Suhyeon Lee, Tokamak: 180 Days after EIP-4844: Will blob sharing solve small rollups' dilemma?

[Music] uh hello uh I'm s Lee I'm a security researcher at to network also parttime lecture in Korea University so today I will talk about uh The Blob sharing and my simulation resarch so the subtitle is real blob sharing so dilemma for small laurels so I'm not sure you're all uh family with blops but in etherum it's uh this concept was uh proposed by EIP 4844 and it dramatically reduced cost for the data availability for RS so um uh so um the characteristic of blops is that it's the the size The Blob each blob size 128 kilobyte and uh even if unlike cold data you in in cold data you pay for the bytes you use in cold data but in bloves you you always have to pay for 128 kilobyte even if you use just one bite or two bytes so um also another characteristic is blob fee mechanism that is that uh if there are more than blops than some stred this is now the three then BL blood base fee increases exponentially also there are like a lck of lops then it decreases exponentially so uh there are uh there's a dilemma for smalls so my company also has a small layer to main that but the problem is we don't have like a lot of transactions like base or the op painet or like arbitrum so uh and for many small Wars 128 kilobyte actually is quite a big then Smalls have to choose either one of them one first option is using BLS without filling most of 128 kilobytes so you have to pay very inefficiently like you have to pay way more than you use so for example uh Tao the their a average blob size is the rear blob size is um 30 kiloby for jerk kit 6 kilobyte and the metal 0.1 75 kilobyte but like a big rollups like Bas they use 126 k and the op main net also same the other strategy is the use the real long delay for uh data availabilities for example Boba network uh one of small buls but 150 blocks it's the the the gap of The Blob submission and the other example the bank chain is 570 the block Gap but for example big RS base nine just nine blobs and uh op main na 24 blocks so it means big wups uh provides data availab availability just U just on time but small wups for the because of the cost they have really big the the ba gaps so uh so this is uh also dilemma for our the our main net so I was finding some solutions and I found a concept called blob sharing oh but it was uh proposed like last year and was at the time it was very theoretical level so I'm like oh now um there are like six months of real blob submission data so let's check it out so the question was what if rule of share blobs so goal is uh goal goal of this mini project was cost analysis of rups b blop sharing and the target rups are like 28 R rups labeled in Dune because I could the easily analyze the data and the expectation my expectation is small lups may get really good cost reduction um use just small lops not big big rops because big rops are very efficiently using blops so uh then how can I simulate blob sharing so I had to make uh such assumptions for example the first assumption is the core assumption the constant data rate so and okay I'll uh explain Conant data after the three out three the other assumptions and also I assumed the there there are there is a virture separate da contract is can be proxy da contract for other ruup da contract and there's a uniform gas consumption uh because the the rovs have their own da contract and for the The Blob submission they have uh different uh their own the da da function in smart contract usually they just is uh they just consume the minimum gas but some rovs use their own customized function so they cost more but for simulation I assume they use all the same the the function costing the minimal gas and also the last assumption is that the after blob sharing there will be less number of the less number of blobs but I assume that this the change wouldn't um change change the etherum just transaction base fee or priority fee so I adopted all the the medium priority fee for blob sharing simulation so what is the first core assumption so uh uh let's assume that there are like three blob submissions first blob was submitted at the block number 80 and it's 50 kilobyte and second blob is at one blog number 100 and 6 kilobyte and the last one is at uh block 150 and it was uh 100 kiloby then uh I assume that uh between 80 to 100 this rup uh generated the 6 kilobyte then it means at every block number it it generated three kilobyte and the same for the the gap between the 100 and 150 so so with this core assumption I pro pro uh pre-process data and reconstructed the blobs for example uh as I assumed in the a previous slide so uh this rul up uh generated the 3 kilobyte between 80 to 100 and 2 kilobyte for every block in between 100 and 150 so I got like a little Legos of The Blob data and I re uh and I collected all the this the small datab bytes of rollups and I reconstructed blobs uh blobs like a shared blobs so uh with these Legos Legos I made like 128 kilobyte the shared blob for the with using the rear data so uh there are some other processes one is the handling access data so the The Blob maximum size 128 k kiloby so after 128 then I had to I generated another blob and also there's uh another the Restriction there can be only six maximum blobs blops for Block so uh if there there's a generated over like six blobs then I I submitted the blob in the next blob block and also they are unlabeled blobs like I used um the main 28 uh blobs uh 28 rups that labeled in dun but there are non-labeled blobs too then uh for uh so only the 28 rups data was Sim the reconstructed so I made unlid blops should be submitted in uh the prioritized yeah so this is the res uh for me this res very was very interesting because uh uh or not only small rups all the rollups got 82 82 to 99 uh percentage cost saving in USD dollars so uh the one one thing is that uh then why all the blops got this uh dram um dramatic cost saving the thing was the blood base fee um so uh the the the most black part is the cost the where the rollups spend for blood base fees but as you see that for many lops the red part disappeared then why uh it's because of the number of blops so as I told you the blob Bas speed mechanism moves uh exponentially as the as the excessive The Blob counts so if there are uh more than three blobs then the blob base speed increases exponentially but uh here's there are two graphs so left is the just uh number of blobs the number of blobs are not that different only like a 20% difference but in the right side there are the blocks with over three blobs so but uh the the total count of blobs are not that different but blocks with over um three blops are DEC the decrease dramatically that made the blob base fee very smooth so left is the rearward blob base fee graph and the right side is the blob base fee movement after the blob sharing simulation so this graft y AIS is the log scale so it means the Block Base fee difference was uh very very big so that made blood base fee like a like 9 99% cheaper than previous uh the the rear data so uh I I just checked confirm that blob sharing has a really big potential but there are still open questions so that was just simulation so there I made three op open questions one is so still there are not the detailed blob sharing schemes for so we need the detailed the smart contract design for blob sharing and also there should be blob sharing data structure for security for example if we just shared the put uh upload the random the data chunks then the rups sequencers can't like read data so the the last one is the blob sharing cost and re distribution in theor one the study says that big rups uh May will have the bigger negotiation power based on based on scam Theory but as we saw in the simulation res the most beneficient uh from The Blob sharing is the big rollups in the after the blob sharing so so here are key takeaways so um it's uh The Blob sharing is very beneficial and has a really big potential but there are still open questions and the detailed data and description can be uh read in the the my the research paper and the the left QR code and you can discuss with me on the right C code which is the E research website thank you [Music] la