text stringlengths 0 2.35k |
|---|
[693.90 --> 697.86] So once something was written in the history, the database cannot say that, |
[697.86 --> 699.24] oh, it was something different. |
[699.24 --> 700.40] It cannot lie to you. |
[700.42 --> 704.16] Because if it would lie, then you will immediately see this. |
[704.16 --> 706.52] Because of this mathematical proof. |
[706.72 --> 711.98] So if there is something crucial like audit logs, which after some time, |
[712.34 --> 715.58] you may want to do some investigation what happened over time. |
[715.58 --> 718.08] This gives us extra protection. |
[718.08 --> 720.82] But you can rely on this information. |
[721.04 --> 727.22] Because database has proven that up to this point in time, it is consistent with the whole |
[727.22 --> 727.62] history. |
[728.16 --> 728.24] Okay. |
[728.54 --> 728.82] Okay. |
[728.82 --> 734.96] I think you were covering a lot of problems that are addressed by immutability databases. |
[735.56 --> 739.56] First, I would like to clarify, immutability is an overloaded term. |
[740.06 --> 746.46] Because as Shonin, you were mentioning, with immutability, we usually refer to systems or |
[746.46 --> 753.16] data structures that are a pen only, that treat changes or updates as a new data, actually. |
[753.16 --> 759.40] So when we are doing an update of a record, we are not mutating the original record, but |
[759.40 --> 763.64] treating the update as a new record, describing the change. |
[764.38 --> 767.50] So we are used to that for immutability. |
[767.80 --> 774.12] And actually, ImmuDB relies on every component in ImmuDB is an appen only data structure. |
[774.64 --> 778.12] Even the cryptographic data structure are treated as appen only. |
[778.12 --> 784.86] But in immutability in databases or even in blockchain, we tend to refer to another thing, |
[785.06 --> 791.92] not just to append only, but to the possibility to verify that the history hasn't changed. |
[792.66 --> 798.08] So every record is registered and cryptographic linked to what happened before. |
[798.08 --> 805.12] And then you have a way to verify if a given transaction or a given record was present and |
[805.12 --> 808.70] was not modified anymore once it was written. |
[809.32 --> 813.98] Doesn't mean that you cannot have the current state of your balance account. |
[814.62 --> 820.76] And as a traditional database, you will have either as the current, the latest value that |
[820.76 --> 825.92] was placed for a given record, because the record will be the key that identifies the address |
[825.92 --> 827.22] or the balance. |
[827.82 --> 833.66] But also, depending on the use case, it may be a cumulative set of changes, like in Git, |
[833.76 --> 835.48] where we are committing changes. |
[836.14 --> 840.78] So the current state or the history, it's independent of that. |
[841.14 --> 844.88] What we refer to this type of thing is verifiable. |
[844.88 --> 852.12] I prefer the term verifiable database rather than immutable database, because every system |
[852.12 --> 854.04] has integrity checks, right? |
[854.14 --> 860.84] Internal integrity checks to check the consistency of a given record or of a given file, if it |
[860.84 --> 864.86] was consistent, is consistent or not. |
[865.36 --> 871.14] But with tampering detection, it's like giving the possibility to the client application or the |
[871.14 --> 876.60] application that is using the database to do their integrity validation by themselves. |
[877.14 --> 878.46] That is one of the differences. |
[879.00 --> 884.50] It's the application that is receiving the data from the database who is able to run the |
[884.50 --> 891.72] integrity check to validate that the data that was received was not modified since it was |
[891.72 --> 892.08] written. |
[892.38 --> 894.12] Okay, let's pull on that thread a little bit. |
[894.12 --> 901.50] So we're not talking about the clients sort of being or maintaining a copy of whatever |
[901.50 --> 906.90] data you might have at a central sort of immutable database or verifiable database, right? |
[906.92 --> 911.88] You're talking about some sort of a cryptographic verifiability of the data. |
[912.12 --> 918.92] So one of the particularities of an immutable database is that at every moment, the complete |
[918.92 --> 922.38] state of the database is captured by a hash value. |
[922.38 --> 929.28] So that denotes not only the current state, but what the complete history of changes up |
