All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Really bad CPU performance?
alexjplant
Member
I just bought a 4-core VPS from a well-known provider on here and am getting what I believe to be pretty not-great CPU benchmarks. Running phoronix-test-suite benchmark sysbench with the CPU test gets me 3286.35 Events Per Second. The reported CPUs are 4 x Intel Xeon E5-2697 v2 (4 Cores) which seems pretty low for this processor type per https://openbenchmarking.org/test/pts/sysbench. I've already had the hosting provider change the physical box this is running on and the result is the same.
The reason that I was suspicious in the first place is because I'm a few hundred miles away from their datacenter on a Gigabit connection and was getting some slight occasional stuttering in my SSH session. Do these numbers make sense? Am I being too picky? To be fair this was a great deal and I haven't run a real workload on the server yet but I want to see if these servers are being egregiously underprovisioned. It's been a while since I had a VPS but even a decade ago $5/mo Atlantic.net VPSes seemed snappier and more responsive than this one is. If anybody has any insight I'd appreciate it. Thanks a bunch!
Comments
Have you accounted for the fact the listed events are based on utilization of all cores without a hypervisor involved?
I can't speak to your results with that specific benchmark but I can speak to general performance of VPS instances. In my personal experience Vultr, Atlantic.net and other large providers that offer VPS servers tend to perform better. The results between different VPS instances no matter the location or time period (tests months apart) are all almost identical. They have their acts together and know what they are doing. They have solid business models in place and don't oversubscribe nearly as much as many/most/some LET providers do. That's why they are larger and generally have more brand recognition to begin with.
Having said all of that, you will find some gems in the rough with smaller/cheaper providers here on LET. I see a very large delta in performance between some of my identically spec'd VPS servers from different LET providers. Some have really bad CPU steal percentages, some are just very spuradic (most likely noisy neighbors), some crawl at a snail's pace all the time and some [few] perform almost as solid as those larger name brand companies. For me, if I'm paying $1/month instead of $5/month and it performs slightly worse but does everything I need it to I consider that a steal. Why would anyone expect to pay 1/5th the cost but get the same level of performance? That's just not the world we lilve in. So if you think whatever you paid is fair and it does everything you need it to do I wouldn't worry too much about it. That's my two cents anyway...
They are called vCPU cores because they aren't CPU cores, but time slices contended with the neighbors.
Most people here use YABS for benchmarking, so you might get more and better replies if you run the same tool as everybody else and post the result here.
https://github.com/masonr/yet-another-bench-script
That CPU is twice as fast per core versus a beefy Pentium 4 in optimal conditions (geekbench 6). It sucks much power, in order for them to make a profit, many users have to share such cores. It doesn't have much real cores and doesn't scale 1:1 under load. Such dedi's are sold here for $50+ for an entire machine, which I consider a scam.
Have you raised a ticket for support to move you to a better node?
Please name and shame if you can't come to a solution. Let's get rid of these scammers
The GB6 score for single core Intel Xeon E5-2697 v2 is just 552.
Assuming that 4 vCores are FULLY dedicated, the GB6 score would be around 1800-2000 range.
Now if the allocation is 25%, then effective GB6 score for 4 vCPU core is around 500
And if the allocation is just 10% (each CPU core is shared by 10 VPS) then GB6 number would be around 200.
Do the Math.
I have a VPS with just one Ryzen 5900 vCPU core, but it is full dedicated to the VPS.
And I am paying $3.75 per month.
It's worse, with 24 threads it reaches 4284. you can't just multiply single core score, in fact single core score is less once under load
Running this now... it's on
Running GB6 benchmark test... *cue elevator music*so I should have something to report shortly. Thanks a bunch for the actionable insight.This seems like it's within the realm of reasonable given what others have been saying in this thread. Anecdotally speaking the box seems faster as things now appear instantly in my shell session... I wonder if something from the provisioning process lingers for a few hours and soaks up cycles. Either way I'll try re-running the Phoronix benchmark again to see if I get higher numbers than before as another comparison.
BENCHOOMER
As other people have pointed out, you probably bought 4 "vCPU." A "vCPU" can be any percentage of a thread of any CPU. Most providers have a huge incentive to oversell cheap VPSes to a point of profit maximization. And since most people are just hosting websites, a vCPU is a very limited resource because it doesn't take much to run WordPress or similar.
The Xeon E5-2697 v2 you mentioned is a 2013 model you can buy on eBay for $25 all day. It's almost shameful in my mind to put new customers on a CPU this old. But there is nothing uncommon about it.
This following is my growing, alphabetical list of providers that have at least one plan that has dedicated CPU, and that tells you the exact CPU model you're buying. Keep in mind, a dedicated "vCPU," "vCore," or even "Core" probably means a thread unless the provider explicitly says you get the whole 2-thread core.
(I have no affiliation with any of these providers and this is based on nothing more than offers I've seen so YMMV. Also, most of these providers have the same mystery plans as every other provider so be sure to find the dedicated offering(s).)
Keep in mind also, availability on dedicated resources tends to be low. Providers that chop up servers so honestly can't do it profitably without selling most or all the capacity up front, so you might have to watch for sales or events with many of these providers.
You can also consider looking for dedicated servers. An older dedicated server (like the server your provider is using lol) might not cost that much more than your VPS if you shop around.