Verify Performance with YABS

We publish benchmarks on VPSBenchmarks every quarter because a claim you cannot check is marketing. You can run the same test on your own server in about ten minutes, read the numbers that matter, and know what "steal" means when you see it. This guide shows how.

Run YABS#

YABS (Yet Another Benchmark Script) tests disk with fio, network with iperf3, and CPU with Geekbench. On a fresh or idle server, as root:

curl -sL https://yabs.sh | bash

It prints results as it goes and finishes with a Geekbench link. Add -i to skip the network test, -g to skip Geekbench, or -6 to run IPv6 iperf; curl -sL https://yabs.sh | bash -s -- -h lists the flags. Do not run it while your application is busy; you will benchmark your application instead of the server.

Read the numbers#

Disk (fio)

Look at the 4k random read/write IOPS and the 1m sequential throughput. NVMe RAID on the Performance line should show tens of thousands of 4k IOPS; hardware RAID-10 SSD on the Foundational line lower but steady. A number that swings wildly between runs is worth a second look.

Network (iperf3)

Send and receive to test hosts on several continents. West Coast and Asia-Pacific targets from Seattle should be strong; distant targets are limited by the far end as much as by us. Your port is 1 Gbps; a single TCP stream to a distant host will not saturate it, and that is physics, not a fault.

CPU (Geekbench)

The single-core score is what most web, database, and game workloads feel. Compare it with the public results for the same plan on VPSBenchmarks; if yours is far off, something on your server is competing for the core (see steal, below), or you have found something we want to hear about.

What CPU steal means#

Steal time is CPU your VM wanted but the host gave to another VM at that instant. A little is normal on any shared host; a lot means an oversold node. We cap average node load at 60% and stop selling on a full node precisely so this number stays low.

vmstat 1 10        # last column "st" is steal, as a percentage
top                # the %Cpu line shows "st" too

Single digits under load are fine. If you see steal in the double digits for sustained periods, take a screenshot of vmstat and the portal graphs and open a ticket; that is exactly the kind of report we want.

Compare fairly#

• VDS plans have dedicated cores and should show near-zero steal; VPS plans share cores, and the cap keeps them honest.
• Foundational plans are built for value, not peak clocks; compare them with Foundational results, not Performance ones. See Performance VPS, Foundational VPS, or VDS?.
• Run at a quiet time and again at a busy one; the difference tells you more than either number alone.

Publish yours#

If you like what you see, submit the YABS output to VPSBenchmarks; if you do not, tell us. Public numbers keep everyone honest, us included.