Netflix & Uber Interview
Cloud Data Platform Cost Optimization Strategies
Reducing cloud costs while maintaining performance
Interview Question
"Your company spends $500K/month on cloud data platform costs. Design a cost optimization strategy that: (1) reduces costs by 30%, (2) maintains performance SLAs, (3) provides visibility into spending, (4) automates cost controls. Include specific tools and techniques for AWS, GCP, and Azure."
Difficulty: Hard | Frequently asked at Netflix, Uber, Amazon, Google
Theoretical Foundation
Cloud Cost Components
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Cloud Data Platform Costs β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β Compute (40-60%): β
β - Virtual warehouses (Snowflake, Redshift) β
β - EMR/Dataproc clusters β
β - Kubernetes pods β
β - Serverless functions β
β β
β Storage (20-30%): β
β - S3/GCS/Blob storage β
β - Database storage β
β - Backup and snapshots β
β β
β Data Transfer (10-20%): β
β - Cross-region transfer β
β - Internet egress β
β - API calls β
β β
β Other (5-10%): β
β - Monitoring and logging β
β - Security and compliance β
β - Support β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Cost Optimization Strategies
1. Right-Sizing
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Right-Sizing β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β Problem: Over-provisioned resources β
β β
β Current: β
β - 10 x r5.4xlarge (64 vCPUs, 256GB) = $4,380/month β
β - Average utilization: 30% β
β β
β Optimized: β
β - 5 x r5.2xlarge (32 vCPUs, 128GB) = $2,190/month β
β - Average utilization: 60% β
β - Savings: $2,190/month (50%) β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
2. Reserved Instances / Savings Plans
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Reserved Instances β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β On-Demand: $0.50/hour β
β 1-Year Reserved: $0.35/hour (30% savings) β
β 3-Year Reserved: $0.25/hour (50% savings) β
β β
β Example: β
β - 10 instances running 24/7 β
β - On-Demand: 10 Γ $0.50 Γ 730 = $3,650/month β
β - 3-Year Reserved: 10 Γ $0.25 Γ 730 = $1,825/month β
β - Savings: $1,825/month (50%) β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
3. Spot Instances / Preemptible VMs
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Spot Instances β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β On-Demand: $0.50/hour β
β Spot: $0.15/hour (70% savings) β
β β
β Use Cases: β
β - Batch processing β
β - Data transformations β
β - CI/CD pipelines β
β - Development/testing β
β β
β Limitations: β
β - Can be reclaimed (2-minute notice) β
β - Not suitable for stateful workloads β
β - May have availability constraints β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
4. Storage Optimization
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Storage Optimization β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β Tiered Storage: β
β - Hot: S3 Standard ($0.023/GB/month) β
β - Warm: S3 Infrequent Access ($0.0125/GB/month) β
β - Cold: S3 Glacier ($0.004/GB/month) β
β - Archive: S3 Glacier Deep Archive ($0.00099/GB/month) β
β β
β Lifecycle Policies: β
β - Move to IA after 30 days β
β - Move to Glacier after 90 days β
β - Delete after 365 days β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
5. Serverless Optimization
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Serverless Optimization β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β Traditional: β
β - Always-on cluster: $2,000/month β
β - Utilization: 40% β
β β
β Serverless: β
β - Pay per query: $0.01/GB scanned β
β - 1000 queries/day Γ 10GB Γ 30 days = $3,000/month β
β β
β When to use serverless: β
β - Variable workload β
β - Ad-hoc queries β
β - Development/testing β
β β
β When to use provisioned: β
β - Predictable workload β
β - High-throughput requirements β
β - Latency-sensitive applications β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Cost Allocation and Tagging
Architecture Diagram
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Cost Allocation β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β Tags: β
β - environment: production, staging, development β
β - team: data-engineering, data-science, analytics β
