The cheapest solution for hosting?


(ljpp) #40

Benchmark results, or did not happen :wink:


(James Mc Mahon) #41

There used to be a site called serverbear which gave a great comparison of performance/dollar. You ran a shell benchmark they provided, it uploaded the results to their website.


(ljpp) #42

Vultr is matching your initial deposit dollar to dollar for new customers. So if you deposit 100$, they’ll give you another 100$ which is effectively a 50% discount for a limited perioid of time.

In practice, for 100$ you get 20 months of 10$ hosting, 10 months of 20$ plan and so forth. Not bad.


(Rafael dos Santos Silva) #44

One server from our hosting (web type):

CPU model:  Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz
Number of cores: 8
CPU frequency:  800.000 MHz
Total amount of RAM: 32118 MB
Total amount of swap: 32767 MB
System uptime:   106 days, 11:36,       
I/O speed:  478 MB/s
Bzip 25MB: 2.41s
Download 100MB file: 31.3MB/s

Almost twice as fast when compared with the best VPS CPU from other benchs.


Benchmarks: VPS vs dedicated/bare metal?
(ljpp) #45

Hm…if I ever need big resources again, it’s time to re-think about a dedicated server. I just don’t like hardware failures.

For those looking for a bargain, I moved one instance to Hetzner (4.80€) just for kicks, and the initial experience is very good. But there’s a separate thread for that.


(Zsolt Ero) #46

I just wanted to share some budget hosting experiences.

1. Scaleway VC1S - 2.99 EUR / month, 2 GB

  • Works reliably, but very slow for rebuilt. For low trafic website it is ok.
  • Scaleway uses custom kernels in Ubuntu, and you have to select a docker compatible custom kernel in the GUI to let Docker install on Ubuntu 16.04 without errors.
  • Huge SSD for 2.99 EUR, 50 GB!
  • It is actually so slow that templates/postgres.template.yml had to be manually modified from sleep 5 to sleep 60 for postgres.

Rebuild time: 17m 14s

2. OVH VPS SSD 1 - 2.99 EUR / month, 2 GB

  • Amazing performance for 3 EUR / month. Actually seriously powerful!
  • North American and European locations as well.
  • Uses stock Ubuntu kernels, docker can be installed out of the box.
  • OVH account creation is a mess. I mean seriously a mess, it took me 2 weeks of emailing and support tickets just to open an account. You also need photo of passport / proof of address as part of registration.
  • Admin GUI is a mess. Feels like government websites built with PHP 10 years ago.

Rebuild time: 5m 45s

Here is the benchmark I used:

./launcher rebuild app  # cache docker image 
mv shared/standalone{,.bak}   # make sure you run it on a test server
time ./launcher rebuild app   # time the second time

Where app is a standard/stock config.

BTW, what were your benchmarks from? Still, I strongly recommend benchmarking via a rebuild, VPS CPU “cores” are a joke and short syntetic benchmarks are usually not reflecting the real world performance on a VPS. A longer rebuild can give a true image.


Upgrade failed due to unclean database shutdown
(ljpp) #47

OVH’s VPS SSD series have serious oomph for the price, but the SSD 1 particularly is limited by it’s 10GB disk space. One will run out of space very soon, if the community gains any content or traction. Personally used it for a few months but then had to move.

But yes, performance is very good.


(Zsolt Ero) #48

Shared the problems I’ve experienced on a Scaleway and the needed fixes here:


(ljpp) #49

I am not sure whether its ok to promote posts written elsewhere, but I summarized my experiences with the lowest price category VPSs:

LeaseWeb is a new experiment for me, and I will report back once I have long term experience.


#50

What’s the command you use for these informations ?

I can add infos on OVH, 1&1 Cloud (french) and probably on Hetzner CX10 very soon


(Jeff Atwood) #51

Hmm, I found a place that will rack a Scooter Computer for $13/month, including public IP, 3TB xfer on 100mbps port…? That sounds… excellent?

Considering I can get crazy performance – Kaby Lake 2 core / 4 thread i7, 16GB RAM, 250GB / 500GB SSD – for a one-time investment of around $400-$500 that seems like a pretty good plan??


