After Tor 0.4.9.15: HSDir refusals and continuing BEGIN_DIR stream pressure

By 1AEO Team · November 9, 2026 · Updated November 14, 2026

Version scope: Tor 0.4.9.15. Instrumented: October 30–November 5; stock: October 29–November 5, 2026, Coordinated Universal Time (UTC). Custom patch: metrics and hardening. HSDir: hidden-service directory.

Definitions#

A circuit is a Tor path carrying many streams (logical connections). A custom patch adds metrics that count streams closed while waiting for HTTP. Stock Tor’s tor_relay_congestion_control_total{state="cc_circuits",action="circs_closed"} counts closure marks only for circuits with congestion-control state.

HTTP-wait-dominated windows have at least 80% of classified streams closed while waiting for HTTP. HSDir GET-largest windows have GET requests as their largest classification, not necessarily a majority.

Relay labels are anonymized.

Takeaways#

1. Two high-rate directory-stream patterns were observed on Tor 0.4.9.15. Active HTTP-wait-dominated windows averaged about 50,000 admissions/s with few defense events: 208 HSDir request-limit refusals/s and 5 BEGIN_DIR rate-limit events/s. Active HSDir GET-largest windows averaged 1,583 HSDir request-limit refusals/s. The HSDir limit helps; substantial stream pressure remains.

2. Tor 0.4.9.15’s HSDir request limit did not prevent 100% main-thread CPU. Stock 0.4.9.15 Relay LF sustained about 56,000 additional admissions/s and 474 Mbit/s more read bandwidth than before the high-rate surge, for about 45 minutes.

3. Tor 0.4.9.15’s HSDir request limit refused half of HSDir GET requests across 206 HSDir GET-largest windows on 118 relays. Relay BP averaged 7,962 HSDir GET requests/s.

4. Later BEGIN_DIR cell positions within a circuit were more common in HTTP-wait-dominated windows. Positions 101+ held most cells in 111 HTTP-wait-dominated windows on 91 relays, versus no GET-largest window.

1. Two high-rate directory-stream patterns were observed on Tor 0.4.9.15#

Active groups: 110 HTTP-wait-dominated windows on 93 relays and 117 HSDir GET-largest windows on 82 relays. HSDir GET-largest averaged 3,170 GET requests/s and <1 BEGIN_DIR rate-limit events/s. Lifetime-limit refusals were zero in both groups.

Time-weighted average stream activity, DoS defense counter events per second, whole-relay traffic and main-thread CPU across active windows from 93 HTTP-wait group and 82 GET-largest-group relays. — **Footnote:** Active comparisons require matching active/observed spans and at least 90% CPU coverage, without selecting CPU values. Stream admission precedes HTTP parsing; streams waiting for HTTP do not reach the HSDir request check. Full groups remain in the supporting data. Streams use processing and memory without needing a socket each. Tor .15 includes the connection-list indexing fix. October 30–November 5, 2026 study, Tor 0.4.9.15 with a custom patch; qualifying pattern windows end November 4. Weighting follows Methods. These are window averages, not peaks or fleet totals. The 93 and 82 relay counts are distinct within each group; relays can contribute multiple windows. The DoS (denial-of-service) row shows how often defense counters incremented, averaged per second. HSDir and lifetime-limit refusals count enforced refusals; BEGIN_DIR rate-limit events can be recorded without blocking. Whole-relay traffic includes legitimate and forwarded traffic. CPU is observed main-thread use, not request-type attribution; 100% means one occupied core. Mbit/s means megabits per second. Full-size figure.

The same time-weighted averages as the chart.
Active window group	Windows / distinct relays	Admitted BEGIN_DIR mean/s	Streams closed while waiting for HTTP (per second)	Read / written Mbit/s	Main-thread CPU (% of one core)
HTTP-wait-dominated	110 / 93	50,238	49,418	446 / 216	98
HSDir GET-largest	117 / 82	6,361	1,365	104 / 77	77

2. Tor 0.4.9.15’s HSDir request limit did not prevent 100% main-thread CPU#

Versus LF’s baseline: +56,145 admitted streams/s, +474 Mbit/s read and +217 Mbit/s written; CPU averaged 100%.

CPU and both traffic directions rose and fell on 24 of 28 qualifying relays, including 23 besides LF, during October 29–November 4. LF runs stock Tor 0.4.9.15.

Relay LF on stock Tor 0.4.9.15: the highlighted late high-load period sustained about 56,000 additional admitted BEGIN_DIR streams/s and 474 Mbit/s more read traffic than before; main-thread CPU averaged 100%. Original timelines show the rise and fall. — **Footnote:** Late high-load comparison below; fixed-phase averages and comparable-relay evidence remain in supporting data. November 4, 2026, approximate UTC. The darker late high-load band is compared with the fixed ten-minute pre-surge sample; the band was selected from admissions alone. Main-thread CPU averaged 100% after rounding; the same period recorded 83 HSDir request-limit refusals/s. Steps and hatched gaps follow Methods. Outlined ten-minute samples still define the separate 24-of-28 resource comparison. The rise-and-fall pattern recurred across relays; magnitudes vary. The same process/build and sampled flags were retained. Whole-relay traffic includes other and forwarded traffic; stock LF cannot identify HTTP-wait classifications. Full-size figure.

Before is the fixed ten-minute pre-surge sample. High load is the longest consecutive run of one-minute admission averages above 50,000 streams/s; its public duration rounds to about 45 minutes. Values are time-weighted means; differences use unrounded values. CPU changes use percentage points (pp).
Metric	Before	High load	Added
Admitted BEGIN_DIR streams/s	197	56,342	56,145
Whole-relay read (Mbit/s)	36	510	474
Whole-relay written (Mbit/s)	33	250	217
Main-thread CPU (% of one core)	46	100	54 pp

3. Tor 0.4.9.15’s HSDir request limit refused half of HSDir GET requests#

“One HSDir request per circuit” shares one allowance between GET and POST. The second request is refused before descriptor lookup and closes its circuit. GET plus POST refusals match its counter in 382 complete instrumented windows across 129 relays, October 30–November 5, 2026 (UTC). GET totals include refusals.

Comparable high-load HTTP-wait and HSDir GET profiles appeared on 92 and 15 other relays, respectively.

Stream activity, HSDir request-limit refusals, traffic and main-thread CPU for Relays KI and BP. Comparable high-load HTTP-wait and GET profiles occurred on 92 and 15 other relays, respectively. — **Footnote:** Key measurements for these examples are in the Relay KI data and Relay BP data. Relay KI: 01 Nov 2026 16:00–18:15 UTC (approx.); Relay BP: 01 Nov 2026 00:45–01:15 UTC (approx.). These selected windows cover different clock spans. Profile matching uses the October 30–November 5 study: mean admissions of at least 10,000/s, mean main-thread CPU of at least 90%, at least 90% CPU coverage, and matching active/observed spans. HTTP-wait profiles require at least 80% of classified streams to have closed while waiting for HTTP; GET profiles require HSDir GET to be the largest category with 45–55% of GET requests refused. Other-relay counts exclude all observations from the corresponding example relay; 11 other relays qualify for both profiles, so counts are not additive. These are comparable category/load profiles, not identical timelines or bandwidth behavior. Steps and shaded gaps follow Methods; grey marks missing CPU at the edges. Whole-relay traffic includes forwarding; CPU is main-thread use, not request-type attribution. Scales match between relays. Full-size figure.