β - project: etl-pipeline, ml-training, reporting β
β - cost-center: engineering, research, operations β
β β
β Cost Allocation: β
β - Track costs by team/project β
β - Chargeback to business units β
β - Budget alerts and thresholds β
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Code Implementation
Cost Monitoring Dashboard
import boto3
import pandas as pd
from datetime import datetime, timedelta
import json
class CostMonitor:
"""Monitor and optimize cloud costs"""
def __init__(self, aws_access_key, aws_secret_key, region='us-east-1'):
self.ce_client = boto3.client(
'ce',
aws_access_key_id=aws_access_key,
aws_secret_access_key=aws_secret_key,
region_name=region
)
self.cloudwatch = boto3.client(
'cloudwatch',
aws_access_key_id=aws_access_key,
aws_secret_access_key=aws_secret_key,
region_name=region
)
def get_cost_summary(self, start_date, end_date):
"""Get cost summary for date range"""
response = self.ce_client.get_cost_and_usage(
TimePeriod={
'Start': start_date,
'End': end_date
},
Granularity='MONTHLY',
Metrics=['UnblendedCost', 'UsageQuantity'],
GroupBy=[
{'Type': 'DIMENSION', 'Key': 'SERVICE'},
{'Type': 'TAG', 'Key': 'team'}
]
)
return response['ResultsByTime']
def get_cost_by_service(self, days=30):
"""Get cost breakdown by service"""
end_date = datetime.now().strftime('%Y-%m-%d')
start_date = (datetime.now() - timedelta(days=days)).strftime('%Y-%m-%d')
response = self.ce_client.get_cost_and_usage(
TimePeriod={
'Start': start_date,
'End': end_date
},
Granularity='MONTHLY',
Metrics=['UnblendedCost'],
GroupBy=[
{'Type': 'DIMENSION', 'Key': 'SERVICE'}
]
)
costs = {}
for result in response['ResultsByTime']:
for group in result['Groups']:
service = group['Keys'][0]
cost = float(group['Metrics']['UnblendedCost']['Amount'])
costs[service] = costs.get(service, 0) + cost
return sorted(costs.items(), key=lambda x: x[1], reverse=True)
def get_cost_trends(self, months=6):
"""Get cost trends over time"""
trends = []
for i in range(months):
end_date = datetime.now() - timedelta(days=30*i)
start_date = end_date - timedelta(days=30)
cost = self.get_cost_summary(
start_date.strftime('%Y-%m-%d'),
end_date.strftime('%Y-%m-%d')
)
trends.append({
'month': end_date.strftime('%Y-%m'),
'cost': cost
})
return trends
def identify_waste(self):
"""Identify cost waste and optimization opportunities"""
waste = []
# Check for idle resources
response = self.cloudwatch.get_metric_statistics(
Namespace='AWS/EC2',
MetricName='CPUUtilization',
Dimensions=[],
StartTime=datetime.now() - timedelta(hours=24),
EndTime=datetime.now(),
Period=3600,
Statistics=['Average']
)
for datapoint in response['Datapoints']:
if datapoint['Average'] < 10: # Less than 10% CPU
waste.append({
'type': 'idle_instance',
'instance_id': datapoint['Dimensions'][0]['Value'],
'utilization': datapoint['Average'],
'estimated_savings': 100 # Example
})
return waste
AWS Cost Optimization
# ============================================================
# AWS COST OPTIMIZATION
# ============================================================
import boto3
class AWSCostOptimizer:
"""AWS cost optimization strategies"""
def __init__(self):
self.ec2 = boto3.client('ec2')
self.s3 = boto3.client('s3')
self.rds = boto3.client('rds')
def rightsize_instances(self):
"""Rightsize EC2 instances"""
# Get instance utilization
response = self.ec2.describe_instances()
recommendations = []
for reservation in response['Reservations']:
for instance in reservation['Instances']:
instance_id = instance['InstanceId']
instance_type = instance['InstanceType']
# Get CPU utilization
cpu_util = self.get_cpu_utilization(instance_id)
if cpu_util < 30: # Under-utilized
recommended_type = self.get_recommended_type(instance_type, cpu_util)
savings = self.calculate_savings(instance_type, recommended_type)
recommendations.append({
'instance_id': instance_id,
'current_type': instance_type,
'recommended_type': recommended_type,
'cpu_utilization': cpu_util,
'monthly_savings': savings
})
return recommendations
def optimize_s3(self):
"""Optimize S3 storage costs"""
# Get S3 bucket inventory
buckets = self.s3.list_buckets()['Buckets']
recommendations = []
for bucket in buckets:
