Persistent Disks & Local SSDs

Storage is a critical component of every Compute Engine VM. Google Cloud offers two primary disk types: Persistent Disks (network-attached, durable storage) and Local SSDs (physically attached, high-performance ephemeral storage). Understanding the differences, performance characteristics, and use cases for each is essential for designing reliable and performant architectures.

---

Persistent Disks

Persistent Disks (PDs) are durable, network-attached block storage devices. They are independent of the VM lifecycle — you can detach, reattach, resize, and snapshot them at any time. Data on a Persistent Disk is automatically encrypted at rest and replicated across multiple physical disks within the zone or region for durability.

Persistent Disk Types

Type	Description	Max IOPS (read)	Max Throughput (read)	Use Case
pd-standard	Standard HDD	7,500	180 MB/s	Bulk storage, cold data, backups
pd-balanced	Balanced SSD	80,000	1,200 MB/s	General-purpose boot and data disks
pd-ssd	Performance SSD	100,000	1,200 MB/s	Databases, high-IOPS workloads
pd-extreme	Extreme performance SSD	120,000	2,400 MB/s	Tier-1 databases (Oracle, SAP HANA)
Hyperdisk Extreme	Next-gen extreme	350,000	5,000 MB/s	Highest-performance workloads
Hyperdisk Balanced	Next-gen balanced	160,000	2,400 MB/s	General workloads with predictable performance
Hyperdisk Throughput	Next-gen throughput	N/A	2,400 MB/s	Large sequential reads (data analytics)

Creating and Attaching Persistent Disks

# Create a standalone Persistent Disk
gcloud compute disks create data-disk-1 \
  --zone=europe-west2-a \
  --size=500GB \
  --type=pd-ssd

# Attach the disk to a running VM
gcloud compute instances attach-disk web-server-1 \
  --zone=europe-west2-a \
  --disk=data-disk-1 \
  --mode=rw

# Inside the VM: format and mount the disk
sudo mkfs.ext4 -m 0 -E lazy_itable_init=0,lazy_journal_init=0 /dev/sdb
sudo mkdir -p /mnt/data
sudo mount -o discard,defaults /dev/sdb /mnt/data

Performance Scaling

Persistent Disk performance scales with disk size. Larger disks provide higher IOPS and throughput, up to the per-VM limits of the machine type. For example, a 1 TB pd-ssd provides roughly 30,000 read IOPS and 30,000 write IOPS, while a 500 GB pd-ssd provides roughly 15,000 each.

Important: Each machine type has maximum disk IOPS and throughput limits. Even with a very large disk, the VM itself may be the bottleneck. Check the machine type documentation for per-VM disk performance limits.

Resizing Persistent Disks

You can increase the size of a Persistent Disk without stopping the VM:

gcloud compute disks resize data-disk-1 \
  --zone=europe-west2-a \
  --size=1000GB

After resizing, extend the filesystem from within the VM:

sudo resize2fs /dev/sdb