206 GET-largest windows; request-type refusals reconcile exactly with the Tor 0.4.9.15 HSDir request-limit counter.
GET-largest population	Count	Meaning
HSDir GET requests	7,650,105,933	HSDir GET requests in these windows
HSDir GET requests refused	3,820,044,729	50% of HSDir GET requests
HSDir POST requests refused	87,693	Separate from GET requests
Tor 0.4.9.15 HSDir request-limit refusals	3,820,132,422	GET refusals + POST refusals

4. Later BEGIN_DIR cell positions within a circuit were more common in HTTP-wait-dominated windows#

October 30–November 4, 2026 UTC: positions 101+ held a majority in 111 HTTP-wait-dominated windows on 91 relays; positions 1–10 held a majority in all 206 GET windows on 118 relays.

Shares of counted BEGIN_DIR cells by position within a circuit, comparing HTTP-wait-dominated and HSDir GET-largest windows. — **Footnote:** Both bars use the window groups defined above. Their stream-category mixes remain in the HTTP-wait group and GET-largest-group supporting tables. October 30–November 4, 2026. Bars pool all 149 HTTP-wait-dominated windows (94 relays) and 206 GET-largest windows (118 relays). The subtitle counts windows with more than half their BEGIN_DIR cells in the specified position range, using unrounded shares. The preceding chart’s 10,000/s and CPU filters are not applied here. BEGIN_DIR cell position within a circuit counts only BEGIN_DIR cells over that circuit’s lifetime, before admission; stream closure does not reset it. It is not a path-hop number. Bands are not rate or policy settings. Shares count cells, not circuits. Full-size figure.

Each percentage uses all counted BEGIN_DIR cells in its own window group as the denominator; rows compare the same BEGIN_DIR cell position band within a circuit.
BEGIN_DIR cell position	HTTP-wait group share	GET-largest share
1	<1%	25%
2–10	4%	75%
11–100	21%	<1%
101–1,000	60%	<1%
1,001+	14%	<1%

Methods#

October 30–November 5: the full pattern groups contain 355 of 385 selected instrumented windows and share 86 relays. Windows were selected when BEGIN_DIR stream/cell or individual defense-counter rates reached 500/s. All 9 windows overlapping November 5 were Other/mixed.

Rates divide combined counts or bytes by combined observation time; longer windows contribute more. CPU averages weight its own measured intervals. Timeline steps and peaks use roughly one-minute averages. Bandwidth covers all relay traffic. Public times use 15-minute rounding. Grey hatching marks unavailable measurements, not zero traffic or CPU use.

HTTP-wait closure classifications count a previously unclassified linked directory stream once, with empty input at cleanup or on a closing-stream/circuit path. They identify neither closure cause nor waiting duration. Effective BEGIN_DIR rate/lifetime settings were not verified throughout. These observations do not measure upgrade savings; the extended period also includes Tor’s November 4 report of directory load from older-client restart loops.

KI and BP in the instrumented relay population#

The labeled KI and BP observations are the examples used above. This separate comparison uses the same eligibility rules and axes as the stock Tor scatter below.

Mean admitted BEGIN_DIR streams and main-thread CPU across 299 eligible custom-patch observations on 128 relays, with the exact Relay KI and Relay BP example windows labeled. — **Footnote:** October 30–November 5, 2026. Tor 0.4.9.15 with a custom patch adding metrics and hardening. Each point is an observation selected using the rate rule in Methods with complete admission counts and at least 90% CPU measurement coverage, matching the stock scatter’s eligibility. No request-category or CPU-utilization threshold selects these points. Rates are whole-window means. Windows can include lower-activity portions, and relays can recur. Labels mark the exact KI and BP examples above. Matching axes aid comparison, but the build populations contain different relays and workloads; this is not a controlled patch-performance test. Full-size figure.

Exact selected windows from the earlier findings; repeated observations of the same relay remain separate points.
Labeled example · approximate UTC	Mean admitted BEGIN_DIR streams/s	Main-thread CPU (% of one core)
Relay KI · 01 Nov 2026 16:00–01 Nov 18:15 UTC	62,104	100
Relay BP · 01 Nov 2026 00:45–01 Nov 01:15 UTC	15,921	94

High pressure also appeared on stock Tor 0.4.9.15#

Among active stock observations with qualified CPU and admission measurements, 118 windows on 99 relays averaged at least 10,000 admitted streams/s and at least 90% main-thread CPU. Relay JA averaged 60,833 admissions/s. Stock counters do not identify streams closed while waiting for HTTP.

Admitted BEGIN_DIR streams, whole-relay traffic and main-thread CPU over time for stock Relay JA and instrumented Relay KI. — **Footnote:** Relay JA is included in the highlighted stock table below; Relay KI’s measurements remain in the supporting data. Relay JA: 1 Nov 2026 12:30–14:45 UTC; Relay KI: 1 Nov 2026 16:00–18:15 UTC (approx.). JA runs stock Tor 0.4.9.15; KI uses a custom patch adding metrics and hardening changes. These are different relays and clock periods. Both columns show the same measurements and scales. Averaging, time rounding and shaded gaps follow Methods; grey hatching marks missing CPU measurements. CPU is main-thread use as a percentage of one core. Full-size figure.

The four highlighted stock observations. Rates are whole-window means; CPU uses supported measurement time.
Highlighted stock relay · approximate UTC	Admitted BEGIN_DIR/s	Read Mbit/s	Written Mbit/s	Main-thread CPU %
Relay JA · 01 Nov 2026 12:30–01 Nov 14:45 UTC	60,833	529	256	99
Relay II · 01 Nov 2026 12:30–01 Nov 15:00 UTC	60,191	524	254	99
Relay CK · 01 Nov 2026 12:30–01 Nov 15:00 UTC	46,460	408	197	96
Relay HU · 01 Nov 2026 12:30–01 Nov 15:00 UTC	35,895	311	153	94

Mean admitted BEGIN_DIR streams and main-thread CPU across 349 eligible stock observations on 153 relays, highlighting Relays JA, II, CK and HU. — **Footnote:** 29 October–5 November 2026. Each point is one stock observation with complete admission counts and at least 90% CPU measurement coverage. Relays can recur, and window durations differ. The four highlighted examples are listed in the table above. Rates are whole-window means. Eligibility and result counts use unrounded values. Selection and weighting follow Methods; these observations do not describe the whole network. Full-size figure.

Window counts are not counts of distinct attacks.
Stock sample	Count
Eligible observation windows	349
Distinct relays	153
Windows averaging at least 90% main-thread CPU	121

Supporting measurements#

Key evidence for the four findings, followed by compact summaries of all 355 full-window comparisons. The 27 Other/mixed and 3 incomplete windows are outside those comparisons. Unavailable measurements remain distinct from zero.

Streams closed while waiting for HTTP before .15

Both samples use an 80% HTTP-wait threshold, but the older custom patch has nine stream categories and the newer patch has ten, with an additional early-closing classification path. These separate samples show pre-existing activity, not a controlled upgrade effect. Historical totals use whole intervals with matching process and loaded-build observations.
Measurement	Tor 0.4.9.14 · custom patch	Tor 0.4.9.15 · custom patch
Windows / relays	38 / 37	149 / 94
Average measured period (minutes)	146	161
Streams closed while waiting for HTTP (per second)	3,730	23,368

A separate 85-window Tor 0.4.9.14 sample with a custom patch recorded zero GET or POST refusals. The retained before/after observations do not provide a closely matched upgrade comparison for HTTP-wait-dominated windows.

