VDBench (Virtual Disk Benchmark): Enterprise Storage Validation
Origin: Developed by Oracle / Sun Microsystems storage architects.
Applicability: Enterprise SAN/NAS arrays, raw block LUNs, distributed filesystems, deduplication and compression validation, and multi-host cluster testing.
Vdbench is a multi-threaded benchmark tool written in Java and C, widely recognized as the industry standard by storage array vendors (NetApp, Pure Storage, Dell EMC, IBM). It excels at strict data integrity validation, automated deduplication and compression ratio generation, and multi-host client execution.
1. Core Architecture & Primitives
Vdbench separates hardware definitions from workload generation using four structural building blocks:
+----------------------------------------------------------------------------+
| Vdbench Architecture |
| |
| +------------------------------------+ |
| | Host Definitions (HD) | (Master / Slave JVMs) |
| +-----------------+------------------+ |
| | |
| +-----------------v------------------+ |
| | Storage / Filesystem (SD / FSD) | (Targets: /dev/sdb or /mnt/data) |
| +-----------------+------------------+ |
| | |
| +-----------------v------------------+ |
| | Workload Definitions (WD / FWD) | (Block sizes, R/W mix, threads) |
| +-----------------+------------------+ |
| | |
| +-----------------v------------------+ |
| | Run Definitions (RD) | (Runtime, warmup, reporting) |
| +------------------------------------+ |
+----------------------------------------------------------------------------+
- SD (Storage Definition): Defines raw block devices or disks (e.g.,
sd=sd1,lun=/dev/sdb,openflag=o_direct). - WD (Workload Definition): Defines block sizes, read/write ratios, and random/sequential access.
- RD (Run Definition): Defines test duration, warm-up intervals, and ties WDs to SDs.
- FSD / FWD: Filesystem counterparts to test directory trees, file sizes, and metadata operations.
2. Prerequisites & Installation
Vdbench requires a Java Runtime Environment (JRE 8 or newer).
# Ubuntu / Debian
sudo apt-get update && sudo apt-get install -y default-jre
# RHEL / Rocky Linux
sudo dnf install -y java-latest-openjdk-headless
# Download and unpack Vdbench (from Oracle official release)
mkdir -p ~/vdbench && cd ~/vdbench
# Extract vdbench50407.zip into directory
unzip vdbench50407.zip
chmod +x vdbench
# Run built-in sanity self-test
./vdbench -t
3. The 10 Essential Vdbench Workload Snippets
Snippet 1: Built-in Sanity Self-Test (Validation)
Objective: Validate Java environment, JNI C library bindings, and test directory permissions.
cd ~/vdbench
./vdbench -t
# Expected output: "Vdbench execution completed successfully"
Snippet 2: Raw Block Sequential Write (Throughput Maxima)
Objective: Benchmark sustained sequential write throughput using 1 MiB blocks on a raw block device.
Create seq_write.par:
sd=sd1,lun=/dev/sdb,openflag=o_direct
wd=wd1,sd=sd1,xfersize=1m,rdpct=0,seekpct=0
rd=rd1,wd=wd1,iorate=max,elapsed=60,interval=5
Execute:
./vdbench -f seq_write.par
Snippet 3: Raw Block Sequential Read (Cache Bypass)
Objective: Measure sequential read throughput with Direct I/O cache bypass.
Create seq_read.par:
sd=sd1,lun=/dev/sdb,openflag=o_direct
wd=wd1,sd=sd1,xfersize=1m,rdpct=100,seekpct=0
rd=rd1,wd=wd1,iorate=max,elapsed=60,interval=5
Execute:
./vdbench -f seq_read.par
Snippet 4: Raw Block 4 KiB Random Write (Steady-State IOPS)
Objective: Measure maximum random write IOPS using 32 concurrent threads across the LUN.
Create rand_write_4k.par:
sd=sd1,lun=/dev/sdb,openflag=o_direct,threads=32
wd=wd1,sd=sd1,xfersize=4k,rdpct=0,seekpct=100
rd=rd1,wd=wd1,iorate=max,elapsed=120,warmup=15,interval=5
Execute:
./vdbench -f rand_write_4k.par
Snippet 5: 8 KiB Mixed 70/30 Read/Write Workload
Objective: Simulate enterprise transaction processing with 70% reads, 30% writes, and 100% random seeks.
