Benchmarks

Measured runtimes for the core operations of the Ocelescope base tool on public object-centric event logs: import, export, filtering, and discovery.

Each configuration was run 5 times on the machine below, and the reported value is the median in seconds. Charts show these medians; the complete per-run numbers are in the All runs table under each section.

A separate suite at the end of this page compares the ocelescope library against pm4py.

Environment

Ocelescope	v1.0.3
OS	Arch Linux
Kernel	7.1.3-arch1-2, x86_64
CPU	12th Gen Intel Core i7-1260P
Cores / threads	12 cores (4 P + 8 E) / 16 threads
Memory	32 GB (31.0 GiB)

Logs

Short name	Log variant	Source	Events	Objects
p2p	p2p	10.5281/zenodo.8412920	14,671	9,543
order-mgmt	order-management	10.5281/zenodo.18373906	21,008	10,840
container	container logistics	10.5281/zenodo.18373888	35,372	13,882
aoe2	aoe2	10.5281/zenodo.13365584	2,372,505	630,590
steel	steel logistics	proprietary, not public	865,859	110,834
hinge	hinge	10.5281/zenodo.13638681	38,528	23,771
angular	angular github comments	10.5281/zenodo.8430332	27,847	28,317
BPI 2017	BPI Challenge 2017	4121/uuid:5f3067df…	1,202,267	31,509

Import

Median import time by file format Logs ordered by size.

• xml
• json
• sqlite
• xes

Logarithmic time axis: bar length is not proportional to the value.

All runs

Log variant	Format	File size	Run 1	Run 2	Run 3	Run 4	Run 5	Median
p2p	xml	17.6 MB	0.527	0.474	0.530	0.487	0.485	0.487
p2p	json	13.6 MB	0.604	0.607	0.622	0.603	0.600	0.604
p2p	sqlite	13.1 MB	0.446	0.427	0.429	0.435	0.442	0.435
order-mgmt	xml	15.2 MB	0.317	0.309	0.302	0.321	0.296	0.309
order-mgmt	json	12.6 MB	0.422	0.421	0.419	0.416	0.414	0.419
order-mgmt	sqlite	36.6 MB	0.251	0.237	0.249	0.256	0.253	0.251
container	xml	13.4 MB	0.276	0.278	0.282	0.277	0.285	0.278
container	json	10.0 MB	0.346	0.352	0.340	0.336	0.343	0.343
container	sqlite	23.0 MB	0.234	0.248	0.235	0.237	0.231	0.235
aoe2	xml	1.3 GB	44.579	44.279	45.856	45.031	45.914	45.031
aoe2	json	1.6 GB	46.261	46.675	46.921	46.647	47.226	46.675
aoe2	sqlite	891.9 MB	8.662	8.542	8.444	8.519	8.506	8.519
steel	xml	1.1 GB	46.423	47.019	47.365	47.822	46.719	47.019
steel	json	922.7 MB	51.213	51.482	52.912	51.361	48.163	51.361
steel	sqlite	580.1 MB	17.796	17.833	17.920	17.566	17.945	17.833
hinge	xml	40.2 MB	0.654	0.649	0.644	0.647	0.644	0.647
hinge	json	29.1 MB	0.951	0.935	0.955	0.974	0.961	0.955
hinge	sqlite	19.8 MB	0.496	0.495	0.491	0.498	0.497	0.496
angular	xml	178.1 MB	2.238	2.312	2.260	2.236	2.340	2.260
BPI 2017	xes	29.7 MB	5.524	5.594	5.558	5.544	5.605	5.558

Export

Median export time by file format Native OCEL exports. The flattened xes export writes a single object type and is listed in the table below only.

• xml
• json
• sqlite

Logarithmic time axis: bar length is not proportional to the value.

All runs

A blank Flattened type means no flattening (native OCEL export).

