Planning and Configuring a GCP Solution

• link to the course
• back to GCP for SysAdmins

Compute

3 main approaches to compute:

• Compute Engine (VMs)
• App Engine (Serverless - App)
• Kubernetes Engine (clusters)

Options available for specific use cases

• Cloud Functions (Serverless - standalone function, event-based)
• Cloud Run (Serverless - runs containers, invoke stateless containers using events)

Machine Types

3 main families:

• Compute Optimized
• Memory Optimized
• Sole Tenant: isolates your VMs and workloads on their own physical servers

Instances are billed based on uptime.

2 other types:

• Custom Machine Types: you can specify the amount of memory and number of vCPUs
• Preemptible VMs (aka Spot Instances): Short-lived VM instances that run for up to 24 hours at a time (not suitable for fault-tolerant workloads).

By combining custom machine types with preemptible VMs, you can really optimize costs for fault-tolerant jobs.

Important discounts:

• Committed use discounts: pay upfront for 1-3 years and get 50-70% discount.
• Sustained use discounts: kick in automatically if you run instances for a certain percentage of time over the course of a month (not applicable for App Engine flexible environment and Cloud Dataflow)

Ways to save money on GCP

• Preemptible VM
• Sustained use discounts
• Committed use discounts

Price Calculator

GCP exposes an API for querying pricing data and offers a simplified web page.

Storage

• Database
  • Data stored with a certain structure
  • Software accessing data
• Storage
  • Generic term
  • GCP has distinct offerings for simple storage

Offerings for simple storage:

• Persistent Disk Storage
  • Usually attached to Compute Engine instances
  • scalability and automatic encryption
• Cloud Filestore
  • Network-Attached Storage (NAS)
  • Critical thing to consider: latency
  • Performance tiers:
    • Standard
    • Premium: offers higher read/write throughput
• Cloud Storage
  • Object-Storage Systems. Work like S3 buckets
  • See also GCP Cloud Developer - 01. GCP Overview#Storage.
  • Cloud Storage Nearline: for data accessed relatively rarely, cheaper than regular Cloud Storage
  • Cloud Storage Coldline: archival purposes, cheapest option.
  • UPDATE: GCP now offers Archive Storage, which is similar to Coldline, but has lower storage costs, higher access costs, no SLA.

SQL-like:

• Cloud SQL
• Cloud Spanner: expensive, scalable, high availability
  • Horizontal scaling is accomplished by adding nodes (more compute resources)
• Bigtable
  • Wide column database
• Firestore
  • Document database
• Memorystore
  • Redis
• Firestore
  • real-time syncing database
  • stores data in JSON

Choosing the right storage option

Issues to keep in mind:

1. What's the data model?
   • Example: Relational vs. NoSQL
   • Cloud SQL: if you need a robust managed relational database service.
   • Cloud Spanner if you need global scalability and can tolerate a bit more latency (and high costs)
   • Memorystore: redis service for ultra-low latency in-memory storage
   • Cloud Firestore: document based
   • Bigtable: wide column storage offering
2. What are my access patterns?
   • read-heavy vs. write-heavy vs. balanced
   • What sort of queries to support?
   • Strict SLA requiring very fast data retrieval? (Cloud SQL)
   • Access from many different regions? (Cloud Spanner)
   • Do we need cache?
3. What is the expected amount of data now and into the foreseeable future?
4. Are there any external constraints around cost, compliance, data location, etc.?

Network

Main services:

• Cloud DNS
  • GCP solutions for configuring DNS records for domains that you own.
• Cloud Load Balancing
  • Traffic routing layer

Network Peering Services

Peering is the process of connecting two separate networks so they can exchange traffic.

• Carrier Peering
  • A service provider acts as a middleman
• GCP Direct Peering
  • Creates a peering connection between your internal network and Google.
• Dedicated Interconnect Solution
  • Similar to Direct Peering, however it creates a dedicated physical connection.
• GCP Partner Interconnect
  • When you need a dedicated physical connection but your company's data center cannot access a GCP colocation facility.
  • Similar to Carrier Peering solution, works through a thrid-party service provider.

Network Security Solutions

• Virtual Private Cloud
  • Designed to protect traffic traveling between GCP and some other endpoint.
• Cloud VPN
  • A simple way of connecting an outside network to GCP endpoints.

A VPC is an isolated subset of a larger network with firewalls to block unauthorized access. Generally it is the key unit of measure for defining your application's network resources.

A GCP VPC can span multiple geographic regions. Instances in US and in Brazil can talk to each other without accessing the public internet.

Niche networking services:

• Cloud CDN
• Traffic Director (service mesh)
• Cloud Armor (DDoS prevention)

About Traffic Director:

A network traffic management tool specifically for service mesh architectures.

A service mesh is an additional software infrastructure layer that controls service-to-service communication.

GCP Load Balancing

There are 6 types, each designed for a particular type of traffic workload. Three for global distribution, and three for regional.

3 Global types:

• HTTP Load Balancer
• SSL Proxy
• TCP Proxy: connections that are not SSL and not HTTP (note: doesn't preserve client IP addresses).

3 Regional types:

• Network TCP/UDP Load Balancer
  • Used for external traffic.
  • 2 specific scenarios where it's the best choice:
    • you need to load balance UDP traffic
    • you are load balancing TCP traffic and need to preserve client IP addresses.
• Internal HTTP(S) Load Balancer: non-accessible from public internet
• Internal TCP/UDP Load Balancer: non-accessible from public internet

Cloud DNS

DNS providers let you publish domain names and route traffic to specific servers and infrastructure.

Start by creating managed zones within Cloud DNS:

• Public Zones
• Private Zones