Create oltp_70_30.par:
sd=sd1,lun=/dev/sdb,openflag=o_direct,threads=64
wd=wd1,sd=sd1,xfersize=8k,rdpct=70,seekpct=100
rd=rd1,wd=wd1,iorate=max,elapsed=180,warmup=20,interval=5
Execute:
./vdbench -f oltp_70_30.par
Snippet 6: Filesystem Definition (FSD) Directory Creation & File Write
Objective: Benchmark filesystem metadata creation by generating a tree of 1,000 files across 10 directories.
Create fsd_write.par:
fsd=fsd1,anchor=/mnt/storage_test,depth=2,width=5,files=100,size=1m
fwd=fwd1,fsd=fsd1,operation=write,xfersize=64k,threads=8
rd=rd1,fwd=fwd1,fwdrate=max,format=yes,elapsed=60,interval=5
Execute:
./vdbench -f fsd_write.par
Snippet 7: Filesystem Random Read & Hotspot Skew
Objective: Measure filesystem read latency when 80% of I/O targets a 20% "hot" working set.
Create fsd_read_skew.par:
fsd=fsd1,anchor=/mnt/storage_test,depth=2,width=5,files=100,size=1m
fwd=fwd1,fsd=fsd1,operation=read,xfersize=16k,threads=16,skew=80
rd=rd1,fwd=fwd1,fwdrate=max,format=no,elapsed=90,interval=5
Execute:
./vdbench -f fsd_read_skew.par
Snippet 8: Filesystem File Delete & Teardown Benchmark
Objective: Measure filesystem metadata unlink and directory reclamation rate by deleting pre-existing files.
Create fsd_delete.par:
fsd=fsd1,anchor=/mnt/storage_test,depth=2,width=5,files=100,size=1m
fwd=fwd1,fsd=fsd1,operation=delete,threads=16
rd=rd1,fwd=fwd1,fwdrate=max,format=clean,elapsed=60,interval=5
Execute:
./vdbench -f fsd_delete.par
Snippet 9: Deduplication & Compression Ratio Qualification
Objective: Generate synthetic data blocks guaranteed to achieve an exact 3:1 deduplication ratio and 2:1 compression ratio to validate storage array efficiency engines.
Create dedup_compress.par:
sd=sd1,lun=/dev/sdb,openflag=o_direct,threads=32
wd=wd1,sd=sd1,xfersize=16k,rdpct=0,seekpct=100
rd=rd1,wd=wd1,iorate=max,elapsed=120,interval=5,dedupratio=3,compratio=2,dedupunit=8k
Execute:
./vdbench -f dedup_compress.par
Snippet 10: Multi-Host Distributed Benchmark (Master/Slave Mode)
Objective: Coordinate multiple client servers against a centralized storage target (e.g., SAN or All-Flash array).
Create multi_host.par:
hd=default,vdbench=/home/admin/vdbench,user=admin
hd=host1,system=192.168.10.101
hd=host2,system=192.168.10.102
sd=sd1,hd=host1,lun=/dev/mapper/mpatha,openflag=o_direct,threads=32
sd=sd2,hd=host2,lun=/dev/mapper/mpatha,openflag=o_direct,threads=32
wd=wd1,sd=(sd1,sd2),xfersize=8k,rdpct=50,seekpct=100
rd=rd1,wd=wd1,iorate=max,elapsed=300,warmup=30,interval=5
Execute from the Master host:
./vdbench -f multi_host.par
4. Analyzing Vdbench Results
Vdbench writes tabular reports to output/summary.html and standard output:
May 28, 2026 ...
interval i/o MB/sec bytes read resp resp resp cpu%
rate i/o pct time max stddev sys+usr
avg_2-120 120 48214.2 376.68 8192 70.0 1.327 14.82 0.412 18.2
resp time: Total service latency in milliseconds ().resp max: Worst recorded latency spike during the interval ().resp stddev: Standard deviation of response times; lower values signify consistent, predictable latency.cpu% sys+usr: Client host CPU consumption, verifying if the test was CPU-bound or storage-bound.