@chimichurri said:
Hosthatch BF2022 2TB Amsterdam, 110 dollars every 3 years
... @dosai Big thanks for the tip! Just like you said, network is fast, my man
I clearly wasn't paying enough attention to everything that was going on! I thought the Polar Bear was the best value 2TB backup option!
Indeed, I'd missed noticing it too. Unfortunately now gone; maybe back at New Years?
What's interesting about this is that the small 4k IO performance is awful in the VM, about 1/4 speed. But, what I didn't expect is that the blocks that are page size and up are somewhat quicker on the VM. I'm not sure why this is, but maybe the readbacks are more likely to end up in RAM cache.
depending on how you deploy your VMs and especially if you're using qcow2 images, it might be worth it to play around with preallocation and the cluster_size. I usually set the latter to something like 512k or 1M which can help quite a bit with IO.
though this all is much more noticable on spinning rust setups compared to ssd/nvme nowadays.
What jumped out at me was the tiny 5 point difference between the single core Geekbench 5 scores, 1441 for the Host and 1436 for the VM. I would have expected the difference to be between 50 and 150. How do you have such a low virtualization penalty?
@Mason said: Hey @ralf - just wanted to say this was a very awesome and interesting read! Thanks for taking the time to test out the host vs a host-sized VM and compare the two results! If I'm not mistaken, I believe @Not_Oles has done some lengthy experiments testing YABS VM performance vs host performance as well in the past and had similar findings.
@Not_Oles said: What jumped out at me was the tiny 5 point difference between the single core Geekbench 5 scores, 1441 for the Host and 1436 for the VM. I would have expected the difference to be between 50 and 150. How do you have such a low virtualization penalty?
I'll try another yabs run again when I get some free time, but I wonder if you'd lost performance because you'd not added the necessary things to the grub command line.
In /etc/default/grub find GRUB_CMDLINE_LINUX and add iommu=pt intel_iommu=on
WARNING: I think this should be iommu=pt amd_iommu=on for AMD, but I've not tried it.
Oh, and the other thing... I've used isolcpus=1,2,....,7,9,10,...,15 so that linux can only schedule on cores 0 and 8 (the same physical core). Then in my vcpupin, I explicitly bound vcpu 0 to 1, vcpu 1 to 2, etc... This means that the core will be completely idle except for the VM process. If you don't do that, it could up end up competing with all the normal kernel things on the host. Remember that the linux scheduler will always try to put tasks on the lowest numbered cores it can.
@Not_Oles said: What jumped out at me was the tiny 5 point difference between the single core Geekbench 5 scores, 1441 for the Host and 1436 for the VM. I would have expected the difference to be between 50 and 150. How do you have such a low virtualization penalty?
I'll try another yabs run again when I get some free time, but I wonder if you'd lost performance because you'd not added the necessary things to the grub command line.
In /etc/default/grub find GRUB_CMDLINE_LINUX and add iommu=pt intel_iommu=on
WARNING: I think this should be iommu=pt amd_iommu=on for AMD, but I've not tried it.
Oh, and the other thing... I've used isolcpus=1,2,....,7,9,10,...,15 so that linux can only schedule on cores 0 and 8 (the same physical core). Then in my vcpupin, I explicitly bound vcpu 0 to 1, vcpu 1 to 2, etc... This means that the core will be completely idle except for the VM process. If you don't do that, it could up end up competing with all the normal kernel things on the host. Remember that the linux scheduler will always try to put tasks on the lowest numbered cores it can.
Thanks for your 🌟 suggestions! 🌟 I will study up, try some stuff, and post something. Thanks again!
@ralf said: In /etc/default/grub find GRUB_CMDLINE_LINUX and add iommu=pt intel_iommu=on
WARNING: I think this should be iommu=pt amd_iommu=on for AMD, but I've not tried it.
