- Pay for each PUT
- Pay for each byte (ie pay less for small chunks)
Per PUT means paying a fixed storecost regardless if the chunk size is 1 MB or 1KB.
Per byte means paying 10x more for a 1MB chunk compared to a 1KB chunk.
The current proposal (rfc0012 extended by rfc0057) is per PUT.
What are some advantages and disadvantages?
I think that the main difference is that per byte will be better(cheaper) for a lot of people, but per put is better for the network (higher burn rate).
I would not do it on bytes, but a slightly higher unit. On bytes might make the cost unnecessarily small when store cost is lowest
For example 128 bytes, or 256 byte or 512 bytes or 1K etc.
I might suggest 512 byte similar to traditional unix/HD sector size, or 4K that windows/modern HD block uses
EDIT: @mav also there has to be a cost for the work to be done by the network to store the data.
So my suggestion is that there is a component for overheads and a component depending on size.
For a 1MB chunk the store cost might be 0.00001
Now maybe we decide 1/2 the cost is overheads and the rest dependent on size
- 500K stored might be 0.000005 overhead and 0.0000025 for data size.
- 100K stored might be 0.000005 overhead and 0.0000005 for data size.
If store cost of 1MB is 0.01 then 500K cost is 0.005+0.0025 and 100K is 0.005+0.001
I voted PUT on the assumption that there is a fixed cost per PUT.
That’s only an approximation, so my suggestion would be that the unit storage charge be designed to keep the fee per unit as close to the network cost per unit as possible. The aim of this is to be the fairest to users and the network (farmers), and I think also the hardest to game or exploit as an attack.
Yes no matter the size of the PUT chunk there is a certain overhead.
I meant to include that in my post, but its only 3 hours from next year and I’m not exactly thinking of everything.
I think though that some benefit in cost to store for say forum comments of say 200 bytes might be good for those who are not doing big file storage but love tweeting or FB posts.
(I am editing my previous post to suggest a base cost for put + variable amount depending on size)
Yes, something like this sounds good.
Cost based on MegaPut, KiloPut, Put. Done.
Pay for each PUT because it sounds easier as it is current proposal.
In some update it can be changed.
Anyway to be sure 1024 times 1kB is 1 PUT if stored at ones ?
I woukd say it depends on which best reflects the work done by the network to both store and retrieve them. This will make the network more resistant to attack and more economically viable.
From a user perspective, I’m not sure it makes a huge difference, as some stuff will be more than one PUT. The user may not understand why this is the case. So, we gain some simplicity with a standard PUT cost, but lose it again as data stored gets larger.
The more that is unfamiliar, the more steps required for the user to action what matters.
Per byte or per kb is immediate to many, Put is new to all.
Per size also makes comparisons to competitors immediate… before noting one off costs etc. Also, guessing at the expected costs would be easier.
So, based on bias only for removing new jargon, by size seems better for the user.
You raise good points here. I voted per PUT for obvious reasons. However, I think that Puts should effectively be a fixed size so that it can by marketed as a cost per MB and PUT never mentioned outside of the inner geekdom.
Given the 4kB sector size of storage and the 1kB optimal packet size for network transmission, I would say go with an easy multiple of that. The challenge is that smaller chunk sizes place more load on the network vaults. The first size up from the 4kB packet that feels right and reasonable from both a marketing and technical perspective is the 1MB chunk we have now.
I’m not sure that it’s obvious. Can you elaborate?
I can understand the desire for per PUT since there’s fixed overheads to deal with.
I can understand the desire for per byte since hitting ‘like’ is surely less costly than uploading a minute of audio, despite both taking only 1 PUT.
So I’m really unsure which way this should go. A combo of both ‘fixed’ overheads and ‘variable’ size would be good but is maybe not so user friendly and is also complex to turn into a storecost algorithm.
For me, one issue is forcing folk to fit as much per put as possible, encouraging less chunks and less work for the network. Dust type transactions could be an easy attack.
Mostly network overhead and spam dust protection as dirvine pointed to. Also hdd and ssd sector size considerations. Any PUT less than 4kB will typically consume 4kB on disk anyhow.
It would seem that you might be able to find a happy medium in something like an appendable data chunk. Iirc you would have 1 thousand 1k data lines in the 1M AD chunk. You pay a PUT to create the AD on the network, but after that filling each of the 1k lines one by one is free or again costs 1 milliPUT. Just spitballing here but it seems like it would be rather efficient since each line is roughly the optimal udp packet size. When the AD is full you would need to buy another one and repeat. The end effect is that you are still paying a fixed price per MB stored, and have some control at the dust level. This wouldn’t control spam as well as charging 1 PUT for each 1k line but it isn’t a free for all either.
EDIT: It probably makes more sense to have an AD or MD with 250 lines at 4kB each. Would save a lot of write amplification on vaults with SSD arrays.
Thats why I suggest a cost based on the full 1MB price but discount it is small or large
For example have the store cost and work out the charge to charge as say
0.5 * store cost + ( Size / 4KB ) * ( store cost / 512 )
The 512 is equal to (2 * 1MB / 4KB)
Now I would perhaps suggest using 100KB as the unit size so
0.5 * store cost + ( Size / 100KB ) * ( store cost / 20 )
The 0.5 is the estimated amount that covers the overheads of storing and having unit sizes of 4KB or 100KB reduces issues of “dust” issues
Although that is fine from a technical perspective it loses the simplicity of having a fixed cost per MB for the layman. It would be nice to be able to look at your safecoin balance and the current storecost and know exactly how many MB you can store at that instant.
Also, not sure if it is valid to discount larger sizes unless you use variable chunk sizes up to GB range or higher… Personally, I really like the concept of fixed 1MB sector sizes for the big SAFE hdd in the sky (metaphorically speaking).
The social media nut does not think of MB. So a pretty mute point for the majority of PUTs when the network is worldwide. More emails, FB, Twitter, Forum, Blogs, photos, commenting, and other such things uploads are done than in file uploads. Most photos are reduced size photos of 20K to 500K since most are memes, funny cat photos, social media postings etc. (Yes there will be a lot of full size uploads to places, but dwarfed by the small social media stuff)
So people are more likely to see the number of puts available than the number of MBs they can upload. If they get more Puts than estimated then they will be happy.
Yes initially MB will be the driving factor, but when the 2 billion FB users come over to SAFE then that will be totally different.
I’m not convinced the culture of micro-puts like “likes” will transfer to SAFE at all. Costs add up. Even paying separately for each e-mail will have quite a barrier to entry. Do you all think intermediary services that offer e.g. a bulk of a certain number of mails or “likes” for a fixed cost might get created?
very basic question here:
Max PUT size is 1Mb right? So what happens if I try to upload a 3.1 Mb file? I’m guessing that the file upload succeeds, but I get charged for 4 PUTs. Is that right?
Briefly: Max chunk is 1mb. Larger files are split up into chunks all of which are stored, and a data map created as part of the self encryption process. (On phone so don’t have link to the primer for this just now)