Log variant	Format	Flattened type	Run 1	Run 2	Run 3	Run 4	Run 5	Median
p2p	xml	—	1.194	1.220	1.182	1.209	1.178	1.194
p2p	json	—	0.418	0.418	0.401	0.414	0.406	0.414
p2p	sqlite	—	0.385	0.394	0.403	0.388	0.411	0.394
p2p	xes	material	0.264	0.270	0.258	0.278	0.294	0.270
order-mgmt	xml	—	1.239	1.236	1.234	1.244	1.224	1.236
order-mgmt	json	—	0.375	0.360	0.446	0.384	0.371	0.375
order-mgmt	sqlite	—	0.294	0.317	0.292	0.322	0.301	0.301
order-mgmt	xes	items	0.314	0.325	0.312	0.322	0.313	0.314
container	xml	—	1.188	1.130	1.127	1.148	1.144	1.144
container	json	—	0.303	0.291	0.291	0.300	0.294	0.294
container	sqlite	—	0.316	0.341	0.318	0.289	0.308	0.316
container	xes	not recorded	0.230	0.233	0.237	0.228	0.238	0.233
aoe2	xml	—	90.003	88.705	89.143	—	—	89.143 ⚠️ 3 runs
aoe2	json	—	20.557	20.047	20.613	20.588	20.188	20.557
aoe2	sqlite	—	23.347	22.383	20.361	22.179	19.627	22.179
steel	xml	—	75.897	76.916	76.137	—	—	76.137 ⚠️ 3 runs
steel	json	—	18.590	18.409	18.626	18.673	18.310	18.590
steel	sqlite	—	23.327	23.674	23.942	23.511	22.874	23.511
steel	xes	stack	7.904	7.675	7.676	7.789	7.865	7.789
hinge	xml	—	2.879	2.792	2.958	2.932	2.894	2.894
hinge	json	—	0.725	0.747	0.737	0.746	0.763	0.746
hinge	sqlite	—	0.545	0.596	0.558	0.860	0.543	0.558 ⚠️ outlier
hinge	xes	Workstation	0.485	0.502	0.487	0.489	0.491	0.489

Filtering

Median filter time and how much each configuration removed Nine filter configurations.

• Runtime
• Events excluded
• Objects excluded

Each panel has its own axis: runtime is linear in seconds, exclusions are a logarithmic count, so bar lengths only compare within a panel. A configuration that excluded nothing shows 0 at the axis start.

Summary

ID	Log variant	Filter (summary)	Excl. events	Excl. objects	Median (s)
F1	p2p	Activity: exclude 5 activities	3,551	0	0.312
F2	p2p	Time frame: 2023-04-01 → 2024-04-01	8,726	0	0.303
F3	p2p	Object attributes ×3 (regex + numeric range)	0	2,919	0.288
F4	p2p	E2O count ×2	206	0	0.330
F5	order-mgmt	Object type + time frame + object attr. + E2O count (4 filters)	7,617	10,818	0.157
F6	container	Object type + time frame + 3× object attr. + 3× E2O count (8 filters)	23,110	7,387	0.184
F7	aoe2	Object type + activity + O2O count + time frame + 2× E2O count (6 filters)	2,177,411	386,022	2.856
F8	steel	Object type + time frame + event attr. + 2× O2O count (5 filters)	441,744	31,077	4.986
F9	hinge	Object type + time frame + activity (3 filters)	29,101	2,225	0.319

All runs

ID	Run 1	Run 2	Run 3	Run 4	Run 5	Median
F1	0.309	0.310	0.318	0.320	0.312	0.312
F2	0.303	0.303	0.305	0.296	0.298	0.303
F3	0.293	0.296	0.288	0.285	0.277	0.288
F4	0.321	0.335	0.330	0.317	0.331	0.330
F5	0.150	0.157	0.159	0.153	0.163	0.157
F6	0.174	0.184	0.186	0.186	0.182	0.184
F7	3.018	2.878	2.856	2.840	2.827	2.856
F8	4.929	4.986	5.238	5.199	4.945	4.986
F9	0.321	0.318	0.319	0.317	0.327	0.319

Discovery

Settings: OC-DFG at frequency threshold 1.0, Inductive Miner (flattening) at frequency threshold 0.8. The two largest logs were pre-filtered first; the sizes below are after filtering. aoe2 was reduced to its top 80% activities and 90% objects; steel logistics by a two-part exclusion filter listing 87 object types and 587 activity types explicitly (largely storage locations, installations, and movement events).

Median discovery time by algorithm Logs ordered by size.

• OC-DFG
• Inductive Miner

Logarithmic time axis: bar length is not proportional to the value.

All runs

The Events and Objects columns are the size of the log after the pre-filter was applied.

Log variant	Algorithm	Pre-filter	Events	Objects	Run 1	Run 2	Run 3	Run 4	Run 5	Median
p2p	OC-DFG	none	14,671	9,543	1.836	1.888	1.893	1.866	1.886	1.886
p2p	Inductive Miner	none	14,671	9,543	1.394	1.396	1.486	1.404	1.403	1.403
order-mgmt	OC-DFG	none	21,008	10,840	3.282	3.170	3.174	3.105	3.253	3.174
order-mgmt	Inductive Miner	none	21,008	10,840	4.485	4.456	4.553	4.381	4.537	4.485
container	OC-DFG	none	35,372	13,882	3.382	3.361	3.410	3.422	3.520	3.410
container	Inductive Miner	none	35,372	13,882	2.439	2.445	2.454	2.468	2.463	2.454
aoe2	OC-DFG	top 80% activities / 90% objects	1,204,066	308,295	117.451	116.754	120.701	118.750	118.566	118.566
aoe2	Inductive Miner	top 80% activities / 90% objects	1,204,066	308,295	68.952	70.249	70.437	70.214	70.854	70.249
steel	Inductive Miner	exclude 87 object types + 587 activities	557,063	66,349	55.718	38.956	57.457	39.857	38.959	39.857 ⚠️ bimodal
hinge	OC-DFG	none	38,528	23,771	4.989	4.735	4.909	4.859	4.753	4.859
hinge	Inductive Miner	none	38,528	23,771	8.729	8.849	8.868	8.937	8.728	8.849

Comparison with pm4py

The sections above measure the base tool. This one measures the ocelescope library on its own, against pm4py doing the same three things to the same files: import, filtering, and export. The logs are container, order-mgmt, and aoe2 from the table at the top of the page.

The two libraries hold a log differently, and that is what the numbers are about. pm4py keeps every table as an in-memory pandas DataFrame. ocelescope keeps the log in a DuckDB database and reshapes DataFrames out of it only when something asks for them, so what a log costs in RAM is decoupled from what it costs to work with. Both readers OCEL.read offers are measured: the default variant="r4pm" and variant="streamed". .sqlite logs always stream, so they have no separate r4pm row.

Setup

ocelescope	0.6.0
pm4py	2.7.23.4
Python	3.13.11
Kernel	7.1.8-arch1-3, x86_64
Runs per case	3

Time is the fastest of the three runs and RAM the highest peak, rather than the median-of-five used earlier on the page. The suite that produced these numbers is at promi4s/ocelescope-library-benchmarks.

Import memory

Peak RAM above baseline while importing Three readers per log and format. Lower is better.

• ocelescope (streamed)
• ocelescope (r4pm)
• pm4py

Logarithmic axis: bar length is not proportional to the value. Sub-gigabyte differences between the small logs are real but visually compressed.

Import time

Import wall time by reader Fastest of three runs.

• ocelescope (streamed)
• ocelescope (r4pm)
• pm4py

Logarithmic time axis: bar length is not proportional to the value.

All runs

Wall time per run, in seconds:

Log	Format	Reader	Run 1	Run 2	Run 3	Best
container	json	ocelescope (streamed)	0.377	0.393	0.393	0.377
container	json	ocelescope (r4pm)	0.324	0.261	0.260	0.260
container	json	pm4py	0.121	0.117	0.118	0.117
container	xml	ocelescope (streamed)	0.450	0.454	0.457	0.450
container	xml	ocelescope (r4pm)	0.234	0.220	0.226	0.220
container	xml	pm4py	0.136	0.146	0.140	0.136
container	sqlite	ocelescope (streamed)	0.137	0.144	0.139	0.137
container	sqlite	pm4py	0.896	0.901	0.870	0.870
order-mgmt	json	ocelescope (streamed)	0.514	0.494	0.531	0.494
order-mgmt	json	ocelescope (r4pm)	0.308	0.312	0.300	0.300
order-mgmt	json	pm4py	0.158	0.151	0.152	0.151
order-mgmt	xml	ocelescope (streamed)	0.503	0.529	0.514	0.503
order-mgmt	xml	ocelescope (r4pm)	0.239	0.231	0.239	0.231
order-mgmt	xml	pm4py	0.167	0.164	0.166	0.164
order-mgmt	sqlite	ocelescope (streamed)	0.157	0.155	0.153	0.153
order-mgmt	sqlite	pm4py	0.764	0.762	0.771	0.762
aoe2	json	ocelescope (streamed)	33.97	34.22	33.86	33.86
aoe2	json	ocelescope (r4pm)	16.49	16.76	16.88	16.49
aoe2	json	pm4py	9.53	9.31	9.69	9.31
aoe2	xml	ocelescope (streamed)	36.91	36.38	36.96	36.38
aoe2	xml	ocelescope (r4pm)	9.71	9.78	9.68	9.68
aoe2	xml	pm4py	8.57	8.55	8.49	8.49
aoe2	sqlite	ocelescope (streamed)	7.38	7.42	7.57	7.38
aoe2	sqlite	pm4py	63.41	63.44	63.29	63.29

Peak RAM above baseline per run, in MB:

Log	Format	Reader	Run 1	Run 2	Run 3	Max	Retained
container	json	ocelescope (streamed)	118	119	123	123	124
container	json	ocelescope (r4pm)	200	202	200	202	203
container	json	pm4py	158	159	160	160	161
container	xml	ocelescope (streamed)	132	132	134	134	121
container	xml	ocelescope (r4pm)	216	206	227	227	215
container	xml	pm4py	163	158	160	163	164
container	sqlite	ocelescope (streamed)	118	118	126	126	127
container	sqlite	pm4py	123	123	123	123	102
order-mgmt	json	ocelescope (streamed)	138	134	134	138	138
order-mgmt	json	ocelescope (r4pm)	216	211	223	223	224
order-mgmt	json	pm4py	209	203	207	209	197
order-mgmt	xml	ocelescope (streamed)	149	142	139	149	134
order-mgmt	xml	ocelescope (r4pm)	241	226	243	243	229
order-mgmt	xml	pm4py	209	209	208	209	198
order-mgmt	sqlite	ocelescope (streamed)	127	120	112	127	128
order-mgmt	sqlite	pm4py	126	126	126	126	119
aoe2	json	ocelescope (streamed)	925	874	909	925	913
aoe2	json	ocelescope (r4pm)	6,427	6,426	6,430	6,430	1,789
aoe2	json	pm4py	6,432	6,429	6,433	6,433	3,814
aoe2	xml	ocelescope (streamed)	2,159	2,171	2,227	2,227	926
aoe2	xml	ocelescope (r4pm)	6,428	6,426	6,429	6,429	2,455
aoe2	xml	pm4py	6,431	6,434	6,431	6,434	3,818
aoe2	sqlite	ocelescope (streamed)	980	978	975	980	980
aoe2	sqlite	pm4py	6,657	6,657	6,657	6,657	4,036

Filtering

All on aoe2, both libraries starting from the same sqlite log. The log is loaded once per library and only the filter call is timed, so import cost stays out of the numbers.

Filter	ocelescope	pm4py
activity	EventTypeFilter	filter_ocel_event_attribute
object-type	ObjectTypeFilter	filter_ocel_object_types
timeframe	TimeFrameFilter	filter_ocel_events_timestamp

activity keeps every activity at least as frequent as Start Build Farm, object-type keeps Villager / Farm / House, timeframe keeps 2023-11-30 to 2023-12-14.

Filter time on aoe2 Fastest of three runs, filter call only.

• ocelescope
• pm4py

All timings

Filter	Engine	Time
activity	ocelescope	1.97 s
activity	pm4py	7.65 s
object-type	ocelescope	1.62 s
object-type	pm4py	8.12 s
timeframe	ocelescope	1.81 s
timeframe	pm4py	8.83 s

Export

Also aoe2, also loaded from sqlite, with the clock starting once the log is instantiated. Each library writes all three formats.

Export time by format on aoe2 Fastest of three runs.

• ocelescope
• pm4py