Skip to content
Request a demo
Aurora Integration EngineBENCHMARK · NOVEMBER 2026 EDITION

Aurora Integration Engine 2.5.1:
performance measured, not claimed.

Full campaign from 30 October to 2 November 2026: 144 MLLP sweep steps against Mirth Connect 4.5.2 on the same bench and five eight-hour runs. Every ACK verified, every delivery reconciled with the destinations.

0
messages lost out of more than 3 million in the eight-hour runs and the 144 sweep steps
up to 6.5×
messages per second compared with Mirth Connect 4.5.2 PRODUCTION on the same bench (2.0× to 4.5× compared with RAW)
4.1 ms
median ACK latency at 100 messages per second for eight hours (ADT)
3,020 msg/s
highest throughput measured (simple ADT forwarding, 32 connections)

01 · MLLP sweeps

MLLP sweeps: Aurora and Mirth Connect on the same bench

Four channels (simple forwarding and clinical channel, 0.6 KB ADT and 3.1 KB ORU), 1, 4, 16 and 32 persistent connections in a closed loop, 5 seconds of warm-up and 20 measured seconds per step, three independent repetitions (median). Only a message with a correct AA ACK and all deliveries verified counts. Mirth Connect 4.5.2 with a 2 GiB heap, RAW and PRODUCTION storage; Aurora 2.5.1 with a 4 GiB heap and STANDARD trace.

Chart of successful messages per second by number of connections, Aurora 2.5.1 and Mirth Connect 4.5.2 RAW and PRODUCTION
Successful messages per second by number of connections.
Chart of median ACK latency by number of connections, Aurora 2.5.1 and Mirth Connect 4.5.2 RAW and PRODUCTION
Median ACK latency by number of connections (log scale).

Successful messages per second (median of three repetitions)

ChannelConn.Aurora 2.5.1Mirth RAWMirth PRODUCTIONAurora / RAWAurora / PRODUCTION
Simple ADT1482139863.5×5.6×
41,5494332593.6×6.0×
162,9418114843.6×6.1×
323,0206714634.5×6.5×
Simple ORU1348109703.2×5.0×
41,1373332083.4×5.5×
161,7736203882.9×4.6×
321,7075253723.2×4.6×
Clinical ADT119895522.1×3.8×
47163001492.4×4.8×
161,0744552472.4×4.4×
321,0643892122.7×5.0×
Clinical ORU115778462.0×3.4×
46112421282.5×4.8×
169153802082.4×4.4×
328893711902.4×4.7×

ACK latency, median / 99th percentile (ms)

ChannelConn.Aurora 2.5.1Mirth RAWMirth PRODUCTION
Simple ADT11.8 / 6.76.7 / 13.011.1 / 18.7
328.4 / 33.546.0 / 85.266.7 / 117.3
Simple ORU12.4 / 7.48.4 / 15.113.3 / 22.0
3215.9 / 48.558.8 / 104.982.7 / 142.8
Clinical ADT14.1 / 18.59.9 / 17.518.5 / 29.5
3226.0 / 87.679.1 / 146.4144.8 / 273.9
Clinical ORU15.0 / 22.211.9 / 19.420.6 / 33.3
3231.1 / 94.083.5 / 144.8161.2 / 300.1
  • Aurora delivers 2.0 to 4.5 times the messages per second of the RAW profile and 3.4 to 6.5 times those of the PRODUCTION profile; the median latency is 2.4 to 5.5 times lower than RAW and 4.2 to 7.9 times lower than PRODUCTION.
  • From 16 to 32 connections throughput stays flat or drops for all profiles and latency roughly doubles: beyond that point, in the closed loop, adding connections no longer increases messages per second.
  • At the 99th percentile Aurora stays below Mirth in every step except the clinical channel with one connection, where it is slightly above the RAW profile (18.5 vs 17.5 ms for ADT, 22.2 vs 19.4 ms for ORU).

02 · Eight hours

Eight hours of continuous load

Synthetic traffic, 100 msg/s

Synthetic traffic at 100 messages per second: clinical ADT channel at 80 msg/s over 8 connections and ORU at 20 msg/s over 4, three HTTP destinations each; 2,880,000 messages and 8,640,000 deliveries.

MeasureADTORU
Messages lost00
ACK median / 95th / 99th4.1 / 6.9 / 9.9 ms4.7 / 7.9 / 11.1 ms
Maximum1,464 ms in the first minute, 211 ms afterwards1,357 ms in the first minute, 205 ms afterwards

Average CPU 31.9% of one vCPU, maximum RSS 813 MiB, no Full GC; database between 6.64 and 8.53 GiB over the last four hours.