HTTP-wait stream pressure: full group and active subset

The subset requires matching active/observed spans and at least 90% CPU coverage. CPU is a percentage of one core.
Population	Windows / relays	Streams closed while waiting for HTTP (per second)	Mean window peak (streams/s)	Read / written Mbit/s	Main CPU %
Full HTTP-wait-dominated group	149 / 94	23,368	49,237	237 / 132	70
Active, well-covered subset	110 / 93	49,418	58,590	446 / 216	98

Relay KI · 01 Nov 2026 16:00–01 Nov 18:15 UTC

The same measured window; peaks are supported interval averages. The circuit row covers only circuits with congestion-control state.
Measurement	Total events	Mean/s	Observed peak/s
Admitted BEGIN_DIR streams	480,682,798	62,104	68,787
Streams closed while waiting for HTTP	473,076,653	61,121	68,664
HSDir GET requests	3,915,212	506	9,815
HSDir request-limit refusals	1,956,134	253	4,908
Congestion-control circuits marked for closure	9,347,132	1,208	5,172

This older circuit counter can include closures triggered by .15 limits and other causes, without identifying which cause applies. It is neither an all-circuit total nor linked to individual stream closures.

CPU and traffic are window means; connection/socket maxima are independently sampled. Whole-process CPU includes the main thread.
Resource or request handling	Measurement
HSDir GET requests refused	50%
Whole-relay read / written	558 / 263 Mbit/s
Main-thread / whole-process CPU	100 / 146% of one core
CPU measurement coverage	99%
Maximum directory connections waiting for HTTP	116,917
Maximum open sockets	9,100

An early-window heartbeat reported BEGIN_DIR rate checking enabled and the BEGIN_DIR lifetime limit disabled.

Relay BP · 01 Nov 2026 00:45–01 Nov 01:15 UTC

The same measured window; peaks are supported interval averages. The circuit row covers only circuits with congestion-control state.
Measurement	Total events	Mean/s	Observed peak/s
Admitted BEGIN_DIR streams	27,703,090	15,921	18,283
Streams closed while waiting for HTTP	6,386,101	3,670	4,172
HSDir GET requests	13,853,884	7,962	9,144
HSDir request-limit refusals	6,922,210	3,978	4,569
Congestion-control circuits marked for closure	7,027,989	4,039	4,650

CPU and traffic are window means; connection/socket maxima are independently sampled. Whole-process CPU includes the main thread.
Resource or request handling	Measurement
HSDir GET requests refused	50%
Whole-relay read / written	166 / 90 Mbit/s
Main-thread / whole-process CPU	94 / 197% of one core
CPU measurement coverage	97%
Maximum directory connections waiting for HTTP	693
Maximum open sockets	9,107

Relay KI’s open sockets fell from 9,100 to 8,157. Its memory readings also fell between the first and last ten minutes as the rate of streams closed while waiting for HTTP rose.

Whole-process memory medians from nine and eight distinct readings, respectively; MB = 1,000,000 bytes.
Relay KI block	Streams closed while waiting for HTTP (per second)	Median resident memory (MB)
First ten minutes	18,769	2,060
Last ten minutes	67,760	1,708

Stream-category mixes for the two window groups

HTTP-wait-dominated

The group’s full classification mix; these percentages do not describe BEGIN_DIR cell positions.
Stream classification	Share of classified streams
Streams closed while waiting for HTTP	98%
HSDir GET requests	<1%
Unread input discarded on a closing stream/circuit	<1%
Consensus GET requests	<1%
HSDir POST requests	<1%
Microdescriptor GET requests	<1%
Server / extra-info descriptor GET requests	<1%
Unparsed directory streams	<1%
Other GET requests	<1%
Other POST requests	0%

HSDir GET-largest

The group’s full classification mix; these percentages do not describe BEGIN_DIR cell positions.
Stream classification	Share of classified streams
HSDir GET requests	<50%
Unread input discarded on a closing stream/circuit	29%
Streams closed while waiting for HTTP	21%
Consensus GET requests	<1%
HSDir POST requests	<1%
Microdescriptor GET requests	<1%
Server / extra-info descriptor GET requests	<1%
Other GET requests	<1%
Unparsed directory streams	<1%
Other POST requests	0%

All comparison windows: key measurements

One row per window. Rates are whole-window means. GET refusal percentages use that window’s complete HSDir GET request-handling counts. CPU is a percentage of one core; coverage is the measured fraction of the window. Selected cell-position shares use all counted BEGIN_DIR cells; the three displayed ranges are not a complete partition. Times are approximate UTC.

