Skip to content
Register

CloudCone HDD storage VPS beats SSD on sustained sequential writes

VPS Hosting by Emre10 17 replies 5.1K views
Emre10

Member OP

#1

So I run this backup script since 2 week on my CloudCone storage VPS. 4TB HDD plan. Friend say I stupid for not getting SSD. But my sequential write test with `fio` show HDD doing 180MB/s sustained, his SSD plan from same Provider drop to 40MB/s after cache full.

I test with 100GB file. No joke. His "SSD" is DRAM cache + TLC, cache exhaust and then garbage collection kill performance.

This Important for backup workload. Sustained matter more than burst. Anyone else see this?

https://cloudcone.com

HieuTce

Member

#2

Yeah this is a thing and it keeps coming up and people keep being surprised and I keep explaining it and here we go again so the thing is that modern consumer and even prosumer ssd architectures are fundamentally optimized for random access patterns and low queue depth responsiveness and they achieve this through aggressive caching strategies and overprovisioning and sophisticated firmware that treats sequential writes as essentially random at the flash translation layer anyway and the result is that when you push sustained sequential throughput beyond the size of the dram cache and the slc cache tier you end up with direct-to-tlc or direct-to-qlc speeds which are often abysmal and meanwhile your hard disk drive is just spinning and seeking in orderly fashion and the sequential pattern is exactly what the drive was born to do and the firmware is not getting in the way and the throughput

zoekyo

Member

Less Is More

3 #3

Simple is better, but
The disk that matches your pattern wins

rm -rf / --no-preserve-root ☯
marchhopper

Member

3 #4

The phenomenon is well understood. The SSD cache exhaustion creates a cliff in performance curves. For backup machines, sustained throughput outweighs peak figures. This is not controversial among those who operate storage infrastructure at scale. The HDD achieves its rated speed through mechanical predictability. The SSD must manage erase blocks and translation tables. Both have valid roles. Neither is universally superior.

-- Marcus

LA

Member

#5

What fio job file did you use

sarahjey

Member

#6

Same provider but my ssd plan chokes on big tarballs too

learning on $5 VPS and prayers
IC

Member

#7

What fio flags you used, direct=1 or cached?

DmitriEng

Member

#8

I had same thing on CloudCone, ssd plan tanked after 60GB

larryjeong

Member

#9

This is why I stopped falling for the "SSD storage VPS" marketing on backup boxes

DmitriEng said:
I had same thing on CloudCone, ssd plan tanked after 60GB

60GB is a hilariously small window for a "storage" plan. That's not even a 4K Blu-ray

HDDs don't lie about their speed. What you see after ten seconds is what you get after ten hours. For rsync and tar.gz dumps that's honestly the better deal

I keep a HostHatch HDD box for exactly this. Never once missed an SSD for that job

$3/year. 128MB RAM. Pure happiness.
zoekyo

Member

Less Is More

#10
DmitriEng said:
I had same thing on CloudCone, ssd plan tanked after 60GB

60GB is a generous cache at least

My Leaseweb HDD box sustains 150MB/s for the whole 4TB
No surprises, no cliff

For bulk storage the boring choice wins

rm -rf / --no-preserve-root ☯

Post a reply

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

Post reply Preview Save draft