A Verkle tree is a dedication scheme that works just like a Merkle tree, however has a lot smaller witnesses. It works by changing the hashes in a Merkle tree with a vector dedication, which makes wider branching elements extra environment friendly.
Thanks to Kevaundray Wedderburn for suggestions on the submit.
For particulars on how verkle timber work, see:
The intention of this submit is to elucidate the concrete structure of the draft verkle tree EIP. It is aimed toward consumer builders who need to implement verkle timber and are in search of an introduction earlier than delving deeper into the EIP.
Verkle timber introduce a lot of adjustments to the tree construction. The most important adjustments are:
- a swap from 20 byte keys to 32 byte keys (to not be confused with 32 byte addresses, which is a separate change);
- the merge of the account and storage tries; and eventually
- The introduction of the verkle trie itself, which makes use of vector commitments as a substitute of hashes.
As the vector dedication scheme for the verkle tree, we use Pedersen commitments. Pedersen commitments are based mostly on elliptic curves. For an introduction to Pedersen commitments and how one can use them as polynomial or vector commitments utilizing Inner Product Argumentss, see right here.
The curve we’re utilizing is Bandersnatch. This curve was chosen as a result of it’s performant, and likewise as a result of it is going to enable environment friendly SNARKs in BLS12_381 to motive in regards to the verkle tree sooner or later. This might be helpful for rollups in addition to permitting an improve the place all witnesses might be compressed into one SNARK as soon as that turns into sensible, without having an additional dedication replace.
The curve order/scalar area dimension of bandersnatch is p = 13108968793781547619861935127046491459309155893440570251786403306729687672801, which is a 253 bit prime. As a results of this, we are able to solely safely decide to bit strings of at most 252 bits, in any other case the sector overflows. We selected a branching issue (width) of 256 for the verkle tree, which suggests every dedication can decide to as much as 256 values of 252 bits every (or to be exact, integers as much as p – 1). We write this as Commit(v₀, v₁, …, v₂₅₅) to decide to the listing v of size 256.
Layout of the verkle tree
One of the design objectives with the verkle tree EIP is to make accesses to neighbouring positions (e.g. storage with virtually the identical deal with or neighbouring code chunks) low cost to entry. In order to do that, a key consists of a stem of 31 bytes and a suffix of 1 byte for a complete of 32 bytes. The key scheme is designed in order that “close” storage areas are mapped to the identical stem and a distinct suffix. For particulars please take a look at the EIP draft.
The verkle tree itself is then composed of two kinds of nodes:
- Extension nodes, that symbolize 256 values with the identical stem however totally different suffixes
- Inner nodes, which have as much as 256 youngsters, which might be both different internal nodes or extension nodes.
The dedication to an extension node is a dedication to a 4 component vector; the remaining positions might be 0. It is:
C₁ and C₂ are two additional commitments that decide to all of the values with stem equal to stem. The motive we have to commitments is that values have 32 bytes, however we are able to solely retailer 252 bits per area component. A single dedication would thus not be sufficient to retailer 256 values. So as a substitute C₁ shops the values for suffix 0 to 127, and C₂ shops 128 to 255, the place the values are cut up in two to be able to match into the sector dimension (we’ll come to that later.)
The extension along with the commitments C₁ and C₂ are known as “extension-and-suffix tree” (EaS for brief).
Figure 1 Representation of a stroll by a verkle tree for the important thing 0xfe0002abcd..ff04: the trail goes by 3 inside nodes with 256 youngsters every (254, 0, 2), one extension node representing abcd..ff and the 2 suffix tree commitments, together with the worth for 04, v₄. Note that stem is definitely the primary 31 bytes of the important thing, together with the trail by the interior nodes.
Commitment to the values leaf nodes
Each extension and suffix tree node comprise 256 values. Because a price is 256 bits extensive, and we are able to solely retailer 252 bits safely in a single area component, 4 bits can be misplaced if we merely tried so retailer one worth in a single area component.
To circumvent this drawback, we selected to partition the group of 256 values into two teams of 128 values every. Each 32-byte worth in a bunch is cut up into two 16-byte values. So a price vᵢ∈ 𝔹₃₂ is became v⁽ˡᵒʷᵉʳ⁾ᵢ ∈ 𝔹₁₆ and v⁽ᵘᵖᵖᵉʳ⁾ᵢ∈ 𝔹₁₆ such that v⁽ˡᵒʷᵉʳ⁾ᵢ ++ v⁽ᵘᵖᵖᵉʳ⁾ᵢ= vᵢ.
A “leaf marker” is added to the v⁽ˡᵒʷᵉʳ⁾ᵢ, to distinguish between a leaf that has by no means been accessed and a leaf that has been overwritten with 0s. No worth ever will get deleted from a verkle tree. This is required for upcoming state expiry schemes. That marker is about on the 129th bit, i.e. v⁽ˡᵒʷᵉʳ ᵐᵒᵈⁱᶠⁱᵉᵈ⁾ᵢ = v⁽ˡᵒʷᵉʳ⁾ᵢ + 2¹²⁸ if vᵢ has been accessed earlier than, and v⁽ˡᵒʷᵉʳ ᵐᵒᵈⁱᶠⁱᵉᵈ⁾ᵢ = 0 if vᵢ has by no means been accessed.
The two commitments C₁ and C₂ are then outlined as
Commitment of extension nodes
The dedication to an extension node consists of an “extension marker”, which is simply the #1, the 2 subtree commitments C₁ and C₂, and the stem of the important thing resulting in this extension node.
Unlike extension nodes within the Merkle-Patricia tree, which solely comprise the part of the important thing that bridges the father or mother inside node to the kid inside node, the stem covers the entire key as much as that time. This is as a result of verkle timber are designed with stateless proofs in thoughts: if a brand new secret’s inserted that “splits” the extension in two, the older sibling needn’t be up to date, which permits for a smaller proof.
Commitment of Internal nodes
Internal nodes have the less complicated calculation technique for his or her commitments: the node is seen as a vector of 256 values, which might be the (area illustration of the) root dedication of every of their 256 subtrees. The dedication for an empty subtree is 0. If the subtree will not be empty, then the dedication for the interior node is
the place the Cᵢ are the youngsters of the interior node, and 0 if a toddler is empty.
Insertion into the tree
Figure 2 is an illustration of the method of inserting a brand new worth into the tree, which will get fascinating when the stems collide on a number of preliminary bytes.
Figure 2 Value v₁₉₂ is inserted at location 0000010000…0000 in a verkle tree containing solely worth v₁₂₇ at location 0000000000…0000. Because the stems differ on the third byte, two inside nodes are added till the differing byte. Then one other “extension-and-suffix” tree is inserted, with a full 31-byte stem. The preliminary node is untouched, and C²₀ has the identical worth as C⁰₀ earlier than the insertion.
Shallower timber, smaller proofs
The verkle tree construction makes for shallower timber, which reduces the quantity of saved knowledge. Its actual energy, nevertheless, comes from the flexibility to provide smaller proofs, i.e. witnesses. This might be defined within the subsequent article.