GCP Professional Cloud Architect Practice Question

A media analytics company is re-platforming an Apache Spark batch job to run on Preemptible Compute Engine VMs. Each VM processes 3 TB of video files stored on attached block storage and must sustain at least 800 MB/s total read throughput and 40,000 random read IOPS during the shuffle phase. Cost optimization is a priority, but the team cannot sacrifice the required performance. Which disk configuration should you recommend for each worker VM to meet the throughput and IOPS goals most economically?

Attach a 3 TB Extreme Persistent Disk (pd-extreme) provisioned for 40,000 read IOPS.
Attach a single 3 TB Balanced Persistent Disk (pd-balanced) to each VM.
Attach three 1 TB Local SSD partitions in a RAID 0 array on each VM.
Attach a single 3 TB Standard Persistent Disk (pd-standard) and enable read caching.

Report Issue

Answer Description

Balanced Persistent Disks reach a maximum of 240 MB/s per TB and 30,000 read IOPS, so a 3 TB pd-balanced volume would deliver roughly 720 MB/s throughput, which is below the 800 MB/s target. Standard Persistent Disks offer even lower throughput and IOPS per GB. Local SSDs provide up to 2,400 MB/s sequential read throughput and more than 680,000 random read IOPS per instance, easily satisfying the performance requirement. Creating a RAID 0 stripe of three 1 TB local SSD partitions offers the needed 3 TB capacity while retaining local SSD linear performance, and it is priced lower per-GB than Extreme Persistent Disk. Extreme Persistent Disks can meet the IOPS requirement but incur a significantly higher $/GB cost and require separate reservation of provisioned IOPS, making them less economical for this workload.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.