Real production channel, 2 reports per second

Real production channel at 2 reports per second: ORU^R01 with a 31 KB PDF, validation, remote SOAP signature, token and multipart upload to the document repository, update for repeated reports, MLLP notification to the departmental system and an ACK for every outcome. 57,600 reports per run, with 1,152 malformed and 224 injected faults answered with AE.

MeasureSTANDARD traceDETAILED trace
Reports lost00
ACK AA / AE56,224 / 1,37656,224 / 1,376
ACK median / 95th / 99th28.8 / 37.0 / 42.2 ms30.0 / 38.9 / 45.6 ms
Maximum1,647 ms in the first minute, 111 ms afterwards1,554 ms in the first minute, 142 ms afterwards
Database over the last four hours1.36 – 1.80 GiB6.40 – 8.50 GiB
Space per retained report50.6 KiB237.7 KiB
Chart of database size over eight hours of continuous load, Aurora 2.5.1
Database size over eight hours: hourly partitions and four-hour retention keep the database within a fixed band.
Chart of 99th percentile ACK latency per minute over eight hours, Aurora 2.5.1
99th percentile ACK latency per minute: no drift over eight hours.

Realistic reporting channel: 2.5.0 and 2.5.1

Three correlated channels (relay to the departmental system, reporting with remote PAdES signature, publication to a document repository with a queue of 5 per minute), 0.10 reports per second for eight hours with 1, 3 and 5 MiB PDFs, controlled errors and injected faults.

MeasureAurora 2.5.0Aurora 2.5.1
Reports / signed / published2,880 / 2,704 / 2,7002,880 / 2,704 / 2,700
ORU report with signature, median / 95th / 99th437 / 1,712 / 1,819 ms445 / 1,749 / 1,836 ms
ORM orders28.8 / 36.4 / 42.0 ms28.6 / 36.0 / 42.0 ms
ADT patient messages30.8 / 39.3 / 58.7 ms30.8 / 38.9 / 50.4 ms
Average CPU / average heap6.2% / 926 MiB6.5% / 929 MiB
Full GC00

The report time includes synchronous conversion, signature and delivery and depends on the PDF size. In the 2.5.1 run a single ADT ACK took 5.2 s (1 of 2,880), with no garbage collector pause above 22 ms.

03 · September

Compared with the September edition

The bench has changed since September (host, VM operating system, Java 25, direct network, new measurement tools): absolute figures from the two editions are not directly comparable. On the real production channel we separated environment and code with four short 15-minute runs on the same bench: the Aurora 2.0 pre-release build used in the September tests, the 2.0.0 release build (9 September), 2.5.0 and 2.5.1.

Real channel ACK, 15-minute runsAurora 2.0, September buildAurora 2.0.0Aurora 2.5.0Aurora 2.5.1
Median30.9 ms31.2 ms34.4 ms35.0 ms
95th percentile51.0 ms52.8 ms61.2 ms64.9 ms
99th percentile73.8 ms78.2 ms93.0 ms99.8 ms
Chart of the stages of the real channel chain, medians by Aurora version
Stages of the real channel chain, medians by version.
  • The build used in September, which then measured 16.2 ms in the eight-hour run, measures 30.9 ms on the current bench (31.2 ms for the 2.0.0 release build): the distance from the first edition depends mostly on the environment.
  • On the same bench 2.5.x needs about 3-4 ms more than the 2.0 builds at the median (10-13%); the largest part, about 2 ms, is in closing the message before the ACK, the rest over the other stages, with increases of at most 1 ms each.
  • Space per report is in line with September for STANDARD (50.6 vs 48.8 KiB, September recalculated with the same method) and higher for DETAILED (237.7 vs 99.4 KiB): worth considering when sizing a DETAILED installation.

04 · Limitations

Stated limitations

  • Mirth Connect “as installed”: 2 GiB heap versus 4 GiB for Aurora, no tuning, storage without pruning; channels rebuilt to be equivalent, with parsing, persistence and ACK handling native to each engine. With an optimised configuration Mirth's results may improve.
  • Closed loop: each connection waits for the ACK before the next message, throughput depends on latency and latencies are not corrected for the waiting effect (coordinated omission).
  • Short steps with a freshly started JVM: absolute values reflect the cold start, the comparison between engines does not.
  • In the fastest steps the generator used with Aurora reaches about two thirds of one vCPU: Aurora's peaks may be partly limited by the bench.
  • Downstream systems simulated over local HTTP without TLS (with a fixed 20 ms delay per call on the realistic channel); a single virtualised host: absolute values indicative, relative comparisons reliable.
Full sweep data: Show all 48 groups (engine, channel, connections)
EngineChannelConn.msg/sMedian95th99thCPU %RSS MiB
Aurora 2.5.1Simple ADT14821.792.896.7130.0536
Aurora 2.5.1Simple ADT41,5492.233.737.3750.9714
Aurora 2.5.1Simple ADT162,9414.479.5615.4299.0927
Aurora 2.5.1Simple ADT323,0208.4121.4133.54126.71,006
Aurora 2.5.1Simple ORU13482.363.807.4431.6533
Aurora 2.5.1Simple ORU41,1372.934.968.3060.6865
Aurora 2.5.1Simple ORU161,7737.4914.8021.51124.71,104
Aurora 2.5.1Simple ORU321,70715.8936.4348.53127.41,149
Aurora 2.5.1Clinical ADT11984.138.7018.4562.4680
Aurora 2.5.1Clinical ADT47164.759.4815.85105.8868
Aurora 2.5.1Clinical ADT161,07412.8228.7340.71186.41,180
Aurora 2.5.1Clinical ADT321,06425.9562.9387.57218.61,213
Aurora 2.5.1Clinical ORU11574.9611.0422.1852.1724
Aurora 2.5.1Clinical ORU46115.4010.4916.35101.8969
Aurora 2.5.1Clinical ORU1691515.0032.1245.65175.01,222
Aurora 2.5.1Clinical ORU3288931.1271.3794.03173.81,291
Mirth RAWSimple ADT11396.709.8213.0333.0590
Mirth RAWSimple ADT44338.6812.6817.4163.0762
Mirth RAWSimple ADT1681118.4629.3636.71108.1830
Mirth RAWSimple ADT3267146.0470.1785.23147.8827
Mirth RAWSimple ORU11098.4311.8615.0831.8726
Mirth RAWSimple ORU433311.1016.2920.1878.5914
Mirth RAWSimple ORU1662024.3737.2045.61140.7830
Mirth RAWSimple ORU3252558.7687.80104.94161.5930
Mirth RAWClinical ADT1959.8913.6617.4530.0640
Mirth RAWClinical ADT430012.6317.7821.7857.7799
Mirth RAWClinical ADT1645533.8451.7061.4398.2753
Mirth RAWClinical ADT3238979.05123.13146.37133.4865
Mirth RAWClinical ORU17811.8716.3419.4432.8641
Mirth RAWClinical ORU424215.5021.6926.1877.1823
Mirth RAWClinical ORU1638040.3460.4771.20122.1947
Mirth RAWClinical ORU3237183.45122.25144.83160.51,176
Mirth PRODUCTIONSimple ADT18611.0615.4618.6924.1745
Mirth PRODUCTIONSimple ADT425914.5820.7725.3551.9922
Mirth PRODUCTIONSimple ADT1648431.0348.3660.4889.6878
Mirth PRODUCTIONSimple ADT3246366.7399.93117.28129.1993
Mirth PRODUCTIONSimple ORU17013.3218.4521.9727.3593
Mirth PRODUCTIONSimple ORU420818.0825.5131.8167.4876
Mirth PRODUCTIONSimple ORU1638839.1858.6271.00103.8929
Mirth PRODUCTIONSimple ORU3237282.69122.55142.75156.9851
Mirth PRODUCTIONClinical ADT15218.4724.7829.4824.9602
Mirth PRODUCTIONClinical ADT414925.4734.7245.5062.0663
Mirth PRODUCTIONClinical ADT1624762.8288.52106.8787.2925
Mirth PRODUCTIONClinical ADT32212144.76225.20273.91115.5934
Mirth PRODUCTIONClinical ORU14620.6127.2833.2531.3733
Mirth PRODUCTIONClinical ORU412829.6539.8251.9569.2842
Mirth PRODUCTIONClinical ORU1620874.50103.47119.26103.2815
Mirth PRODUCTIONClinical ORU32190161.17247.43300.08130.8809

Environment: server with two AMD EPYC 7313 and 94 GiB RAM; Debian 13 VM with 8 vCPU and PostgreSQL 16.15 for the system under test; generators and simulated destinations on a separate Rocky Linux 9.7 VM. Aurora 2.5.1 on Temurin 25.0.5, Mirth Connect 4.5.2 on Temurin 17. Aurora's system database is PostgreSQL 14 or later. Internal tests run by TECHEMV, without the involvement of independent third parties, using synthetic data only; raw data, scripts and channels available on request.

The previous edition (September 2026, Aurora 2.0) remains available: September report (PDF, Italian).

Mirth Connect is a trademark of NextGen Healthcare, Inc., which did not take part in these tests and has not endorsed them; the name is used only to identify the product compared. Aurora Integration Engine is a product of TECHEMV SRL.

Evaluating an integration engine, or moving off Mirth Connect?

Bring your projects: the demo runs on your own cases.