bucket_name = bucket['Name']
# Analyze access patterns
lifecycle_rules = self.analyze_lifecycle(bucket_name)
if lifecycle_rules:
recommendations.append({
'bucket': bucket_name,
'lifecycle_rules': lifecycle_rules,
'estimated_savings': self.estimate_s3_savings(bucket_name)
})
return recommendations
def create_savings_plans(self):
"""Create savings plans for predictable workloads"""
# Analyze usage patterns
usage_patterns = self.analyze_usage_patterns()
recommendations = []
for pattern in usage_patterns:
if pattern['utilization'] > 70 and pattern['commitment'] > 12:
recommendations.append({
'instance_family': pattern['family'],
'recommendation': 'Compute Savings Plan',
'savings': pattern['potential_savings']
})
return recommendations
def get_cpu_utilization(self, instance_id):
"""Get CPU utilization for instance"""
# Simplified - in production, use CloudWatch
return 25.0 # Example
def get_recommended_type(self, current_type, utilization):
"""Get recommended instance type based on utilization"""
# Simplified recommendation logic
if utilization < 20:
return 't3.medium'
elif utilization < 40:
return 't3.large'
else:
return current_type
def calculate_savings(self, current_type, recommended_type):
"""Calculate monthly savings"""
# Simplified pricing
prices = {
'm5.xlarge': 0.192,
'm5.2xlarge': 0.384,
't3.medium': 0.0416,
't3.large': 0.0832
}
current_cost = prices.get(current_type, 0.192) * 730
recommended_cost = prices.get(recommended_type, 0.0416) * 730
return current_cost - recommended_cost
Snowflake Cost Optimization
-- ============================================================
-- SNOWFLAKE COST OPTIMIZATION
-- ============================================================
-- 1. Monitor credit usage
SELECT
warehouse_name,
SUM(credits_used) AS total_credits,
SUM(credits_used) * 3 AS total_cost_usd
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
WHERE start_date >= DATEADD(day, -30, CURRENT_DATE())
GROUP BY warehouse_name
ORDER BY total_credits DESC;
-- 2. Identify expensive queries
SELECT
query_id,
query_text,
execution_time_ms,
bytes_scanned / 1024 / 1024 AS mb_scanned,
credits_used
FROM SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY
WHERE start_date >= DATEADD(day, -7, CURRENT_DATE())
ORDER BY credits_used DESC
LIMIT 10;
-- 3. Right-size warehouses
-- If average utilization < 50%, use smaller warehouse
SELECT
warehouse_name,
AVG(used_clusters) AS avg_clusters,
MAX(cluster_number) AS max_clusters
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_LOAD_HISTORY
WHERE start_date >= DATEADD(day, -7, CURRENT_DATE())
GROUP BY warehouse_name;
-- 4. Auto-suspend and auto-resume
ALTER WAREHOUSE analytics_wh
SET AUTO_SUSPEND = 60
AUTO_RESUME = TRUE;
-- 5. Multi-cluster warehouse for concurrency
ALTER WAREHOUSE analytics_wh
SET MIN_CLUSTER_COUNT = 1
MAX_CLUSTER_COUNT = 5;
-- 6. Use serverless for ad-hoc queries
-- Serverless auto-scales and charges per second
-- 7. Compact micro-partitions
ALTER TABLE orders COMPACT;
-- 8. Drop unused tables
SELECT table_name, last_altered
FROM INFORMATION_SCHEMA.TABLES
WHERE last_altered < DATEADD(day, -90, CURRENT_DATE());
Cost Alerting
# ============================================================
# COST ALERTING
# ============================================================
import boto3
import json
class CostAlerting:
"""Set up cost alerts and budgets"""
def __init__(self):
self.ce_client = boto3.client('ce')
self.sns = boto3.client('sns')
def create_budget(self, budget_name, limit, email):
"""Create a cost budget"""
response = self.ce_client.create_budget(
AccountId='123456789012',
Budget={
'BudgetName': budget_name,
'BudgetLimit': {
'Amount': str(limit),
'Unit': 'USD'
},
'TimeUnit': 'MONTHLY',
'BudgetType': 'COST',
'CostFilters': {
'TagKey': ['team'],
'TagValues': ['data-engineering']
}
},
NotificationsWithSubscribers=[
{
'Notification': {
'NotificationType': 'ACTUAL',
'ComparisonOperator': 'GREATER_THAN',
'Threshold': 80,
'ThresholdType': 'PERCENTAGE'
},
'Subscribers': [
{
'SubscriptionType': 'EMAIL',
'Address': email
}
]
}
]
)
return response
def create_cost_anomaly_detection(self):
"""Create cost anomaly detection"""
response = self.ce_client.create_anomaly_detector(
