Brains4Cloud
ESEN
Practice

Topic 10 of 11

D2 · Planning and implementing a cloud solution

Connectivity and load balancing

The rest of section 2.3: establishing network connectivity (Cloud VPN, VPC Network Peering, and Cloud Interconnect), choosing and deploying load balancers, and differentiating Network Service Tiers. VPC peering is in the previous article.

Hybrid connectivity options

OptionTraffic pathWhen
Cloud VPNThe internet, encrypted with IPsecA quick, inexpensive connection; doesn't reduce egress costs
Dedicated InterconnectA direct physical connection with GoogleHigh bandwidth and you can reach a colocation facility
Partner InterconnectThrough a service providerYou can't reach the colocation facility or need less capacity
Cross-Cloud InterconnectA physical connection between Google and another cloudConnect to AWS, Azure, OCI, or Alibaba Cloud
Direct Peering and Carrier PeeringPeering with Google's edge networkSpecific cases; Google recommends Cloud Interconnect instead

It extends your network to the VPC through an encrypted IPsec connection. It can also connect two VPCs. It can't route traffic to the internet: it is for communication between private networks.

HA VPNClassic VPN
SLA99.99% for most topologies99.9%
RoutingDynamic only (BGP)Static only (policy-based or route-based)
External IPsTwo, picked automatically from separate pools; no forwarding rulesYou create the IP and forwarding rules
IPv6Yes (dual stack or IPv6-only)No
API resourcevpn-gatewaytarget-vpn-gateway
  • For the 99.99% SLA, configure two tunnels, one from each gateway interface; with a single active interface there is no SLA.
  • Each tunnel supports about 250,000 packets per second, between 1 and 3 Gbps depending on packet size.
  • It requires a VPN gateway on the other side (on-premises or another cloud) with adequate redundancy.
  • HA VPN over Cloud Interconnect adds IPsec encryption to Interconnect traffic, which never crosses the internet.

Low-latency, high-availability connections where traffic doesn't cross the internet. The VPC's internal IPs are reachable directly from on-premises, without NAT or VPN tunnels.

Dedicated Interconnect

  • A physical connection between your network and Google's in a colocation facility.
  • 10, 100, or 400 Gbps Ethernet circuits, up to 8 circuits per connection.
  • You must install and maintain routing equipment in the colocation facility.

Partner Interconnect

  • A connection through a supported service provider, with more points of connectivity.
  • VLAN attachments from 50 Mbps to 50 Gbps.
  • No equipment of your own in the colocation facility. For compliance with third parties, add IPsec encryption.
Reliability levelSLA
Critical production99.99%
Non-critical production and development99.9%
No level specifiedNo SLA

Interconnect can reduce egress costs; Cloud VPN alone doesn't. With Private Google Access for on-premises hosts, they reach Google APIs over internal IPs. Cross-Site Interconnect connects your on-premises sites to each other through Google's network.

Cloud Router

A distributed, fully managed service that speaks BGP with your network to create dynamic routes in the VPC. It doesn't forward packets (Andromeda does that): it only manages BGP sessions and route advertisements.

ProductCloud Router
Dedicated, Partner, and Cross-Cloud InterconnectRequired
HA VPNRequired
Classic VPNOptional (usually with static routes)
Cloud NATUses it as a control plane, without BGP sessions

To choose a load balancer, you first decide the traffic type:

Application Load Balancer

HTTP or HTTPS, Layer 7, with a flexible feature set.

Proxy Network Load Balancer

TCP with optional SSL offload, to backends in one or more regions.

Passthrough Network Load Balancer

TCP, UDP, ESP, GRE, ICMP. Preserves the client source IP.