Oh, and the other thing... I've used isolcpus=1,2,....,7,9,10,...,15 so that linux can only schedule on cores 0 and 8 (the same physical core). Then in my vcpupin, I explicitly bound vcpu 0 to 1, vcpu 1 to 2, etc... This means that the core will be completely idle except for the VM process. If you don't do that, it could up end up competing with all the normal kernel things on the host. Remember that the linux scheduler will always try to put tasks on the lowest numbered cores it can.
Webhosting24 BlackFriday 2021 Special KVM NVMe VPS, upgraded from 10x10x10, in Munich, $13.52/year.
Compared to previous benchmark from 2021-April, both CPU score and I/O score decreased.
Comments
Indeed, I'd missed noticing it too. Unfortunately now gone; maybe back at New Years?
I have an older deal
$84/y
2 CPU core
2.5 TB storage
2.5 GB RAM
6 TB bandwidth
This new deal would have been saving me nearly $50 a year 👀🤦
1 CPU core
2 TB storage
1 GB RAM
10 TB bandwidth
$50 per year
$110 per 3 years (doubled RAM and bandwidth for three year payments)
@ralf @tarasis You may still be able to get it (quote from OGF HH BF thread)
Contribute your idling VPS/dedi (link), Android (link) or iOS (link) devices to medical research
nice testing and writeup!
depending on how you deploy your VMs and especially if you're using qcow2 images, it might be worth it to play around with preallocation and the cluster_size. I usually set the latter to something like 512k or 1M which can help quite a bit with IO.
though this all is much more noticable on spinning rust setups compared to ssd/nvme nowadays.
AWS lighstail, 5$ p.m., last benchmark before final shutdown after rockstable 3+ years of uptime
>Geekbench 5 Benchmark Test:
Test | Value
|
Single Core | 1441
Multi Core | 7982
Full Test | https://browser.geekbench.com/v5/cpu/18922424
and now the VM:
>Geekbench 5 Benchmark Test:
Test | Value
|
Single Core | 1436
Multi Core | 7633
Full Test | https://browser.geekbench.com/v5/cpu/18967843
Hi @ralf!
Thanks for posting this. Very interesting!
What jumped out at me was the tiny 5 point difference between the single core Geekbench 5 scores, 1441 for the Host and 1436 for the VM. I would have expected the difference to be between 50 and 150. How do you have such a low virtualization penalty?
Hi @Mason!
Thanks for your kind words! In case anybody is interested, here are links to previous posts about the difference between host and VM performance:
When running qemu on Hetzner AX101, how much difference does -cpu host and -enable-kvm make?
results with qemu's default processor (as in the first test, above) but now flying the -enable-kvm flag
Here is a link to your previous comment:
This is why I always request for CPU passthrough
Best wishes from the desert! 🏜️
I hope everyone gets the servers they want!
I'll try another yabs run again when I get some free time, but I wonder if you'd lost performance because you'd not added the necessary things to the grub command line.
In
/etc/default/grubfindGRUB_CMDLINE_LINUXand addiommu=pt intel_iommu=onWARNING: I think this should be
iommu=pt amd_iommu=onfor AMD, but I've not tried it.Oh, and the other thing... I've used
isolcpus=1,2,....,7,9,10,...,15so that linux can only schedule on cores 0 and 8 (the same physical core). Then in my vcpupin, I explicitly bound vcpu 0 to 1, vcpu 1 to 2, etc... This means that the core will be completely idle except for the VM process. If you don't do that, it could up end up competing with all the normal kernel things on the host. Remember that the linux scheduler will always try to put tasks on the lowest numbered cores it can.Thanks for your 🌟 suggestions! 🌟 I will study up, try some stuff, and post something. Thanks again!
I hope everyone gets the servers they want!
Amsterdam - KVM Memory VPS - KVM Beta High
Prior to transfer to @iaecm - https://lowendspirit.com/discussion/4983/post-bf-buyers-remorse-tuesday-vps-giveaway
EDIT: realised that this was after a fresh re-install, so the net speeds might be slightly lower than they should as I didn't change the MTU.
