PetraSuper
Member
OP
network boot believer
- Joined:
- Jun 2024
- Posts:
- 127
- From:
- Bratislava, Slovakia
I am run test on server. All synthetic. No lie, but no truth also. Hej?
I am collect data from forum. 200 people send result. We make application benchmark. Compile LLVM repeatedly. Surprise: this is best predictor for real workload. No articles needed, numbers speak.
Malý server with good YABS can compile slow. Malý server with bad Geekbench can compile fast. Why? Cache. Memory bandwidth. Not in synthetic.
I am post spreadsheet. Into config you put your result. We see who has real speed.
boot anything, anywhere, anytime
moebig
Member
RAID is not backup
- Joined:
- Jul 2024
- Posts:
- 115
- From:
- Casablanca, Morocco
Wesh PetraSuper
Le server حلو but benchmark is haram incha'allah. I am see this before. Synthetic lie, real world cry.
Your LLVM test : good idea. But incha'allah uptime is better metric. Who cares speed if down ?
I am run test on GreenCloudVPS (https://greencloudvps.com) and CloudCone. GreenCloudVPS : YABS good, LLVM bad. CloudCone : YABS bad, LLVM good. Le provider knows ? Probably not. They tune for bench, not for build.
French spacing is important ! Never forget.
RAID 1: because paranoia pays
olespete
Member
Trust No One
- Joined:
- Jun 2024
- Posts:
- 270
- From:
- Unknown
200 people sending WHO KNOWS WHAT from WHO KNOWS WHERE. This is a SECURITY NIGHTMARE.
What could go wrong:
- Fake results injected by providers themselves
- Malicious spreadsheet macros (are you sanitizing inputs?)
- Correlation attacks linking forum identity to server performance
- Supply chain compromise in LLVM source
I recommend fail2ban on whatever server hosts this. Firewall everything. Validate submissions against known-good IP ranges. CHECK THE SIGNATURES.
For what it's worth, the benchmark is fine. The trust model is BROKEN!
airgapped, encrypted, faraday'd, still worried
mediaaustin
Member
Deliverability Nerd
- Joined:
- Jul 2024
- Posts:
- 276
- From:
- Austin, US
You're all measuring the wrong thing.
For what it's worth, I've spent fifteen years on deliverability. The only benchmark that matters for a VPS is: can it send email that lands in the inbox? Not spam folder. Inbox.
SPF alignment, DKIM signing, DMARC policy — these need consistent IP reputation more than they need LLVM compile speed. A "fast" server on a UCEPROTECT-listed /24 is worthless. A "slow" server on clean warm IP? Gold.
The synthetic vs. Real-world debate misses the point. Application-specific means YOUR application. If you're hosting mail, test mail. If you're compiling, test compile. If you're running WordPress, test WordPress.
Correcting a misconception: YABS disk tests with fio are actually decent predictors for I/O-bound mail queues. But only if you run them with `fsync=1`, which nobody does. I used https://github.com/masonr/yet-another-bench-script for this.
SPF, DKIM, DMARC — holy trinity ✉️
garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
Llvm compile on malý server? What cpu, what ram
CUDA cores are my love language
PetraSuper
Member
OP
network boot believer
- Joined:
- Jun 2024
- Posts:
- 127
- From:
- Bratislava, Slovakia
Le provider knows ? Probably not. They tune for bench, not for build.
This is what I am suspect. But how to prove? We need more data.
I am add column to spreadsheet: provider, location, YABS disk, YABS network, LLVM time. If pattern show same provider always YABS good LLVM bad, we have evidence. Not proof, but evidence.
Also I am add kernel version. And `lscpu` output. Too many variable.
Good catch. I am use `nproc` to detect. But if provider throttle, this hurt. Maybe fixed `-j2` for all? Slower but fair?
boot anything, anywhere, anytime
