Skip to content

Benchmark honesty—when providers optimize for YABS

VPS Hosting by priya 6 replies 430 views
5 #1

Kindly I have observed that many providers optimize their systems for YABS benchmark only, while real-world performance itself suffers. The disk cache is tuned for 1GB test, network buffers inflated for iperf3, and then we wonder why our database crawls under actual load.

I am requesting the community kindly to share real workload numbers. I propose we test the following:

  • PostgreSQL pgbench, scale 100, 10-minute run
  • nginx static file serving with wrk, 1000 concurrent connections
  • Linux kernel compilation time, make -j$(nproc)

I shall maintain a spreadsheet with all submissions. Cheap VPS only need not apply if it cannot handle actual work. Please share your provider name and plan details.

Thanking you,
Priya

#2

Disk's fine, pipe's meh.

# YABS for comparison
fio 1GB: 1.2 GB/s read, 890 MB/s write
iperf3: 2.1 Gbps single thread

Real workloads from my RackNerd 2-core:

pgbench tps: 847
wrk req/sec: 12,430
kernel compile: 18m 42s

Your pgbench scale is too small. Scale 1000 or nothing.

fio, iperf, geekbench. results or gtfo.
1 #3

Scale 1000 would take forever on these budget boxes. Five hours just to init the database. I tried on my HostHatch 1GB plan and it OOM'd during init itself. Maybe we need tiered categories. Small. Medium. Large. Otherwise the spreadsheet becomes meaningless for the low end. Which is most of us...

#4

Your RackNerd plan, SingaporeRep? I'm on Vultr 4GB and honestly anything under 8GB feels like a toy now. My numbers:

pgbench tps: 1,204
wrk req/sec: 31,800
kernel compile: 11m 15s

This is with zfs and compression on, too. Budget people don't know what they're missing.

#5

What compression ratio on that zfs

#6

I had the same OOM on a 1GB box.

#7

Which kernel were you running, 5 or 6

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft