Now you are at 2x switch, 2x target, 2x port, 2x config to drift. Your savings on NVMe went where exactly?
I count parts. More parts, more status pages. More status pages, more 3 AM.
My edge nodes boot local. NVMe is €4.15/month at Hetzner. I sleep.
Member
I break things for a living so here's how I break network boot:
All four happened to me. All four were "theoretically prevented by proper design." Proper design is a vacation I have not taken.
Local NVMe: one less thing.
Following from Auckland. 180ms to anywhere that matters. Network boot is theoretical here.
My NAS boots off a CF card. 47 nodes over the network sounds like science fiction and witchcraft.
Member
CS student perspective: Hetzner auction server with 2x NVMe is €34.60/month. Boot NVMe is not the expensive part. The expensive part is learning iSCSI at 2 AM because your provider had a storage hiccup.
I bought local boot. I bought sleep.
Member
Frankfurt perspective: I track uptime in nine nines because clients demand it. The spreadsheet says network boot costs me 0.03% annual availability from target maintenance windows I do not control. Local boot: my maintenance window, my choice.
€4.15 is rounding error. 0.03% is contract breach.
Member
Sofia, mostly reading. Nuyekigor here.
Ceevul, you are also Sofia. Do you run this? Which provider gives you MLAG in Sofia? I ask because I looked.
Hetzner has no Sofia presence. Neither does Equinix for dedicated at price point. Local options are single-homed or expensive.
If you have found MLAG+redundant target at Bulgarian pricing I would genuinely like to know.
I do not run in Sofia. I run in Hetzner Falkenstein https://www.hetzner.com with MLAG on my own colo cross-connect. My location in profile is where I live, not where my racks live.
The architecture is real. The geography is irrelevant.
Member
Geography is latency is physics is relevant. You said Sofia. People read Sofia. Now you say Falkenstein. Which is it?
Also: own colo cross-connect at Hetzner? They do colo in Falkenstein, yes. But you are not in their colo list price sheet with MLAG. Custom contract? Then your €3/month iSCSI math is even more broken than GeorgeNmp's.
BGP is my love language but I speak accounting too.
This thread has become a game of telephone where everyone argues against their own invention.
Let me restate what I actually run:
I said "proper LACP to your TOR" in my first post. Hetzner's TOR. Their risk, their SLA, their maintenance. I have had one switch failure. Hetzner moved the NIC to redundant fabric. 8 minutes. 4 nodes affected, not 47, because not all on same switch.
The €8/month NVMe savings are real on 47 nodes. €3768/year. I spend it on engineers, not storage.
Database workloads still excluded. Still.
