Kaustinen analytics

Last updated: 2023-11-05

The following is an automatically generated report from a modified geth node that captures metrics from the Kaustinen network.

This modified geth node is planned to continue running in Kaustinen, so as the state grows and have bigger state we can keep track if the metrics makes sense considering the design.

Once in a while I'll be updating this document with fresher reports.

Warn: take numbers with a grain of salt – we're in early stages of Kaustinen and the geth-vkt version is underly heavy development. There might be bugs, optimizable code, etc. If something looks odd, ask!

Tree stats report

The analyzer tracks every 200 blocks:

• block_number: the block number where these metrics where collected.
• depthX_count: how many leaf nodes are at this particular depth.
• leaf_node_count: how many leaf nodes are in the tree.
• internal_node_count: how many non-leaf nodes are in the tree.
• keyvalue_count: how many key-values are stored in the tree.

The following are the stats for the latest registered block:

block_number	depth1_count	depth2_count	depth3_count	depth4_count	depth5_count	leaf_node_count	internal_node_count	keyvalue_count
153200	0	1849	335473	7080	30	344432	67136	1377975

Interpretation hints:

• We could get an average leaf-node fill factor by computing kevalue_count/leaf_node_count. Currently, this value is ~5 which makes sense since this is the fill factor for accounts.
• The distribution of leaf node depths should be coherent with some ~expected depth regarding log_256(keyvalue_count), with reasonable outliers.

Witness stats report

For each witness in a block we register:

• block_numer: the block number of the analyzed witness.
• gas_used: gas used in the block.
• ssz_total_size: the total bytes size of the SSZ encoded witness.
• ssz_statediff_size: the total bytes of the SSZ encoded statediff part of the witness.
• ssz_verkleproof_size: the total bytes of the SSZ encoded verkleproof par of the witness.
• statediff_stem_count: how many stems are in the statediff.
• statediff_currentvalue_non-nil_count: how many currentValue in the statediff aren't null.
• statediff_newvalue_non-nil_count: how many newValue in the statediff aren't null.

The following are the top 10 witnesses ordered by ssz_total_size desc:

block_number	gas_used	ssz_total_size	ssz_statediff_size	ssz_verkleproof_size	statediff_stem_count	statediff_currentvalue_nonnil_count	statediff_newvalue_nonnil_count
136133	3430000	41914	26960	14946	102	408	103
136127	3430000	41914	26960	14946	102	408	103
132149	3430000	41914	26960	14946	102	408	103
132140	3430000	41914	26960	14946	102	408	103
132136	3430000	41914	26960	14946	102	408	103
132115	3430000	41914	26960	14946	102	408	103
132111	3430000	41914	26960	14946	102	408	103
132092	3430000	41914	26960	14946	102	408	103
131814	3430000	41914	26960	14946	102	408	103
131782	3430000	41914	26960	14946	102	408	103

Interpretation notes:

• Note that ssz_total_size ~= ssz_statediff_size + ssz_verkleproof_size, the separation is done to know how many bytes from the total size come from the statediff vs the proof.

Proof verification report

This report shows some experimental metrics of the e2e proof verification times:

• num_keys: how many keys are proven in this proof.
• numopenings: how many openings are part of this Multiproof.
• prestate_getelementsforproof_ms: how many ms takes to collect proof elements from the reconstructed pre-state tree.
• verify_multiproof_ms: how many ms takes to do the Multiproof verification.
• e2e_verif_time_ms: how many ms takes for all witness verificatio work (deserializing, reconstructing tree, verifying Multiproof, etc).

The following are the top 10 results ordered by num_keys desc:

num_keys	numopenings	prestate_getelementsforproof_ms	verify_multiproof_ms	e2e_verif_time_ms
510	1620	14	8	36
510	1620	13	9	35
510	1620	14	8	33
510	1620	13	9	32
510	1620	14	7	35
510	1620	13	10	34
510	1620	14	9	35
510	1620	13	8	33
510	1620	14	11	35
510	1620	19	8	41

Interpretation notes:

• The block number for each case isn't registered since it's mostly irrelevant for the metrics. What matters is num_keys and/or numopenings.

Proof generation report

This report shows some experimental metrics of the e2e proof generation times:

• num_keys: how many keys are proven in this proof.
• numopenings: how many openings are part of this Multiproof.
• prestate_getelementsforproof_ms: how many ms takes to collect proof elements from the reconstructed pre-state tree.
• create_multiproof_ms: how many ms takes to do the Multiproof proof generation.
• e2e_gen_time_ms: total time to generate the proof to be inserted in the block.
• e2e_gen_time_ms: how many ms takes to generate the witness (get elements from tree, generate Multiproof, serialize, etc).

num_keys	numopenings	prestate_getelementsforproof_ms	create_multiproof_ms	e2e_gen_time_ms
510	1620	115	51	166
510	1620	110	50	161
510	1620	128	49	178
510	1620	108	51	159
510	1620	137	55	194
510	1620	109	48	159
510	1620	112	48	161
510	1620	137	48	186
510	1620	136	52	189
510	1620	108	48	158

Interpretation notes:

• The block number for each case isn't registered since it's mostly irrelevant for the metrics. What matters is num_keys and/or numopenings.