Naranjatech: 1 CPU, 1GB RAM and 16GB NVMe 11,50 €/year
MaxKVM ADX-2 Singapore: Monthly: $2.90
1x vCPU (EPYC 7702p, dedicated)
2GB RAM (ECC DDR4)
60GB SSD (PCIe4 NVMe SSD)
2TB Transfer (Inbound + Outbound, monthly)
10Gbps+ Port Speed (Dual 40G uplink)
https://dttech.top - Personal Site.
Server Factory - Ryzen S (2 dedicated vCores)
Prior to transfer to @contactwajeeh - https://lowendspirit.com/discussion/4983/post-bf-buyers-remorse-tuesday-vps-giveaway
I'm a bit sad to let this one go, it's been great!
HostCram KVM-3C - $21 Quarterly (same as BF, but slightly higher price and bigger drive)
Prior to transfer to @webkesh - https://lowendspirit.com/discussion/4983/post-bf-buyers-remorse-tuesday-vps-giveaway
Another fine VPS, amazing performance for the price.
Old HH year end deal 2 vCPU, 2GB Ram, 2.5TB disk, 6TB BW $84/y
New HH BF 2TB, $110/3y. 1vCPU, 2GB Ram, 2TB disk, 20TB
Much better net connection, slower disk (although could be hit with others testing), and slower core but not by a lot. $84/y vs $36/y
Psychz Network (Copy from MJJ-loc)
Notice the Disk IO
Ontario Dildo Inspector
HostHatch Sweden
Hi @ralf!
The iommu/VT-d already might be enabled on Darkstar because I can see:
I have to look more at the scheduling and run more tests.
Thanks again for your helpful comments!
Tom
I hope everyone gets the servers they want!
My new firewall - will probably run VMs on it too given that its way overspec'd for pure firewall duty (16gb / fast 1tb nvme)
Spaceberg VDS Nasty - 1 (Stack)
Our Sweety - 1 package. HDD, NVMe cached
Spaceberg.cc - Your favorite Seedbox provider!
greencloudvps
SG Storage
after migration to new DC
I bench YABS 24/7/365 unless it's a leap year.
spaceberg
france
I bench YABS 24/7/365 unless it's a leap year.
greencloudvps
San Jose
$40/3 years (double cpu during pre order)
I bench YABS 24/7/365 unless it's a leap year.
Webhosting24 BlackFriday 2021 Special KVM NVMe VPS, upgraded from 10x10x10, in Munich, $13.52/year.
Compared to previous benchmark from 2021-April, both CPU score and I/O score decreased.
Accepting submissions for IPv6 less than /64 Hall of Incompetence.
holy shit, what a beast
whats wrong with formating. does not work for me!
``
woot@localhost:~# curl -sL yabs.sh | bash
## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ##
Yet-Another-Bench-Script
v2022-12-04
https://github.com/masonr/yet-another-bench-script
## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ##
Thu 08 Dec 2022 06:13:15 AM PST
Basic System Information:
Uptime : 17 days, 5 hours, 15 minutes
Processor : AMD Ryzen 9 5900X 12-Core Processor
CPU cores : 1 @ 3693.060 MHz
AES-NI : ✔ Enabled
VM-x/AMD-V : ✔ Enabled
RAM : 976.4 MiB
Swap : 0.0 KiB
Disk : 270.5 GiB
Distro : Debian GNU/Linux 11 (bullseye)
Kernel : 5.10.0-11-amd64
fio Disk Speed Tests (Mixed R/W 50/50):
iperf3 Network Speed Tests (IPv4):
Geekbench 5 Benchmark Test:
Test | Value
|
Single Core | 1639
Multi Core | 1620
``
Wow
What provider?
add one more ` .. i do it with 3 ```
ServerStatus , slackvpn <-- openVPN auto install script for Slackware 15
Another Post-CM InceptionHosting buy.. this time in Netherlands. Great little box
ServerStatus , slackvpn <-- openVPN auto install script for Slackware 15
holy shit, what a beast
Nice. Which provider? Plan?
buyvm
Premium