Spoiling Hetzner - A very small review


As well as being tired of NetCups’ billing practices, their occasional ~1min/week network blips are becoming annoying.

I think I’ll be spinning up some new Hetzners to start migrating my stuff over, even if I am stuck with NC for ~6 more months. I just don’t have the motivation to do anything on my personal projects anymore- dealing with everyone else’s just takes it right out of ya.


Already have 4 VM’s running @ Hetzner. <.<


I can go find 15 miners for you with 100% success rate. Guarantee Hetzner has a solid abuse team that knows how to spot subtle trends. Example: Signs up, spins up 100 of the most expensive servers, maxes out CPU on all within 30m, all named “miner[1…100]” :stuck_out_tongue:

Lots of ways you can identify trends far beyond reasonable doubt. That’s why they get paid the big bucks.


I’m gonna rename all my miners to “idler[1…100]” and all my idlers “miner[1…100]”. That’ll throw them off my scent! :stuck_out_tongue:


Oh yeah but that’s not how you do it on scale… I mean, that’s just a pattern, doesn’t work with all. :stuck_out_tongue:


On the other side of the spectrum, there are certainly plenty which go for the cheapest OVZ plans they can Google, and try to beat the shit out of it, too.


I’ll take that bet, you know I work at scale :wink:


We use Hetzner dedis for machine learning and its derivative anti-spam filters. Very good performance for the price, better than the competition in that regard.



I did some tests from my CX11 in Nuremberg to my storage box (BX10) in Falkenstein (connected through sshfs):

root@nur ~/storage # dd if=/dev/zero of=test bs=1M count=1024 conv=fdatasync
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 42.3316 s, 25.4 MB/s

root@nur ~/storage # fio --name=randwrite --ioengine=libaio --iodepth=1 --rw=randwrite --bs=4k --direct=0 --size=256M --numjob s=8 --runtime=60 --group_reporting
randwrite: (g=0): rw=randwrite, bs=4K-4K/4K-4K/4K-4K, ioengine=libaio, iodepth=1

Starting 8 processes
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
randwrite: Laying out IO file(s) (1 file(s) / 256MB)
Jobs: 8 (f=8): [w(8)] [100.0% done] [0KB/8152KB/0KB /s] [0/2038/0 iops] [eta 00m:00s]
randwrite: (groupid=0, jobs=8): err= 0: pid=1757: Fri Feb 16 11:08:35 2018
write: io=1257.4MB, bw=21453KB/s, iops=5363, runt= 60002msec
slat (usec): min=12, max=852601, avg=1486.17, stdev=10559.55
clat (usec): min=1, max=403, avg= 2.33, stdev= 2.76
lat (usec): min=13, max=852604, avg=1488.51, stdev=10559.71
clat percentiles (usec):
| 1.00th=[ 1], 5.00th=[ 1], 10.00th=[ 1], 20.00th=[ 2],
| 30.00th=[ 2], 40.00th=[ 2], 50.00th=[ 2], 60.00th=[ 2],
| 70.00th=[ 2], 80.00th=[ 3], 90.00th=[ 3], 95.00th=[ 4],
| 99.00th=[ 13], 99.50th=[ 17], 99.90th=[ 34], 99.95th=[ 53],
| 99.99th=[ 88]
lat (usec) : 2=18.53%, 4=75.29%, 10=4.58%, 20=1.24%, 50=0.30%
lat (usec) : 100=0.05%, 250=0.01%, 500=0.01%
cpu : usr=0.35%, sys=0.72%, ctx=321827, majf=0, minf=55
IO depths : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
issued : total=r=0/w=321802/d=0, short=r=0/w=0/d=0, drop=r=0/w=0/d=0
latency : target=0, window=0, percentile=100.00%, depth=1
Run status group 0 (all jobs):
WRITE: io=1257.4MB, aggrb=21452KB/s, minb=21452KB/s, maxb=21452KB/s, mint=60002msec, maxt=60002msec

If you want I can run other tests as well.

My primary use for that storage box is borg and I’m happy with it


Seems like they kicked off a bunch of miners recently, CPU performance has increased by ~25-35% on my cloud instances in the last couple of days.


I wish Hetzner offered US locations. Not very nice, and extremely prejudiced I feel. Also, it seems a bit unfair to only keep their service allocated to one region. All VPSs are created equal.

With that said, this is a very informative review, thanks!


Oh be nice :stuck_out_tongue:


Good to know. I know our team is working hard to make sure everyone can enjoy their cloud. --Katie


We’re a privately owned company. There’s some pros and cons to that. The disadvantage is not having crazy amounts of cash at hand to expand overnight. The advantage is that we are not beholden to anyone who wants to see a return on their investment.
And this means that we can focus on quality. We’re more of a quality over quantity company. Our first DC in Finland is nearing completion, and if things there go well, I believe we’ll be looking at other sites. --Katie, Marketing


Gotta love your replies and honesty. Keep it going! :grin:


Thank you, Katie. I was just ranting because I wish your servers were in the US. Cause I see a lot of great things about Hetzner all over. The big LET thread, this forum, and other boards.


It’s nice to know that there is a demand from North America. It’s always nice to feel wanted. I’m make sure to pass your comments on to some of my colleagues. --Katie


[quote=“sshd, post:29, topic:86, full:true”]
I did some tests from my CX11 in Nuremberg to my storage box (BX10) in Falkenstein (connected through sshfs)…[/quote]

Fwiw I tried something like that a while back and found I got around 2x the speed with plain scp as with sshfs mount, so I decided to stick with scp.

Meanwhile I finally ran a “production” task on a 32gb cloud server last night, a monthly archive compression thing that took 258 wall clock minutes on my i7-3770, or 1141 user minutes (4 tasks, about 440% utilization because of hyperthreading). Most of this time was spent in bzip2 and xz. On the cloud instance it took 200 wall clock minutes or 1529 user minutes with 8 tasks (load average stayed in the 12-13 range for most of the run). In both cases there were some idle cycles towards the end (not all cores utilized) as the last pieces came straggling in. But 200 wall clock minutes vs 258 is near perfect scaling of 8 cores@2.2 ghz vs 4 cores at 3.4 ghz.

I’m not ready to cancel my dedi yet (I like the 2x 3TB storage space among other things) but these cloud instances have plenty of cpu oomph! I spun up the cloud server, copied some files around and did the compression step, then deleted the server after around 3.8 hours of uptime, so I spent 0.2 euro on the whole operation. It makes me wonder why I’m spending > 100x that much on the usually-idle dedi every month.


I am completely addicted on Hetzner. Seriously. I asked Katie if they could add netboot.xyz to their ISOs list. Hopefully they do. <3

Also got a Windows VM running there, Windows Server 2016, 1 Core, 2GB RAM and its pretty fast.


@Hetzner_OLMay I start pitching Montreal, Quebec, Canada or Dallas Texas US to you in advance?