HTTP-wait-dominated windows · 149 windows

Key measurements for every window in this comparison group.
Relay · approximate UTC	Admitted BEGIN_DIR/s	Streams closed while waiting for HTTP (per second)	HSDir GET/s	GET refused %	Read Mbit/s	Written Mbit/s	Main CPU %	CPU coverage %	Cells 1–10 %	Cells 101–1,000 %	Cells 1,001+ %
Relay BP · 04 Nov 2026 12:30–04 Nov 13:15 UTC	51,911	48,623	1,648	50	458	232	100	98	12	>50	3
Relay LS · 04 Nov 2026 12:30–04 Nov 13:00 UTC	49,188	48,970	109	50	450	210	96	98	2	70	12
Relay LA · 04 Nov 2026 10:15–04 Nov 12:15 UTC	36,647	35,702	482	50	333	164	98	99	4	56	31
Relay LN · 04 Nov 2026 08:45–04 Nov 11:30 UTC	62,240	61,820	218	50	569	265	100	99	2	64	23
Relay LB · 04 Nov 2026 07:00–04 Nov 08:00 UTC	48,884	42,325	3,285	50	445	230	100	99	15	62	9
Relay LX · 04 Nov 2026 06:45–04 Nov 08:00 UTC	44,271	36,199	4,042	50	401	213	100	99	23	44	3
Relay LU · 04 Nov 2026 06:15–04 Nov 08:00 UTC	32,800	32,040	398	50	324	160	99	99	3	51	39
Relay LA · 04 Nov 2026 06:15–04 Nov 08:00 UTC	33,346	32,558	412	50	325	161	99	99	4	59	19
Relay KV · 04 Nov 2026 03:45–04 Nov 04:30 UTC	1,415	1,413	1	47	13	7	15	33	7	32	<1
Relay KW · 04 Nov 2026 03:45–04 Nov 04:30 UTC	1,725	1,708	12	50	17	9	25	20	8	32	<1
Relay LG · 04 Nov 2026 03:30–04 Nov 04:15 UTC	900	872	19	50	9	6	18	41	11	31	<1
Relay MB · 04 Nov 2026 02:45–04 Nov 04:15 UTC	4,928	4,285	420	50	57	35	41	94	23	27	<1
Relay MB · 04 Nov 2026 00:30–04 Nov 02:45 UTC	50,674	50,052	320	50	437	218	100	99	5	58	8
Relay KV · 04 Nov 2026 00:30–04 Nov 03:45 UTC	56,197	55,849	182	50	489	240	100	>99	7	52	4
Relay KW · 04 Nov 2026 00:30–04 Nov 03:45 UTC	52,086	51,774	161	50	454	223	98	>99	6	56	6
Relay LP · 04 Nov 2026 00:15–04 Nov 03:30 UTC	59,484	59,311	92	49	536	252	99	99	2	63	24
Relay LG · 04 Nov 2026 00:15–04 Nov 03:30 UTC	56,701	56,258	231	50	513	244	99	>99	3	63	21
Relay LU · 04 Nov 2026 00:15–04 Nov 03:30 UTC	38,705	38,174	273	50	348	170	98	99	3	58	29
Relay LA · 04 Nov 2026 00:15–04 Nov 02:45 UTC	39,953	39,429	266	50	355	174	99	99	3	61	24
Relay DH · 03 Nov 2026 12:30–03 Nov 13:00 UTC	52,251	50,562	857	50	462	233	100	97	7	59	5
Relay AM · 03 Nov 2026 12:15–03 Nov 13:00 UTC	30,793	29,489	644	50	273	136	93	98	7	61	10
Relay S · 03 Nov 2026 11:30–03 Nov 13:00 UTC	50,851	49,038	943	50	468	232	99	99	6	65	11
Relay JT · 03 Nov 2026 10:00–03 Nov 11:00 UTC	47,903	47,330	298	50	417	208	100	97	7	>50	6
Relay AA · 03 Nov 2026 10:00–03 Nov 11:00 UTC	58,620	58,368	133	49	504	250	100	97	7	52	3
Relay FV · 03 Nov 2026 10:00–03 Nov 11:00 UTC	59,886	59,693	102	50	522	253	99	99	3	61	22
Relay AA · 03 Nov 2026 08:15–03 Nov 10:00 UTC	54,424	51,916	1,261	50	476	240	100	>99	10	51	4
Relay FV · 03 Nov 2026 08:00–03 Nov 09:45 UTC	56,852	56,009	433	50	498	245	99	99	5	61	11
Relay JT · 03 Nov 2026 08:00–03 Nov 10:00 UTC	43,295	41,975	689	50	386	194	100	99	9	51	5
Relay GW · 03 Nov 2026 06:45–03 Nov 10:00 UTC	224	213	7	50	20	19	27	75	13	30	0
Relay IS · 03 Nov 2026 06:15–03 Nov 08:00 UTC	46,615	46,314	151	50	405	198	96	>99	5	57	5
Relay GW · 03 Nov 2026 05:15–03 Nov 06:45 UTC	57,865	57,485	194	50	529	252	100	98	2	66	19
Relay JU · 02 Nov 2026 23:30–03 Nov 00:15 UTC	3,674	3,619	37	50	34	17	50	45	9	31	<1
Relay JU · 02 Nov 2026 21:00–02 Nov 23:30 UTC	56,061	55,466	298	50	512	242	100	99	2	63	25
Relay FQ · 02 Nov 2026 15:30–02 Nov 16:15 UTC	2,257	2,167	59	50	21	13	20	75	12	31	<1
Relay HA · 02 Nov 2026 14:45–02 Nov 17:45 UTC	42,577	42,006	287	50	395	183	95	99	2	59	29
Relay EO · 02 Nov 2026 13:30–02 Nov 16:00 UTC	47,858	46,237	817	50	419	205	99	99	4	55	33
Relay FQ · 02 Nov 2026 13:00–02 Nov 15:30 UTC	45,385	42,425	1,481	50	393	198	99	99	7	53	32
Relay FT · 02 Nov 2026 12:30–02 Nov 14:45 UTC	41,135	40,864	135	50	360	178	99	99	2	61	27
Relay FV · 02 Nov 2026 12:15–02 Nov 14:30 UTC	57,755	57,432	169	50	514	245	99	99	3	65	16
Relay BN · 02 Nov 2026 06:45–02 Nov 08:15 UTC	7,474	6,978	325	50	100	73	58	98	15	29	<1
Relay BN · 02 Nov 2026 06:00–02 Nov 06:45 UTC	54,463	53,441	512	50	490	238	99	98	4	65	13
Relay IA · 02 Nov 2026 02:45–02 Nov 03:30 UTC	4,051	3,968	55	48	53	40	31	94	9	31	<1
Relay EX · 02 Nov 2026 02:15–02 Nov 03:00 UTC	3,883	3,828	36	49	38	24	29	75	9	31	<1
Relay EX · 02 Nov 2026 00:30–02 Nov 02:15 UTC	56,528	56,221	162	49	496	242	100	99	6	53	4
Relay IA · 02 Nov 2026 00:30–02 Nov 02:45 UTC	55,251	53,721	774	50	492	241	100	99	7	59	6
Relay ED · 02 Nov 2026 00:15–02 Nov 02:15 UTC	56,169	55,843	170	50	514	240	100	99	2	62	26
Relay EH · 02 Nov 2026 00:15–02 Nov 02:45 UTC	56,395	56,084	161	50	509	241	99	99	2	66	18
Relay EX · 02 Nov 2026 00:00–02 Nov 00:15 UTC	44,716	40,710	2,170	50	413	218	100	<90	16	44	2
Relay GL · 01 Nov 2026 22:00–01 Nov 23:45 UTC	55,806	54,652	608	50	489	244	100	99	7	56	7
Relay GW · 01 Nov 2026 20:45–01 Nov 21:45 UTC	46,029	44,036	997	50	428	202	100	99	6	59	22
Relay KI · 01 Nov 2026 18:15–01 Nov 19:45 UTC	2,374	2,307	44	50	24	18	32	52	10	32	<1
Relay DE · 01 Nov 2026 18:15–01 Nov 20:45 UTC	44,774	44,591	95	50	398	191	97	99	4	64	7
Relay CT · 01 Nov 2026 17:45–01 Nov 20:45 UTC	57,205	56,551	344	49	496	245	100	99	7	51	6
Relay GT · 01 Nov 2026 17:45–01 Nov 19:00 UTC	3,741	3,203	351	50	47	36	41	98	24	27	<1
Relay KK · 01 Nov 2026 17:30–01 Nov 18:00 UTC	898	883	7	47	57	77	48	98	9	32	<1
Relay IA · 01 Nov 2026 17:30–01 Nov 18:15 UTC	1,953	1,651	199	49	47	43	51	97	25	26	<1
Relay BV · 01 Nov 2026 16:45–01 Nov 19:30 UTC	55,556	55,290	132	50	501	235	98	>99	2	66	19
Relay IA · 01 Nov 2026 16:15–01 Nov 17:30 UTC	35,245	33,279	987	50	311	155	94	99	8	60	7
Relay KI · 01 Nov 2026 16:00–01 Nov 18:15 UTC	62,104	61,121	506	50	558	263	100	99	3	65	20
Relay KK · 01 Nov 2026 16:00–01 Nov 17:30 UTC	39,368	38,486	452	50	347	173	98	98	4	63	22
Relay GT · 01 Nov 2026 16:00–01 Nov 17:30 UTC	48,339	47,894	230	50	438	209	99	>99	2	63	26
Relay HV · 01 Nov 2026 15:30–03 Nov 06:30 UTC	32	29	<1	48	143	160	70	>99	17	28	<1
Relay BW · 01 Nov 2026 15:30–03 Nov 06:30 UTC	124	121	2	29	35	38	42	98	10	31	<1
Relay GG · 01 Nov 2026 14:45–03 Nov 06:30 UTC	229	197	20	50	37	40	42	>99	23	26	<1
Relay FZ · 01 Nov 2026 14:00–01 Nov 15:30 UTC	55,462	54,634	412	50	503	236	98	98	4	66	11
Relay JP · 01 Nov 2026 13:30–01 Nov 15:30 UTC	36,936	36,805	66	49	321	157	96	99	4	61	8
Relay HV · 01 Nov 2026 12:30–01 Nov 15:30 UTC	42,782	42,284	254	50	368	182	99	>99	2	61	25
Relay BW · 01 Nov 2026 12:15–01 Nov 15:15 UTC	60,382	59,977	206	49	528	256	99	>99	3	66	15
Relay FK · 01 Nov 2026 12:15–01 Nov 14:45 UTC	58,941	58,468	248	50	534	255	100	99	2	63	25
Relay GG · 01 Nov 2026 12:15–01 Nov 14:45 UTC	57,593	57,264	169	50	508	245	98	99	2	71	11
Relay EG · 01 Nov 2026 06:00–01 Nov 08:00 UTC	61,381	60,819	294	50	546	262	100	99	2	68	19
Relay JM · 01 Nov 2026 06:00–01 Nov 08:00 UTC	51,894	51,511	198	50	454	223	99	>99	6	57	4
Relay KR · 01 Nov 2026 06:00–01 Nov 08:00 UTC	56,894	56,608	150	50	506	243	99	99	2	67	17
Relay HR · 01 Nov 2026 06:00–01 Nov 06:45 UTC	56,894	55,198	884	50	514	250	100	99	5	70	11
Relay CD · 01 Nov 2026 06:00–01 Nov 08:00 UTC	58,614	56,532	1,084	50	528	255	100	99	5	65	20
Relay GR · 01 Nov 2026 06:00–01 Nov 08:00 UTC	55,999	54,808	621	50	492	244	100	99	8	53	3
Relay HD · 01 Nov 2026 04:15–01 Nov 05:15 UTC	5,287	5,077	135	50	52	50	42	82	11	31	<1
Relay GA · 01 Nov 2026 03:45–01 Nov 05:15 UTC	6,996	6,375	404	50	64	42	46	71	15	55	<1
Relay BP · 01 Nov 2026 03:45–01 Nov 05:00 UTC	2,506	2,337	112	50	26	14	36	60	15	32	<1
Relay GO · 01 Nov 2026 03:45–01 Nov 04:15 UTC	690	680	5	42	20	34	24	98	8	32	<1
Relay DG · 01 Nov 2026 03:30–01 Nov 04:15 UTC	2,593	2,515	50	50	24	16	20	65	10	31	<1
Relay DH · 01 Nov 2026 03:30–01 Nov 04:15 UTC	2,386	2,374	8	50	22	18	42	26	7	32	<1
Relay HD · 01 Nov 2026 01:15–01 Nov 04:00 UTC	50,608	44,060	3,294	50	449	231	100	>99	17	46	4
Relay GO · 01 Nov 2026 01:15–01 Nov 03:30 UTC	55,744	54,255	760	50	507	239	100	99	4	61	26
Relay GA · 01 Nov 2026 01:15–01 Nov 03:45 UTC	57,081	55,611	763	50	507	247	99	99	4	64	20
Relay EQ · 01 Nov 2026 01:15–01 Nov 03:45 UTC	42,714	42,188	267	50	371	183	99	99	3	63	23
Relay BP · 01 Nov 2026 01:15–01 Nov 03:45 UTC	57,015	55,991	535	50	498	247	100	99	8	48	3
Relay DH · 01 Nov 2026 00:45–01 Nov 03:30 UTC	43,726	43,504	112	50	382	184	96	>99	3	65	6
Relay KC · 01 Nov 2026 00:45–01 Nov 01:15 UTC	47,140	38,453	4,339	50	432	225	100	97	20	59	5
Relay DG · 01 Nov 2026 00:45–01 Nov 03:30 UTC	36,839	36,482	174	50	317	157	95	99	4	63	7
Relay DH · 01 Nov 2026 00:15–01 Nov 00:30 UTC	31,685	29,842	912	50	283	143	94	93	9	61	5
Relay GA · 01 Nov 2026 00:15–01 Nov 00:30 UTC	39,686	38,658	513	50	357	173	91	93	5	70	8
Relay BP · 01 Nov 2026 00:15–01 Nov 00:30 UTC	36,914	32,788	2,040	50	334	170	96	94	15	53	2
Relay GB · 01 Nov 2026 00:15–01 Nov 00:30 UTC	41,491	40,730	377	50	374	179	92	93	4	69	5
Relay BS · 01 Nov 2026 00:15–01 Nov 00:30 UTC	42,395	41,747	343	50	390	185	94	93	4	72	6
Relay KU · 31 Oct 2026 23:15–31 Oct 23:45 UTC	30,296	26,774	1,831	50	288	142	92	98	14	62	11
Relay EW · 31 Oct 2026 23:15–31 Oct 23:30 UTC	32,682	28,837	1,966	50	312	154	94	93	14	61	8
Relay FC · 31 Oct 2026 23:15–31 Oct 23:45 UTC	56	52	7	3	17	15	13	74	17	46	30
Relay GY · 31 Oct 2026 22:15–31 Oct 23:15 UTC	56,681	56,562	59	49	501	238	98	99	2	70	11
Relay FC · 31 Oct 2026 22:00–31 Oct 23:15 UTC	55,532	54,312	618	50	505	240	100	98	4	70	15
Relay BO · 31 Oct 2026 21:45–31 Oct 22:45 UTC	55,356	55,244	56	49	504	232	98	97	2	68	12
Relay EL · 31 Oct 2026 20:30–31 Oct 21:15 UTC	946	942	2	5	27	30	24	97	7	33	<1
Relay DP · 31 Oct 2026 20:15–31 Oct 22:00 UTC	57,032	56,960	40	49	509	241	98	99	2	67	16
Relay JQ · 31 Oct 2026 18:30–31 Oct 20:30 UTC	55,792	55,662	64	49	498	235	98	99	2	68	16
Relay EL · 31 Oct 2026 18:15–31 Oct 20:30 UTC	53,501	53,282	106	49	469	226	99	99	4	60	6
Relay IV · 31 Oct 2026 18:00–31 Oct 19:00 UTC	4,183	4,054	83	49	43	33	38	66	10	32	<1
Relay IV · 31 Oct 2026 16:00–31 Oct 18:00 UTC	47,011	46,835	91	49	407	199	99	99	1	60	30
Relay AK · 31 Oct 2026 15:30–01 Nov 13:45 UTC	93	85	2	50	43	65	37	99	15	30	<1
Relay HS · 31 Oct 2026 14:45–01 Nov 13:30 UTC	38	36	<1	44	23	28	39	97	9	55	<1
Relay BZ · 31 Oct 2026 13:15–31 Oct 15:15 UTC	56,980	56,582	196	50	504	240	98	>99	3	69	11
Relay HS · 31 Oct 2026 13:15–31 Oct 14:45 UTC	58,125	57,860	136	50	507	245	100	99	6	>50	2
Relay HW · 31 Oct 2026 13:15–31 Oct 15:15 UTC	36,327	36,244	42	49	309	153	94	99	5	59	4
Relay AK · 31 Oct 2026 13:15–31 Oct 15:30 UTC	42,832	42,350	237	50	368	183	99	99	3	63	22
Relay HI · 31 Oct 2026 12:30–31 Oct 13:00 UTC	41,088	40,932	81	50	356	176	99	98	3	66	8
Relay BW · 31 Oct 2026 12:15–31 Oct 13:00 UTC	53,275	52,933	157	50	467	225	97	98	3	70	6
Relay BY · 31 Oct 2026 12:15–31 Oct 13:45 UTC	59,957	59,834	61	50	534	251	99	99	2	68	13
Relay IC · 31 Oct 2026 12:15–31 Oct 13:00 UTC	33,863	33,105	371	50	296	147	94	98	6	60	5
Relay JR · 31 Oct 2026 12:15–31 Oct 13:00 UTC	34,284	33,358	463	50	299	149	95	95	7	54	3
Relay BN · 31 Oct 2026 12:15–31 Oct 13:00 UTC	47,688	46,931	360	50	424	204	95	95	3	70	9
Relay CJ · 31 Oct 2026 12:15–31 Oct 13:00 UTC	53,606	52,976	299	50	471	228	98	96	4	70	7
Relay CL · 31 Oct 2026 10:45–31 Oct 12:15 UTC	56,254	56,093	79	49	494	240	99	99	3	68	10
Relay JZ · 31 Oct 2026 06:15–31 Oct 08:00 UTC	55,776	54,784	500	50	489	242	100	99	8	51	3
Relay DM · 31 Oct 2026 00:45–31 Oct 02:00 UTC	54,316	54,182	66	49	484	230	98	99	2	71	11
Relay KQ · 31 Oct 2026 00:30–31 Oct 02:30 UTC	44,553	44,502	26	49	380	188	99	99	2	67	17
Relay DJ · 31 Oct 2026 00:00–31 Oct 00:15 UTC	27,468	25,942	698	49	245	123	88	79	8	64	4
Relay JK · 30 Oct 2026 22:30–30 Oct 23:15 UTC	5,931	5,872	46	50	66	41	32	96	7	44	<1
Relay FU · 30 Oct 2026 22:00–30 Oct 23:30 UTC	35,944	35,493	219	50	312	154	95	99	5	64	5
Relay JK · 30 Oct 2026 22:00–30 Oct 22:15 UTC	30,096	28,066	989	50	274	134	93	89	9	66	5
Relay DI · 30 Oct 2026 22:00–30 Oct 23:30 UTC	57,352	57,168	98	49	532	244	99	98	3	69	10
Relay JK · 30 Oct 2026 21:30–30 Oct 22:00 UTC	6,409	5,764	423	50	75	47	60	97	19	29	<1
Relay JK · 30 Oct 2026 20:00–30 Oct 21:15 UTC	40,207	40,114	47	49	343	170	98	98	4	62	9
Relay HJ · 30 Oct 2026 15:45–30 Oct 16:00 UTC	45,917	43,570	1,147	50	403	202	100	91	7	68	5
Relay CM · 30 Oct 2026 15:15–30 Oct 15:45 UTC	796	786	6	49	10	10	51	3	8	33	<1
Relay KB · 30 Oct 2026 15:00–30 Oct 16:00 UTC	8,385	8,279	68	49	77	45	83	31	8	32	<1
Relay CH · 30 Oct 2026 14:45–30 Oct 16:00 UTC	518	513	2	40	15	19	18	<90	8	33	<1
Relay JF · 30 Oct 2026 14:45–30 Oct 16:00 UTC	981	975	3	43	18	22	27	39	7	34	<1
Relay CH · 30 Oct 2026 12:30–30 Oct 14:45 UTC	36,642	36,413	110	49	312	155	95	99	4	62	5
Relay CM · 30 Oct 2026 12:30–30 Oct 15:15 UTC	59,720	59,572	75	50	539	251	100	>99	2	64	23
Relay JF · 30 Oct 2026 12:30–30 Oct 14:45 UTC	42,157	42,049	54	49	366	177	97	99	5	54	2
Relay KB · 30 Oct 2026 12:15–30 Oct 15:00 UTC	58,705	58,425	140	50	519	248	99	99	4	64	13
Relay FI · 30 Oct 2026 12:15–31 Oct 03:00 UTC	124	121	1	26	9	11	14	88	9	32	<1
Relay FI · 30 Oct 2026 10:45–30 Oct 12:00 UTC	37,312	37,037	132	48	322	160	98	98	8	47	3
Relay FO · 30 Oct 2026 09:15–30 Oct 11:45 UTC	50,048	49,796	135	50	445	215	98	99	4	67	11
Relay IL · 30 Oct 2026 07:15–30 Oct 10:30 UTC	1,102	1,025	50	49	16	14	39	24	15	30	<1
Relay IL · 30 Oct 2026 05:15–30 Oct 07:00 UTC	55,201	54,990	114	48	492	239	99	>99	7	58	8
Relay D · 30 Oct 2026 05:15–30 Oct 10:15 UTC	2,706	2,644	39	50	41	31	36	71	9	32	<1
Relay IT · 30 Oct 2026 04:30–30 Oct 05:45 UTC	46,493	46,354	73	49	419	197	100	98	7	52	6
Relay D · 30 Oct 2026 03:15–30 Oct 05:00 UTC	48,582	48,404	99	47	419	207	100	98	11	25	<1
Relay CW · 30 Oct 2026 03:15–30 Oct 04:45 UTC	43,990	43,931	32	48	376	186	99	98	5	59	11

