Nice write up, especially that as predicted your final results do not differ much from @arrogantrabbit reference setup. So just a couple of things.
It would indeed enhance your write up to include details on CPU architectures, memory timings, motherboard specifications, as well as information on NFS and … Kubernetes versions. These additions would provide a more comprehensive understanding of the system configuration and environment, especially … the Kubernetes version info.
I noticed that you are being very brave going straight against the Rabbit’s Lambo in you VW without a helmet [:- )] so let me ask you a question. What do you exactly mean by the reference to the LawrenceSystems? And what exactly is you position on NFS vs Object Storage? Do you agree with the recent opinion expressed by the CEO of Hammerspace at The HPCwire pages? [link]
[slightly off topic] Ah, just wanted to add that I do not think that’s a valid business model long term, looks more like a charity, but I might be wrong.
I don’t think this is of much use, since these are none important factors in my opinion. Again, this is a comparison between different ZFS pools, not node hardware/software guide.
Anyway, here you go: x64, 4 vCPU of a Intel(R) Xeon(R) E-2276G,
DDR4-2666 ECC, NFS 4.2, Debian 12 with Docker 25.0.3.
I don’t get it. @arrogantrabbit and I am not fighting if that is what you try to imply.
Lawrence did some YT videos on how SCALE performs worse than CORE. He did this again a year later and found the differences to be mostly gone. I am too lazy to look it up and I also don’t really care to be honest. Again, this is not a how-to guide but a comparison between ZFS pools.
This is both off topic and also meaningless. You could ask me what my position on birds and waiting in line at the groceries store is. A: my opinion on it does not matter and B: this is an Apples to Oranges comparison. These are not the same thing.
What opinion? He stated multiple opinions
And I am no expert on AI.
Let phrase it like this:
With the knowledge that “Cloud” turned out to be way more expensive then originally expected and now have specialist to “uncloud” companies to save millions of AWS Dollars every year just be installing a rack, and all AI companies I know have a bunch of 4090 installed instead of renting compute, am I surprised that for AI where you need fast and huge amounts of storage, that this can be better achieved by a cheap 100GBit TrueNAS than trying to connect to some Object Storage with my 25Gbit fiber connection? Not really
But again, this is not business analyst insight but a comparison between ZFS pools.
Yeah, that is because you guys don’t understand the downsides of STORJ and what the USP of STORJ is.
Thinking you can compete with Backblaze by buying hardware is not a valid business model and looks more like charity from you to STORJ.
But again, this is not a business model analysis but a comparison between ZFS pools.
That is also why I am going to try to flag our comments off topic, in fairness to other members of the forum.
Regarding the results not differing from the Rabbit’s referential ZFS setup thoroughly described in this forum over recent months, you may refer to your very own explicit thread on this topic, not just the summary here. As for the opinion of the CEO of Hammerspace, I was referring to the main opinion of this article that was underlined by the topic. As for the business model, I was speaking for myself, and as of now, I still stand by my opinion that long-term, this is probably not a valid business model and that capital expenditures (CapEx) should be included into the equation on the side of the storagenode operators. EDIT: Thanks for providing hardware specification and software versions.
Again, link please. Not sure what you are talking about.
Why not form your own opinion and verbalize it then?
I still don’t know what point you exactly wanna discuss.
To me his main point was selling his product.
What is not a valid business model? Using unused resources? Ok. Maybe. We will see. But that does not make “buying hardware and compete with backblaze” a valid business model.
Just because A might be wrong, does not B automatically right.
That would be great for SNO. There is just one problem. STORJ is already loosing money with the current price structure and just surviving on ICO runaway money. Now you want to add additional costs. If STORJ wants to survive longterm, they have to find a price model that creates a profit margin. That is why additional price cuts are not a question of if but when.
This isn’t true. Our runway expands each month from ongoing sales. The company has significant reserves unrelated to the tokens. Even if all the tokens are spent, the company has several years of runway from its existing funds, not including any new and ongoing sales.
The company continues to see positive growth. We are on an upward trajectory.
