Skip to content
Register

Johannesburg box for African CDN node—latency still brutal?

VPS Hosting by HeadRich 25 replies 2.2K views
HeadRich

Member OP

128GB or nothing

#1

Been testing JNB vs EU locations for edge caching. Results are... not what marketing claims.

Test setup: identical 4GB RAM, 2-core, NVMe boxes. Synthetic from Cape Town, Lagos, Nairobi.

  • Hetzner JNB: 185ms CT, 245ms Lagos, 290ms NBO
  • RackNerd FRA: 165ms CT, 280ms Lagos, 310ms NBO
  • CloudCone LHR: 175ms CT, 295ms Lagos, 325ms NBO

But here's the kicker: KnownHost Cairo beats all of them. 140ms CT, 190ms Lagos, 220ms NBO. Not even marketed as African.

So what does "African hosting" even mean if EU-labeled beats JNB and Cairo isn't in the conversation? But how much RAM? These boxes are all 4GB. My home lab has 512GB. This is painful.

SU

Member

#2

HeadRich your testing is amazing 😊 but the EU locations are lying to us they just mean "we peer with Seacom" not actually optimized JNB has power problems too my friend there loses servers monthly

uma

Member

99.99% or bust

#3

I've monitored three "Africa-optimized" providers for 18 months. Their status pages show 99.9% uptime, but my external probes from Nairobi see 12% packet loss during European evening peaks. The status pages don't capture cross-continental reality. Alert fatigue from false "up" signals is real.

436 days. reboot is surrender.
prague1983

Member

btw I use Arch

#4

Just use Cairo skill issue

Seriously though, "Africa-optimized" means "we bought cheap transit to Seacom and put it on the landing page." The actual routing often backhauls through London because someone didn't update their BGP communities. Just use actual African infra or stop pretending.

oops: 0000 [#1] SMP
steveipw

Member

#5

KnownHost uses Samsung PM1733, the JNB options I tested had recycled Intel S4510 with 80% wear.

Specs from my audit:

  • Hetzner JNB: 2x S4510 960GB, RAID-1, no hot spare
  • RackNerd FRA: 2x unknown datacenter pull, SMART errors
  • KnownHost CAI: 2x PM1733 1.92TB, RAID-10, ZFS

The JNB auction servers are clearly ex-mining rigs. Pass.

zfs send | zfs receive. repeat.
olespete

Member

Trust No One

#6

Which kernel on the KnownHost box, mainline or their custom patchset

airgapped, encrypted, faraday'd, still worried
steveipw

Member

#7
olespete said:
Which kernel on the KnownHost box, mainline or their custom patchset

Custom. Based on 5.15 with their own ZFS modules baked in. I asked support, got a one-liner "we maintain stability patches." No source, no changelog.

The SMART on those S4510s in JNB was grim. 38,000 power-on hours, 82% wear. Hetzner's "new" auction tier is just refurbished with fresh labels.

zfs send | zfs receive. repeat.
HeadRich

Member OP

128GB or nothing

#8
steveipw said:
38,000 power-on hours

That's not even a good scam. I get better hardware from the e-waste pile behind Al Aweer. At least those don't pretend to be enterprise.

My compile boxes in DXB run 5950X with 128GB and I still swap on Chromium. 4GB for a CDN node is insulting to the workload.

frankfurt_ops

Member

8 #9
uma said:
12% packet loss during European evening peaks

Same pattern from Frankfurt. I monitor 200+ endpoints for a logistics firm. The "Africa-optimized" routes all converge on Seacom at Marseille or Djibouti, then you're fighting for transit capacity with every streaming service doing regional expansion. The landing page says "direct." The traceroute says "lol."

prague1983

Member

btw I use Arch

#10
frankfurt_ops said:
Just use Cairo skill issue

I stand by it. Cairo has TE North, Telecom Egypt, direct to SEACOM and AFR-IX. The reason it works is nobody in marketing knows where it is. No demand spike, no oversubscription.

Also: KnownHost has no Cairo datacenter. Steveipw where are you getting CAI from?

oops: 0000 [#1] SMP

Post a reply

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

Post reply Preview Save draft