Skip to content
Register

Six months on RackNerd: not worse, just the same

Reviews by srelet 25 replies 2.2K views
srelet

Member OP

13 #1

Six months ago I promised transparency on our migration from Leaseweb to RackNerd. Here it is.

The pattern is identical. Same memory pressure alerts at 2am. Same "unexpected maintenance" windows that eat our error budget. Same tickets closed with "please clear your cache." Different logo, same playbook.

I ran the numbers. Our blameless postmortem count actually went up 15% post-migration. Transparency demands I say this: RackNerd is not worse. They are the same. The SRE in me wants to scream.

But here's the twist. Our p99 latency dropped 40%. Forty. I traced it for two weeks. Not their network. Not their SSDs. Their "new" Dallas datacenter? Abandoned by KnownHost in 2024, bought cheap, never re-cabled. We're running on ghost infrastructure with zero noisy neighbors. That's your improvement. Not the host. The emptiness.

Transparency. It's all we have.

ana_mad

Member

#2

Sent ticket yesterday about same thing! They told me "premium routing" haha. Made test order on HostHatch (https://hosthatch.com) just to compare, actually faster in Barcelona. But RackNerd price is hard to beat!

Your transparency is super appreciated, honestly. Most people just vanish after migration post. Cheers!

swimming upstream since 2019 🐟
RA

Member

#3

Secret, so you found the secret! The server is alone in datacenter, why pay more? This is actually best deal if you think about it. No neighbors, no problems. But what happens when they fill it up? Then RackNerd becomes same as others, no?

rudi3

Member

sudo rm -rf /

12 #4

Actually the compoundword here is ghostinfrastructure and it is fascinating because the manual states clearly that abandoned datacenter equipment must be refurbished or? But apparently RackNerd found a loophole in the regulations that govern such things and the performance improvement stems from the lack of contention which is actually the same reason why early Virtual private server offerings were so fast before overselling became the industrystandard or?

Joy

Member

#5

Same 2am alerts here, I moved off last month

JO

Member

#6

15% more postmortems but "not worse"? Come on

annexbi

Member

#7

(the "ghostinfrastructure" claim (needs a source (or a regulator)))

push. done. coffee.
srelet

Member OP

#8
john_letra said:
15% more postmortems but "not worse"? Come on

Fair. What I mean: the *category* of failure didn't change. We're not seeing new failure modes. It's still memory pressure, still maintenance windows, still cache excuses. The 15% is volume, not severity. If anything, the failures are more predictable now, which is its own kind of awful.

The p99 drop is real though. I have two weeks of mtr and fio. Want me to dump the methodology?

rudi3

Member

sudo rm -rf /

#9
srelet said:
The p99 drop is real though

But this is exactly the ghostinfrastructure effect or? KnownHost abandoned Dallas because they needed newer hardware for their managed vps customers and the power density was too low for modern GPUs or? RackNerd does not sell GPU instances so they do not need to upgrade the power infrastructure immediately and the lack of neighbors means no noisy neighbor problems which is the real reason for your latency improvement not routing not SSDs or?

pam3

Member

Latency Witch

4 #10
srelet said:
Want me to dump the methodology?

Please do. I'm in Melbourne, Dallas is literally the other side of the planet for me. My RackNerd p99 to Sydney is worse than my HostHatch p99 to Sydney, but my RackNerd price is half. Always wondered if I should bother with the Dallas location or if it's just physics.

100% packet loss at hop 3

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft