# Testnet process-memory observations — September 2026 This note records the source and scope of the process-RSS figures quoted on the Quaxar website. It is a transcription of preserved acceptance artifacts, not a universal benchmark or a claim about production readiness. ## Scope - Environment: one 32 GiB Linux host connected to public XRPL Testnet. - Workload: non-validator synchronization and bounded `full` observation windows with NuDB. Measurements were made on 10–11 September 2026 UTC. - Sampling: process resident set size (RSS), reported in KiB by Linux `/proc//status` (`VmRSS`) or `ps -o rss`, then converted using 1 GiB = 1,048,576 KiB. - Comparison boundary: profile, synchronization peak, and settled `full` RSS are reported separately. These observations do not predict another host, ledger range, storage device, build, network condition, or workload. ## Recorded results | Runtime | Profile and phase | Window | Preserved result | | --- | --- | --- | --- | | Quaxar | Medium, bounded `full` window | 20 samples at 30-second intervals, 2026-09-10 19:55:07–20:04:32 UTC | 3,875,360–4,415,552 KiB (3.70–4.21 GiB); 4,018,600.2 KiB (3.8324 GiB) average; 179 validated ledgers advanced; zero restarts | | Quaxar | Huge, early post-sync `full` window | 20 samples at 30-second intervals, 2026-09-11 00:17:26–00:26:56 UTC | 6,332,620–6,872,396 KiB (6.04–6.55 GiB); 6,501,367.4 KiB (6.2002 GiB) average; 183 validated ledgers advanced; zero restarts | | Quaxar | Huge, synchronization run | Preserved summary high-water plus 15-second interval samples | Summary peak: 12,619,464 KiB (12.0349 GiB); maximum raw interval sample: 12,175,000 KiB (11.6109 GiB) | | `rippled` 3.3.0 | Medium, late `full` comparator | Preserved same-host acceptance comparator | 12.60 GiB | The profile labels are not inferred from directory names. The medium run used configuration SHA-256 `92346c039d81da9bc7e9a71a757e6723405dd0f6ed4d551af259c5aa9d8ecb04`; its preserved `server_info` reports `node_size=medium`. The huge run used configuration SHA-256 `1203b4c9cade993dea0581f1c9359505d8ceb9a9e4caab7bf6c03aa08267d30b`; each raw settled-window sample reports `node_size=huge`. The medium Quaxar window also required `full`, active service state, one stable PID, non-validator identity, zero restarts, and continuing validated-ledger advancement. The huge 10-minute window required `node_size=huge`, `full`, validated age no greater than 15 seconds, at least five peers, one stable PID, and zero restarts. The 12.0349 GiB field is the peak recorded by the preserved experiment summary. The raw 15-second synchronization series tops out at 11.6109 GiB, so the public page exposes both rather than presenting the summary high-water as an interval sample or a settled `full` average. ## Later live rechecks A read-only recheck on 11 September 2026 sampled a newer huge-profile binary 10 times at five-second intervals from 09:27:54 to 09:28:39 UTC. It remained `full`, used one stable PID, advanced 14 validated ledgers, and measured 9.5836–9.6013 GiB RSS, averaging 9.5928 GiB. `/proc//status` reported a 13.3967 GiB `VmHWM` since that process restarted at 08:13:41 UTC. A further read-only recheck on 11 September 2026, 11:56:59–11:57:59 UTC, sampled the same running service three times over 60 seconds. It reported `node_size=huge`, `server_state=full`, build version 0.8.0, one stable PID 705728, zero restarts, 21 peers, validated age 2–3 seconds, and advanced 18 validated ledgers (20669100 → 20669118). `VmRSS` ranged 7,635,700–7,765,708 KiB (7.28–7.41 GiB) and `VmHWM` was 14,047,480 KiB (13.3967 GiB) since the 08:13:41 UTC restart. Configuration SHA-256 was `1203b4c9cade993dea0581f1c9359505d8ceb9a9e4caab7bf6c03aa08267d30b` with `ledger_history=256` and `[node_db] online_delete=512`; the binary was `/opt/quaxar/bin/quaxar.ae07cd6-offercross-6cec3d2c`, SHA-256 `6cec3d2ca6c8b7c593f39556a22a43b828c0b18b930c73cbfe6c61d7277adcf1`. This is the current fully synced huge-profile reference used on the public page. The 7.28–7.41 GiB figure is a 60-second RSS range, not an average, and 13.40 GiB is the process high-water mark, not current RSS. This later point is not inserted into the historical comparison as if it used the same build, runtime age, or sampling gate. It demonstrates that the earlier 6.2002 GiB huge window was neither a current value nor proof of a long-run plateau. The `rippled` comparator is associated with `xrpld version 3.3.0`, commit `00a178fb92ca49521b937ae1a99d863765ea8a90`. The same pre-cutover bundle also contains an earlier 9.36 GiB point observation after 29 minutes; the website uses the preserved 12.60 GiB late-`full` comparator, not that earlier point. ## Artifact record The acceptance artifacts remain on the audited Testnet host. Paths and SHA-256 digests are included so the transcription can be checked against the source bundle without publishing host credentials or operational configuration. | Artifact | SHA-256 | | --- | --- | | `quaxar-memory-followup-20260910T184657Z/FINAL_RUNTIME_ACCEPTANCE_OK.json` | `272f95c5917fa05f3bfe595e051c1537a0a0cfc7e6de15af37cf2a33d04be497` | | `quaxar-huge-nudb-reset-20260910T225506Z/FULL_SUMMARY.json` | `749c720ccbffe85f82495ceddfe21f1e767154a2ac69a83fe9cad37b3b6d7a1c` | | `quaxar-huge-nudb-reset-20260910T225506Z/MEDIUM_COMPARISON.json` | `68b48a3d87f4a1513a2977fcafb3e88a0d0149609b162dc4b9495cc2af76aa01` | | `memory-layout-20260910T155719Z/RUNTIME_ACCEPTANCE_OK.json` | `2d4bd73d3c3c6d88399f125468d9b21e0c876572614905b315bbf5bfd5c9bf9c` | | `pre-quaxar-20260910T005002Z/rippled-version.txt` | `3d0d28e37fb4a216847242d5fe7e1b3dbccb71f9999249911d768b8550ec35b0` | The accepted medium Quaxar binary SHA-256 was `10841c1ec51e45119006d9e174fb7dff53ed06ecccd59cc35c0abb6bbf534c7f`; its source head was `a36c6d7405a916f4444d60b9b6781c8f77cbb4f9`. Exact figures are rounded to two decimals on the public page.