HSDir GET-largest windows · 206 windows

Key measurements for every window in this comparison group.
Relay · approximate UTC	Admitted BEGIN_DIR/s	Streams closed while waiting for HTTP (per second)	HSDir GET/s	GET refused %	Read Mbit/s	Written Mbit/s	Main CPU %	CPU coverage %	Cells 1–10 %	Cells 101–1,000 %	Cells 1,001+ %
Relay LS · 04 Nov 2026 11:00–04 Nov 12:15 UTC	4,514	1,047	2,257	50	81	57	69	98	>99	0	0
Relay BH · 04 Nov 2026 10:15–04 Nov 12:45 UTC	10,638	2,395	5,320	50	125	82	74	99	>99	0	0
Relay MB · 04 Nov 2026 09:15–04 Nov 12:45 UTC	8,175	1,555	4,087	50	151	113	86	>99	>99	<1	0
Relay LP · 04 Nov 2026 09:00–04 Nov 12:45 UTC	7,444	1,703	3,722	50	113	76	77	99	>99	0	0
Relay LB · 04 Nov 2026 09:00–04 Nov 12:15 UTC	9,507	2,197	4,754	50	123	71	81	>99	>99	<1	0
Relay LA · 04 Nov 2026 09:00–04 Nov 10:00 UTC	6,069	2,332	2,375	50	100	74	72	97	80	7	<1
Relay BP · 04 Nov 2026 08:15–04 Nov 12:30 UTC	10,107	2,347	5,054	50	128	73	85	>99	>99	0	0
Relay LN · 04 Nov 2026 08:15–04 Nov 08:45 UTC	18,818	3,512	9,409	50	206	136	99	98	>99	0	0
Relay LA · 04 Nov 2026 04:00–04 Nov 05:45 UTC	9,374	2,135	4,458	50	122	83	83	99	95	2	<1
Relay LS · 04 Nov 2026 03:45–04 Nov 08:00 UTC	3,923	893	1,962	50	67	48	63	>99	>99	0	0
Relay LU · 04 Nov 2026 03:30–04 Nov 06:15 UTC	5,722	1,524	2,654	50	125	100	74	97	93	2	<1
Relay LN · 04 Nov 2026 02:30–04 Nov 08:00 UTC	4,308	1,010	2,155	50	73	44	74	>99	>99	0	0
Relay BP · 04 Nov 2026 02:30–04 Nov 08:00 UTC	3,750	882	1,875	50	69	45	69	>99	>99	0	0
Relay LX · 04 Nov 2026 01:00–04 Nov 06:45 UTC	7,825	1,615	3,912	50	96	58	70	>99	>99	0	0
Relay LX · 04 Nov 2026 00:45–04 Nov 01:00 UTC	21,321	1,865	10,657	50	223	154	100	93	100	0	0
Relay LB · 04 Nov 2026 00:45–04 Nov 07:00 UTC	9,303	1,853	4,652	50	125	83	79	>99	>99	0	0
Relay MB · 04 Nov 2026 00:15–04 Nov 00:30 UTC	12,451	1,287	6,214	50	175	127	97	80	100	0	0
Relay LP · 04 Nov 2026 00:15–04 Nov 00:30 UTC	5,205	1,016	2,592	50	93	70	Unavailable	0	100	0	0
Relay KV · 04 Nov 2026 00:15–04 Nov 00:30 UTC	5,030	1,154	2,523	50	82	49	Unavailable	0	100	0	0
Relay LX · 04 Nov 2026 00:15–04 Nov 00:30 UTC	16,651	1,989	8,328	50	183	122	94	92	>99	0	0
Relay KW · 04 Nov 2026 00:15–04 Nov 00:30 UTC	2,915	671	1,469	50	56	39	Unavailable	0	>99	0	0
Relay LB · 04 Nov 2026 00:15–04 Nov 00:30 UTC	16,211	3,122	8,112	50	182	125	95	92	>99	0	0
Relay AL · 03 Nov 2026 12:15–03 Nov 13:00 UTC	29,461	5,555	12,228	50	283	182	100	98	83	12	<1
Relay DH · 03 Nov 2026 12:15–03 Nov 12:30 UTC	22,327	79	11,157	50	230	157	100	81	>99	0	0
Relay FV · 03 Nov 2026 12:15–03 Nov 12:30 UTC	826	197	415	50	23	19	29	78	>99	0	0
Relay JT · 03 Nov 2026 12:15–03 Nov 12:45 UTC	198	46	100	49	12	11	12	57	>99	0	0
Relay AA · 03 Nov 2026 12:15–03 Nov 14:30 UTC	36	8	18	46	9	9	12	30	>99	0	0
Relay BH · 03 Nov 2026 12:00–03 Nov 13:45 UTC	9,462	2,126	4,731	50	111	68	68	99	>99	0	0
Relay AL · 03 Nov 2026 11:45–03 Nov 12:15 UTC	4,243	1,002	2,122	50	65	40	67	98	>99	0	0
Relay AV · 03 Nov 2026 11:30–03 Nov 14:15 UTC	4,810	1,065	2,406	50	93	71	78	60	>99	0	0
Relay FQ · 03 Nov 2026 11:15–03 Nov 14:15 UTC	4,640	1,062	2,320	50	68	43	56	92	>99	0	0
Relay DH · 03 Nov 2026 11:00–03 Nov 12:15 UTC	5,033	1,191	2,518	50	88	57	75	98	>99	0	0
Relay BH · 03 Nov 2026 11:00–03 Nov 11:45 UTC	5,270	1,215	2,636	50	85	53	75	99	>99	0	0
Relay S · 03 Nov 2026 11:00–03 Nov 11:15 UTC	6,695	1,479	3,351	50	140	100	86	94	>99	0	0
Relay AV · 03 Nov 2026 10:00–03 Nov 11:30 UTC	6,524	1,511	3,263	50	108	72	83	99	>99	0	0
Relay AM · 03 Nov 2026 10:00–03 Nov 12:15 UTC	6,369	1,451	3,186	50	86	56	74	99	>99	0	0
Relay FQ · 03 Nov 2026 10:00–03 Nov 11:15 UTC	9,180	2,129	4,590	50	113	60	84	99	>99	0	0
Relay AL · 03 Nov 2026 10:00–03 Nov 11:30 UTC	14,701	2,390	7,350	50	168	119	89	>99	>99	0	0
Relay DH · 03 Nov 2026 10:00–03 Nov 11:00 UTC	14,439	3,347	7,220	50	179	108	93	99	>99	0	0
Relay AL · 03 Nov 2026 08:30–03 Nov 10:00 UTC	21,724	2,401	10,794	50	233	163	100	98	99	<1	<1
Relay AL · 03 Nov 2026 08:15–03 Nov 08:30 UTC	25,548	2,182	11,699	50	258	168	100	89	92	3	0
Relay JT · 03 Nov 2026 07:15–03 Nov 08:00 UTC	3,549	808	1,777	50	76	57	69	99	>99	0	0
Relay AV · 03 Nov 2026 06:45–03 Nov 09:45 UTC	10,745	2,374	5,373	50	157	107	93	>99	>99	0	0
Relay AV · 03 Nov 2026 06:30–03 Nov 06:45 UTC	24,960	2,671	11,168	50	254	167	100	93	90	4	<1
Relay FV · 03 Nov 2026 06:00–03 Nov 08:00 UTC	5,525	1,285	2,764	50	117	85	79	99	>99	0	0
Relay AA · 03 Nov 2026 06:00–03 Nov 08:00 UTC	6,381	1,476	3,192	50	98	61	79	99	>99	0	0
Relay AM · 03 Nov 2026 06:00–03 Nov 09:45 UTC	6,439	1,456	3,221	50	101	71	78	99	>99	0	0
Relay IS · 03 Nov 2026 06:00–03 Nov 06:15 UTC	5,960	1,337	2,983	50	96	63	82	93	>99	0	0
Relay FQ · 03 Nov 2026 04:45–03 Nov 10:00 UTC	6,077	1,422	3,039	50	88	52	76	99	>99	0	0
Relay AL · 03 Nov 2026 04:15–03 Nov 08:00 UTC	5,637	1,305	2,820	50	94	62	79	99	>99	<1	0
Relay DH · 03 Nov 2026 04:15–03 Nov 10:00 UTC	6,880	1,600	3,441	50	113	75	78	>99	>99	0	0