[929.28 --> 929.92] to that point. |
[930.46 --> 937.42] So the client in InModb, for instance, or in other immutable databases, is the client who |
[937.42 --> 940.06] needs to keep track of this current state. |
[940.52 --> 947.38] The latest state that is known is like in the example that Bar mentioned regarding your bank |
[947.38 --> 952.74] bank balance account, you may know what was the latest state that you can trust. |
[953.20 --> 957.14] And based on that and the new changes is where you can compare. |
[957.38 --> 962.56] You have the base to compare the new changes or the new results and so on. |
[962.68 --> 968.22] So, but the client only needs to keep track of the state of the database at any given point. |
[968.36 --> 970.10] That is the minimal information. |
[970.10 --> 974.60] So to make sure I understand this, that means that deleting records also isn't permissible. |
[975.00 --> 975.52] Is that true? |
[975.88 --> 977.56] Deletion is actually depending. |
[977.90 --> 979.08] We have two levels. |
[979.46 --> 982.40] We have logical deletion or physical deletion. |
[982.86 --> 988.08] Logical deletion is something that can be handled by the application or by the server. |
[988.48 --> 993.54] But the difference will be in terms of performance because the filtering out of the information will |
[993.54 --> 997.88] be done much faster if it's done directly by the database. |
[997.88 --> 1005.10] In NemoDB, we currently have support for logical deletion in both manners, like deleting a key, |
[1005.18 --> 1008.28] for instance, or by providing an expiration date. |
[1008.90 --> 1011.86] But this currently is just a logical deletion. |
[1012.04 --> 1014.14] This means the data will be still there. |
[1014.60 --> 1018.66] It will be automatically filtered out and the client won't receive it. |
[1018.92 --> 1021.28] But it's not yet physically deleted. |
[1022.02 --> 1026.16] And we are under discussions to incorporate physical deletion of data. |
[1026.16 --> 1032.98] And it's a very, very interesting topic to discuss what involves physically deleting the data |
[1032.98 --> 1035.46] and yet being able to prove. |
[1035.78 --> 1042.20] So depending on the data you delete or you remove, is the possibility you have later on to build proof. |
[1042.76 --> 1044.60] So it's a very, very interesting topic. |
[1045.14 --> 1045.22] Yeah. |
[1045.22 --> 1049.48] And I'm assuming we're going to want to talk about good use cases for an immutable database. |
[1049.84 --> 1053.62] But I guess the first thing that comes to my mind is I feel like you'd have to be careful |
[1053.62 --> 1058.84] as to what applications you use this for because there are rules like GDPR where you have to be able |
[1058.84 --> 1060.18] to forget people, essentially. |
[1060.84 --> 1064.74] And I could imagine a weird situation where you write something to an immutable database |
[1064.74 --> 1067.90] accidentally and then realize, like, how do we fix this? |
[1067.90 --> 1074.98] And actually, GDPR is the main reason why we started actually thinking about physical deletion |
[1074.98 --> 1083.44] because some laws require from you to make sure that the data is not accessible at all after some time. |
[1083.80 --> 1089.04] Of course, the rules are not clear because sometimes you have to hide the data from the users, |
[1089.04 --> 1094.54] but then you have to keep it for a longer time because there may be some kind of investigation later on. |
[1094.54 --> 1098.72] But still, it is possible to actually remove the data. |
[1098.94 --> 1103.78] And maybe there is a different reason for that because if you have append-only structure, |
[1104.00 --> 1109.08] append-only data, and you start putting too much data into it, you will just run out of space. |
[1109.86 --> 1115.16] And after sometimes you want to reclaim maybe the space or you have physical constraints of your server |
[1115.16 --> 1118.48] and you have to deal with that and there is a production system running. |
[1118.48 --> 1126.26] So maybe you want to just wipe data that is older than some point in time in the past. |
[1127.18 --> 1134.60] And still, the state, as Hiromimo said, the state of the database, the hash of the database contains all the history. |
[1135.24 --> 1137.08] So this is a very interesting topic. |
[1137.22 --> 1143.38] So you no longer have the data, but the state needs to calculate this data in |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.