AnomalyDetector={
'DetectorName': 'cost-anomaly-detector',
'MonitorType': 'DIMENSIONAL',
'DimensionValue': {
'Key': 'SERVICE',
'Values': ['Amazon Elastic Compute Cloud - Compute']
}
}
)
return response
Infrastructure as Code for Cost Optimization
# ============================================================
# TERRAFORM FOR COST OPTIMIZATION
# ============================================================
# main.tf
"""
# Spot instances for batch processing
resource "aws_instance" "batch" {
count = 3
ami = "ami-0c55b159cbfafe1f0"
instance_type = "c5.xlarge"
instance_market_options {
market_type = "spot"
spot_options {
spot_instance_type = "persistent"
}
}
tags = {
Name = "batch-worker-${count.index}"
team = "data-engineering"
}
}
# Reserved instances for predictable workloads
resource "aws_instance" "production" {
count = 2
ami = "ami-0c55b159cbfafe1f0"
instance_type = "m5.xlarge"
# Use Reserved Instance pricing
lifecycle {
prevent_destroy = true
}
tags = {
Name = "production-worker-${count.index}"
team = "data-engineering"
}
}
# S3 lifecycle policy for cost optimization
resource "aws_s3_bucket_lifecycle_configuration" "data_lake" {
bucket = aws_s3_bucket.data_lake.id
rule {
id = "move-to-ia"
status = "Enabled"
transition {
days = 30
storage_class = "STANDARD_IA"
}
transition {
days = 90
storage_class = "GLACIER"
}
transition {
days = 365
storage_class = "DEEP_ARCHIVE"
}
}
}
"""
π‘
Production Tip: Start with the biggest cost drivers. Typically, compute is 40-60% of costs. Focus on right-sizing, reserved instances, and spot instances for batch workloads. Then optimize storage and data transfer.
Common Follow-Up Questions
Q1: How do you calculate ROI for cost optimization?
def calculate_roi(optimization_cost, monthly_savings, implementation_hours, hourly_rate):
"""Calculate ROI for cost optimization"""
implementation_cost = implementation_hours * hourly_rate
annual_savings = monthly_savings * 12
roi = (annual_savings - implementation_cost) / implementation_cost * 100
payback_months = implementation_cost / monthly_savings
return {
'annual_savings': annual_savings,
'implementation_cost': implementation_cost,
'roi_percent': roi,
'payback_months': payback_months
}
Q2: How do you handle cost allocation across teams?
-- Use tags for cost allocation
SELECT
tag_value as team,
SUM(unblended_cost) as cost
FROM cost_data
WHERE tag_key = 'team'
GROUP BY tag_value
ORDER BY cost DESC;
Q3: How do you automate cost optimization?
# Auto-scaling based on utilization
def auto_scale_cluster(cluster_id, target_utilization=70):
"""Auto-scale cluster based on utilization"""
current_utilization = get_cluster_utilization(cluster_id)
current_nodes = get_cluster_nodes(cluster_id)
if current_utilization > target_utilization + 10:
new_nodes = current_nodes + 1
elif current_utilization < target_utilization - 10:
new_nodes = max(1, current_nodes - 1)
else:
return
scale_cluster(cluster_id, new_nodes)
Q4: How do you handle multi-cloud cost optimization?
- Use cloud-agnostic tools (Kubecost, CloudHealth)
- Standardize tagging across clouds
- Compare pricing across providers
- Use reserved capacity across clouds
β οΈ
Critical Consideration: Cost optimization is not a one-time projectβit's an ongoing process. Set up regular reviews, automate monitoring, and create a culture of cost awareness.
Company-Specific Tips
Netflix Interview Tips
- Discuss multi-cloud cost optimization
- Explain spot instances for encoding
- Mention CDN optimization for streaming
- Talk about data transfer cost reduction
Uber Interview Tips
- Focus on real-time cost optimization
- Discuss geospatial data storage optimization
- Mention ML training cost reduction
- Talk about multi-region deployment costs
Amazon Interview Tips
- Discuss AWS cost optimization services
- Explain Savings Plans and Reserved Instances
- Mention Spot Instances for batch processing
- Talk about S3 Intelligent-Tiering
βΉοΈ
Final Takeaway: Cloud cost optimization requires continuous monitoring and optimization. Focus on the biggest cost drivers first, automate where possible, and create a culture of cost awareness. The goal is not just to cut costs, but to optimize value.