All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Upgrade dedicated server from Online to Hetzner, yes or no?
DrCornFlakes
Member
I'm undecided if I should upgrade my current server, or not.
Current server: https://documentation.online.net/en/dedicated-server/offers/pro/pro-1-s-15k & https://support.hpe.com/hpsc/doc/public/display?docId=emr_na-c02793595
Upgrade server: https://www.hetzner.com/dedicated-rootserver/ex51-ssd
Geekbench comparison: https://browser.geekbench.com/v4/cpu/compare/9139125?baseline=7730556
Use: LAMP to host a single site with some traffic.
Currently, it doesn't have any significant performance issues, however, top shows MySQL process always rounding 200-300% CPU usage with 2~3 load average.
Had a few 500 apache timeouts that were solved with some tune up.
Some logs still show some slow query/access sometimes (>5s).
Reasons to upgrade:
- better CPU performance, that I hope will decrease momentary slowness episodes
- current promotion with no setup fee
- a more "future proof" investment
Reasons to not upgrade:
- price month 19,99€ vs 44€
- SAS 15k vs SSD reliability, totally unaware of which is more reliable
- HP hardware with hardware RAID vs ??? hardware with software RAID
- server spec vs desktop spec, ECC RAM, Xeon, SAS/SDD
What do you think? Am I forgetting something that I should be considering?
Thanks for your input. 
- What would you do?27 votes
- Keep current server25.93%
- Upgrade it74.07%
Comments
Fix your software issues before thinking cpu is too slow.
Online server has ECC and maybe better network, Hetzner has more hardware per dollar.
If you're running real services on the old server and are srsly thinking of moving, I'd get the Hetzner server and run tests (particularly of network) before letting go of the old one. You could start with one of their cloud servers on hourly rental to see if the network suits you.
Sounds like, you have some software issues and should optimize it.
But a change from online.net to Hetzner is a good choice anyway.
Just enable logging for slow sql queries and take a look at that.
Since it's only 1 website, going with a LEMP stack, plus caching MySQL with memcached should reduce your load by quite a bit.
Getting a server that performs 70% better, when your CPU is at 200%~300 usage won't solve your issue
In my experience, online has better network to the USA than hetzner, but hetzner has better network inside europe, although the two are close.
I've noticed quite a few more drops on the network side with online.net (multiple downtimes over the course of 6 months using scaleway) than with my hetzner dedi (one time, but that was due to a server in the same rack blowing its psu and tripping a fuse, and was fixed quickly and efficiently after notifying support)
online has crap network to the USA. Has been proven a lot.
Hetzner is even worse. They are good, but network to USA is never a good point for any EU providers.
It ain't that bad with my dedicated:
New York 94.130.207.188 3 98.044 ms 98.151 ms 98.086 ms 0.364 ms 0%
Miami 94.130.207.188 3 120.519 ms 120.811 ms 120.710 ms 0.314 ms 0%
Dallas 94.130.207.188 3 130.138 ms 130.469 ms 130.278 ms 0.139 ms 0%
San Francisco 94.130.207.188 3 164.299 ms 164.329 ms 164.317 ms 0.331 ms 0%
Seattle 94.130.207.188 3 159.140 ms 159.190 ms 159.157 ms 0.326 ms 0%
Ping isn't an issue, show us some sustained transfer speeds. I've found OVH much better than Hetzner for this. Not sure about Online.
I wish Hetzner has DC in NA, North America, being the US and Canada.
Well.. France is closer than Germany
Thank you all.
Country of main audience is relatively close to France or Germany.
Online.net stats say monthly 3Mbps average bandwidth and 700GB traffic.
Ping and traceroute are very similar, difference is negligible.
So, network wise, it isn't that important.
The main problem here is that I don't have enough know-how to fix software...
I've enabled all logs (php-fpm, slow query log, slow php log, etc.), searched for causes and solutions but, there's always a point where I hit a wall.
I've read a lot that most of the time, even if it's not the correct way to do things, that it's cheaper and easier to throw better hardware than to fix "bad" software.
Hence, this discussion.
Thank you all, again.
If you are prepared to throw money at the problem why not throw it at someone to look into it for you? Might pay off in the long run.
No really. That's ok as a temporary measure or if the software has a constant-factor slowdown compared to an optimized version. But in your case it's likely that the slowdown will get worse as your dataset gets larger. So I'd advise continuing to figure out the software issue. Looking at the slow queries, using EXPLAIN to find the full table scans, and adding indexes is usually the first thing to try.
Also, moving to bigger servers hits a wall after not that many expansions. If you're trying to run a very large site, you have to scale horizontally (i.e. distribute the workload over lots of normal sized servers) rather than vertically (upgrade normal sized servers to huge ones).
Performance wise, you'd better disclose what kind of website software you are running. I've run a moderate size forum without any issues on an i3 with millions of pageviews per day.
If your application naturally uses lots of resource (those complicated business logic) then a better server might be the only choice. Otherwise, before you hit the CPU spec wall (yes, there's eventually a wall for you) you may have the chance to reduce the resource cost by fixing the software.