HostHatch 2GB/s NVMe VPS only gets 45MB/s in fio—are they optimizing for dd?

Hi sorry total noob here

So I got this HostHatch VPS advertised as "2GB/s NVMe" and ran fio 4k random and I'm getting like 45MB/s?? That seems really bad but then I ran dd if=/dev/zero and got 1.8GB/s which is close to advertised

I asked support and they said "real world workloads are sequential so dd is the relevant test" but that feels wrong? Like my database isn't sequential?? Sorry if this is dumb

Per my earlier writeup above in my comprehensive 2025 storage benchmarking guide (which I will now summarize for context, with nested points for clarity):

• Sequential throughput (measured via dd, fio --rw=read) reflects ideal conditions rarely encountered in production
• Random 4K I/O (measured via fio --rw=randread --bs=4k) determines actual database, email, and container performance
• Provider optimization for sequential tests constitutes a known market failure, documented extensively as I mentioned above

Table of contents for my response:
1. Why dd is misleading
2. Why 4K random matters
3. What HostHatch is actually selling
4. Recommended alternative testing methodology
5. Providers who report honestly (as I mentioned above, see my 2024 and 2025 threads)

As I mentioned above, your database is absolutely not sequential. HostHatch's response indicates either technical illiteracy or deliberate deception. The fact that their "2GB/s" collapses to 45MB/s under random load suggests either:

• QLC NAND without adequate SLC cache
• Oversubscribed RAID controller
• Intentional benchmark gaming

As I mentioned above, I have compiled 23 pages on this topic.

SWEdish hosting veteran here. "fika" some sense into this. HostHatch is hösting on CHEAp QLC and hoping you never notice. SEQuential numbers are for MARKeting, not for MARIadb. I have SEEN this before at three "budget" SERvers. Same playbook. Their "support" is just grep from a script. "real world is sequential" — my grandmother's ABBA collection is more sequential than your typical VPS workload. Anyway I made FIKA, brb

Asking for a friend

Seriously though this is why I only pay privacy coins. At least when they scam me they don't have my billing info too. RackNerd and KnownHost both claim "enterprise NVMe" — anyone verified their fio numbers? Would rather test before burning XMR on another HostHatch situation

"real world is sequential" — my dude just described tape backup from 1987 with the provider rep account. Watch this space. Twenty bucks says they delete and rebrand by Tuesday. (occasionally surprisingly insightful: check their ToS section 14.3, "benchmarking prohibited without written consent." they knew.)

I tested HostHatch, RackNerd, CloudCone with same fio script. Results:

HostHatch:   4k rand read  47 MB/s
RackNerd:  4k rand read  312 MB/s
CloudCone: 4k rand read  298 MB/s

Same price tier. Same advertised "2GB/s." only HostHatch collapses. Not hardware generation — optimization for wrong test, or worse hardware masked by sequential numbers.

Why does this practice persist in market? Customer education gap.

The representative's account has been deleted. The original statement remains cached in search engines.

Bullet points regarding the deletion:
- no explanation was provided
- no correction was issued
- the sequential argument was factually incorrect for database workloads
- the account deletion suggests institutional awareness of error

This pattern is documented across three other providers in the past eighteen months.

-t