(James Mc Mahon) #52

This, so much this. They have to be losing business with how poor their signup\account creation form is. Two issues I came across in December

Phone number not accepted
Solution: The correct way to input your phone number is +[COUNTRY CODE][PHONE NUMBER] without spaces, parenthesis, dashes or brackets. For example, OVH support office’s own phone number is +18556845463.

“subscribe_minimum_18_years_old” message
Workaround: Leave the year field blank.

Incidentally I’m using two interserver VPS “slices” for a total of 2GB RAM which comes out at $13/mo including a fixed IP. Storage is 50GB but that doesn’t matter to me because I’ll likely use S3 for object storage at some point.


(ljpp) #53

Hetzner CX10: Hetzner experiences
I am pretty sure OVH benchmarks are somewhere here too (they are good).

And the benchmark tool is VPSBench.


(Zsolt Ero) #54

@ljpp, VPSBench is the most useless benchmark ever. Literally no line means anything useful in that benchmark.

Let’s go through it:

  • CPU model: symbolic, shows absolutely nothing about how many users are sharing a CPU
  • Number of cores: see above, a single dedicated high frequency core could easily destroy a 2-4-6 “core” VPS
  • I/O speed, aka the sequential read dd test. There are dozens of decent articles about why this test is absolutely meaningless to real world performance.
  • Bzip: so quick that I could get higher value on a 1 GBP / month VPS than on my i5 dedicated server.
  • Download: this is actually testing cachefly’s CDN in your server’s region, not your server’s network connection

If you want a real world test, benchmark rebuilding the same Discourse version on them, from a clean folder. That’s actually by far the best real world test I could come up with.


(ljpp) #55

@zsero

Bullets 1-5: Yes, I know

If you would have actually read the article, you would know that it is based on running live production sites. The VPS bench figures are there just for show. Running Discourse in production involves builds and re-builds. Discourse build times correlate ~1:1 with the CPU benchmark results.


(Zsolt Ero) #56

No, they absolutely don’t. Scaleway VC1 range has quite good CPU benchmarks, and a fast SSD as well. Yet, Discourse build times are abysmal.


(ljpp) #57

Well, Scaleway scores around 35-40% slower Bzip test results in VPSBench, in comparison to DigitalOcean. Disk I/O is also on the slow side - I beleive their “local SSDs” are not directly plugged to the server hardware.

But agreed, it is definitely slowest of the pack.


(Zsolt Ero) #58

OK, a couple of examples:

  • Scaleway VC1L (8 GB RAM, 6 cores) does 9000 iops and finishes sysbench in 20 seconds -> builds my docker image in 17:40.
  • iwstack 2 GB instance, (4 cores) SAN HDD, 3290 iops, sysbench in 23.07 -> builds docker image in 08:23

As you can see, Scaleway has 3x IO performance and bit better CPU, yet iwstack is 2.1x faster.

Here is my benchmark I’ve run on a couple of servers:
42 server’s results

apt-get update && apt-get install -y sysbench ioping ;\
ioping -RD -w 10 . | grep iops ;\
sysbench --test=cpu --cpu-max-prime=40000 --num-threads=8 run | grep "total time:"

(Steven-Anthony Winderlich) #59

In Germany there is a VPS Hoster named MyVirtualServer with Prepaid model.

Myvirtualserver.com is their site.

Currfntly ran by 2 people next to full time job and or school and family. :slight_smile:

40 bucks a month for 16 GB RAM 8 Cores and 800 GB Harddrive space is pretty good i guess.

800 GBit DDoS Protection and 3 snapshots included for free as well. 10 GBit shared Network as well.


(Jeff Atwood) #60

I question the value of --num-threads=8 because what you need to show is single-threaded performance. The number of cores is a different variable that should be shown in a different column. Also what is the sysbench^-1 column exactly?

I think the reason you are seeing Discourse rebuild results inconsistent with your data is because you are not isolating single-thread performance and that matters intensely for Discourse.