |
|                     |                                          | URL map -> Backend Sevices -> instance group         |
+---------------------+------------------------------------------+------------------------------------------------------+
|     6. Types of     | Basic network load balancer which        | 1. Cross-region load balancer uses only one          |
|    load balancer    | directs the request based on IP address, | global IP address and routes the request             |
|                     | port and the protocol within the region. | to the nearest region.                               |
|                     |                                          |                                                      |
|                     |                                          | 2. Content-based load balancer is based              |
|                     |                                          | on the URL path. Different path rules need           |
|                     |                                          | different backend services. for eg: /video           |
|                     |                                          | and /static require two separate backend services.   |
+---------------------+------------------------------------------+------------------------------------------------------+
| 7. Session affinity | Session affinity can be set, but only    | 1. Client IP Affinity: This directs the same         |
|                     | during the creation of target pool.      | client ip to same backend instance by                |
|                     | Once it is set, the value                | computing hash of the IP.                            |
|                     | cannot be changed.                       | 2. Generated Cookie Affinity: Load balancer stores   |
|                     |                                          | cookie in clients and directs the same client to     |
|                     |                                          | same instance with the help of retrieved cookie.     |
+---------------------+------------------------------------------+------------------------------------------------------+
|   8. Health check   | Health check is optional, but network    | Health can be verified by either using HTTP          |
|                     | load balancing relies on HTTP Health     | heath check or HTTPS health check.                   |
|                     | checks for determining instance health.  |                                                      |
+---------------------+------------------------------------------+------------------------------------------------------+

上面的表格是基于我的观点。如果有任何不正确的地方，或者如果我遗漏了什么，请随时发表意见，我会将其添加到表格中。

以下是在GCP中设置HTTP负载均衡器的说明的link。

Answer 2

总的来说，下面是网络负载均衡器和Http负载均衡器之间的区别。

网络负载均衡器(第4层)：这是基于网络变量的流量分布，例如IP地址和目标端口。它是第4层(TCP)及更低层，不考虑应用层的任何内容，例如内容类型、cookie数据、自定义标头、用户位置或应用程序行为。它是无上下文的，只关心数据包中包含的网络层信息，它以这样或那样的方式引导。

应用程序负载均衡器(第7层)这是基于多个变量的请求分布，从网络层到应用层。它是上下文感知的，可以根据任何单个变量直接请求，就像它可以轻松地组合变量一样。应用程序根据其特殊的行为进行负载平衡，而不仅仅是服务器(操作系统或虚拟化层) information.Provides能够根据规则、基于主机或基于路径来路由HTTP和HTTPS流量。与NLB类似，每个目标可以位于不同的端口上。

这两者之间的另一个区别很重要，因为网络负载平衡不能保证应用程序的可用性。这是因为它仅根据网络和TCP层变量进行决策，根本不了解应用程序。通常，网络负载均衡器将根据服务器响应ICMP ping或正确完成三次TCP握手的能力来确定“可用性”。应用程序负载均衡器要深入得多，它不仅能够根据特定页面的成功HTTP GET来确定可用性，还能够根据输入参数验证内容是否如预期的那样。

参考：https://medium.com/awesome-cloud/aws-difference-between-application-load-balancer-and-network-load-balancer-cb8b6cd296a4

Answer 3

此外，我想提一下在GCP中选择正确的负载均衡器(LB)时需要考虑的3 main aspects：

1)全球与区域

2)外部与内部

3)流量类型

请在此chart上找到更多信息。