How to choose a load balancer

AspectOptions
External or internalExternal: traffic from the internet. Internal: clients in the same VPC or connected through peering, Cloud VPN, or Interconnect
Global, cross-region, or regionalGlobal: external only, a single anycast IP, and IPv6 termination. Cross-region: internal with backends in several regions. Regional: backends in one region, IPv4 only, or requirements that traffic and TLS stay in one region
Proxy or passthroughProxy: terminates the connection at the load balancer (Google Front Ends or Envoy) and opens a new one to the backend. Passthrough: the packet reaches the backend unchanged and responses go straight to the client (direct server return)
Network tierPremium or Standard (see below)
  • Resilience: a global or cross-region load balancer survives a regional outage; a regional one, a zonal outage. Overall resilience also depends on backend redundancy.
  • Proxy load balancers don't preserve the client IP by default. For authentication on Application Load Balancers, Google recommends Identity-Aware Proxy (IAP).

Load balancer summary

Load balancerModesNetwork tier
Application Load BalancerGlobal external, regional external, classic, regional internal, cross-region internalGlobal external Premium only; regional external Premium or Standard; internal Premium
Proxy Network Load BalancerGlobal external, regional external, classic, regional internal, cross-region internalGlobal external Premium only; regional external Premium or Standard
Passthrough Network Load BalancerRegional external, internal, global external (preview)Depends on the mode

If you don't specify a tier, the load balancer uses Premium. Internal load balancers always use Google's backbone (Premium), without Premium's internet egress charges.

PremiumStandard
RoutingOver Google's backbone; traffic enters and exits at the peering point closest to the clientOver ISP and transit networks; enters and exits near the resource's region
Optimizes forPerformance and latencyCost
FeaturesAllFoundational: Cloud NAT, regional external Application Load Balancer, regional external passthrough Network Load Balancer
SLA99.99%99.9%
  • The tier is chosen per resource (the external IP of a VM or load balancer), so you can mix them.
  • The global external Application Load Balancer and global external proxy Network Load Balancer exist only in Premium. Cloud VPN gateways only support Premium.
  • In either tier, traffic between Google Cloud VMs stays on Google's network.
  • Cloud CDN is always Premium.

Exam scenarios

SituationAnswerWhy
An encrypted, inexpensive connection to on-premises over the internet with 99.99%HA VPNTwo interfaces, BGP, and a 99.99% SLA.
10 Gbps private, not over the internet, with your own equipment in a colocation facilityDedicated InterconnectA direct physical connection with Google.
A private connection but you can't reach a Google colocation facilityPartner InterconnectThrough a service provider.
Connect directly to a VPC in another cloudCross-Cloud InterconnectA physical connection between Google and the other provider.
Encrypt traffic that goes over InterconnectHA VPN over Cloud InterconnectIPsec over the VLAN attachments.
A global HTTPS website with a single anycast IP and backends in several regionsGlobal external Application Load BalancerLayer 7, Premium Tier.
A game that uses UDP and needs the client source IPExternal passthrough Network Load BalancerDoesn't terminate the connection and supports UDP.
Terminate TLS only in one region for complianceRegional load balancerTLS is terminated in the deployed region.
Cut egress costs for an app without strict latency requirementsStandard TierUses ISP networks instead of Google's backbone.

Common exam traps

❓ Does HA VPN support static routes?

No. HA VPN only uses dynamic routing with BGP (Cloud Router). Classic VPN uses static routes.

❓ Is Cloud Interconnect traffic encrypted?

Not by itself. To encrypt it, use HA VPN over Cloud Interconnect.

❓ Does Cloud VPN reduce egress costs?

No. Dedicated and Partner Interconnect can.

❓ Can an internal load balancer be global?

No. For internal load balancers with backends in several regions, use cross-region mode.

❓ Does an Application Load Balancer preserve the client source IP?

Not by default, because it is a proxy. Passthrough Network Load Balancers do preserve it.

❓ Can Standard Tier be used with the global external Application Load Balancer?

No. It exists only in Premium Tier.

Did you get this topic?

Test what you just learned

An online store has backends in us-central1 and europe-west1. It wants a single IP for all its users worldwide, HTTPS terminated close to them, and traffic distributed across the backends. Which load balancer should it deploy?

Ready to test "Connectivity and load balancing"?

Original questions, each with a full explanation.

Start a mock exam