Relay GW · 03 Nov 2026 04:00–03 Nov 05:00 UTC	7,228	1,550	3,615	50	104	67	78	99	>99	0	0
Relay AV · 02 Nov 2026 22:15–03 Nov 06:30 UTC	8,307	1,640	4,154	50	140	101	87	>99	>99	<1	0
Relay JU · 02 Nov 2026 20:00–02 Nov 20:45 UTC	5,705	1,232	2,853	50	91	62	76	99	>99	0	0
Relay AV · 02 Nov 2026 16:15–02 Nov 22:00 UTC	5,813	1,249	2,899	50	114	91	83	>99	>99	<1	0
Relay EO · 02 Nov 2026 16:15–02 Nov 17:15 UTC	869	297	375	50	11	10	21	79	87	5	<1
Relay AV · 02 Nov 2026 15:00–02 Nov 16:00 UTC	4,287	987	2,144	50	112	97	79	99	>99	0	0
Relay HA · 02 Nov 2026 14:15–02 Nov 14:45 UTC	3,898	858	1,950	50	71	58	71	97	100	0	0
Relay EO · 02 Nov 2026 13:15–02 Nov 13:30 UTC	17,570	579	8,802	50	175	115	99	78	100	0	0
Relay FM · 02 Nov 2026 13:00–02 Nov 16:00 UTC	5,270	1,215	2,639	50	82	50	77	>99	>99	0	0
Relay FQ · 02 Nov 2026 12:45–02 Nov 13:00 UTC	18,740	1,732	8,776	50	188	122	100	94	94	3	<1
Relay EO · 02 Nov 2026 12:30–02 Nov 12:45 UTC	17,366	562	8,681	50	178	120	100	75	>99	0	0
Relay EO · 02 Nov 2026 12:15–02 Nov 12:30 UTC	13,573	2,014	6,788	50	147	100	96	93	>99	0	0
Relay FQ · 02 Nov 2026 12:15–02 Nov 12:30 UTC	13,758	1,200	6,879	50	152	104	93	93	>99	0	0
Relay FT · 02 Nov 2026 12:15–02 Nov 12:30 UTC	5,657	597	2,805	50	178	169	94	68	>99	0	0
Relay IM · 02 Nov 2026 11:30–03 Nov 06:30 UTC	369	21	184	50	103	107	99	3	>99	<1	0
Relay IM · 02 Nov 2026 10:00–02 Nov 11:15 UTC	6,094	700	3,046	50	113	98	91	30	>99	0	0
Relay IM · 02 Nov 2026 08:30–02 Nov 09:45 UTC	15,227	1,550	7,610	50	191	142	98	99	>99	0	0
Relay IM · 02 Nov 2026 08:30–02 Nov 08:45 UTC	26,471	331	13,223	50	271	186	100	25	100	0	0
Relay BN · 02 Nov 2026 04:15–02 Nov 05:45 UTC	9,291	1,904	4,645	50	118	77	83	99	>99	0	0
Relay IM · 02 Nov 2026 02:45–02 Nov 08:15 UTC	13,138	902	6,565	50	160	121	97	44	>99	0	0
Relay IM · 02 Nov 2026 02:45–02 Nov 03:00 UTC	9,525	544	4,759	50	130	103	99	51	>99	0	0
Relay IM · 02 Nov 2026 00:30–02 Nov 02:30 UTC	21,414	1,641	10,706	50	236	169	99	79	>99	0	0
Relay ED · 02 Nov 2026 00:00–02 Nov 00:15 UTC	4,621	1,083	2,312	50	83	68	71	92	>99	0	0
Relay EH · 02 Nov 2026 00:00–02 Nov 00:15 UTC	4,121	962	2,062	50	66	47	66	82	100	0	0
Relay EX · 01 Nov 2026 23:15–02 Nov 00:00 UTC	5,172	1,177	2,588	50	85	56	78	99	>99	0	0
Relay IM · 01 Nov 2026 22:30–02 Nov 00:30 UTC	25,185	934	12,589	50	261	181	98	31	>99	0	0
Relay IM · 01 Nov 2026 20:45–01 Nov 22:15 UTC	8,798	487	4,355	50	125	102	99	2	99	<1	0
Relay GW · 01 Nov 2026 19:15–01 Nov 20:30 UTC	5,567	1,084	2,785	50	100	76	79	98	>99	0	0
Relay IA · 01 Nov 2026 19:00–02 Nov 00:15 UTC	3,076	718	1,540	50	69	54	67	>99	>99	<1	0
Relay IM · 01 Nov 2026 17:30–01 Nov 20:30 UTC	5,697	575	2,762	50	105	92	Unavailable	0	97	<1	0
Relay CT · 01 Nov 2026 17:15–01 Nov 17:45 UTC	12,844	2,187	6,422	50	192	160	97	97	>99	0	0
Relay GL · 01 Nov 2026 16:30–01 Nov 22:00 UTC	6,580	1,080	3,289	50	154	136	89	99	>99	0	0
Relay DE · 01 Nov 2026 16:15–01 Nov 18:15 UTC	6,272	1,188	3,136	50	103	75	81	99	>99	0	0
Relay BV · 01 Nov 2026 16:15–01 Nov 16:45 UTC	3,157	740	1,579	50	64	56	63	94	>99	0	0
Relay CT · 01 Nov 2026 16:15–01 Nov 17:00 UTC	5,533	1,208	2,766	50	139	133	87	99	>99	0	0
Relay GL · 01 Nov 2026 16:15–01 Nov 16:30 UTC	24,284	534	12,045	50	244	163	100	89	>99	<1	0
Relay JP · 01 Nov 2026 15:45–03 Nov 06:30 UTC	136	53	54	50	31	33	33	>99	81	7	0
Relay IA · 01 Nov 2026 15:30–01 Nov 16:00 UTC	1,194	274	599	50	45	41	53	97	>99	0	0
Relay KK · 01 Nov 2026 14:15–01 Nov 15:45 UTC	3,375	734	1,686	50	153	160	92	>99	>99	0	0
Relay GT · 01 Nov 2026 14:15–01 Nov 16:00 UTC	9,272	1,786	4,637	50	129	98	87	99	>99	0	0
Relay GL · 01 Nov 2026 14:00–01 Nov 16:00 UTC	6,835	1,402	3,416	50	157	141	93	99	>99	<1	0
Relay KI · 01 Nov 2026 14:00–01 Nov 16:00 UTC	7,206	1,660	3,603	50	103	69	80	99	>99	0	0
Relay IM · 01 Nov 2026 13:15–01 Nov 17:15 UTC	13,745	1,387	6,869	50	190	147	Unavailable	0	>99	0	0
Relay FZ · 01 Nov 2026 13:15–01 Nov 14:00 UTC	4,571	1,027	2,286	50	92	75	68	98	>99	0	0
Relay JP · 01 Nov 2026 13:15–01 Nov 13:30 UTC	4,796	1,067	2,399	50	69	47	65	95	>99	0	0
Relay IM · 01 Nov 2026 13:00–01 Nov 13:15 UTC	27,940	308	13,961	50	295	199	Unavailable	0	100	0	0
Relay HV · 01 Nov 2026 12:15–01 Nov 12:30 UTC	1,364	208	647	50	178	198	Unavailable	0	>99	0	0
Relay IM · 01 Nov 2026 12:15–01 Nov 12:45 UTC	26,059	429	13,020	50	289	201	Unavailable	0	>99	0	0
Relay CT · 01 Nov 2026 10:15–01 Nov 12:15 UTC	4,439	1,021	2,220	50	136	129	89	99	>99	0	0
Relay IQ · 01 Nov 2026 10:00–01 Nov 13:00 UTC	4,149	962	2,077	50	65	44	61	>99	>99	<1	0
Relay BX · 01 Nov 2026 09:30–01 Nov 12:15 UTC	5,221	1,202	2,610	50	129	117	89	>99	>99	0	0
Relay KR · 01 Nov 2026 08:00–01 Nov 09:00 UTC	2,733	979	1,150	50	29	22	53	51	85	5	<1
Relay CD · 01 Nov 2026 08:00–01 Nov 09:00 UTC	2,164	840	869	50	26	22	51	50	82	6	<1
Relay IQ · 01 Nov 2026 07:45–01 Nov 09:00 UTC	5,395	1,241	2,699	50	73	47	70	99	>99	0	0
Relay IE · 01 Nov 2026 06:15–01 Nov 12:15 UTC	8,025	1,799	4,015	50	104	68	79	>99	>99	0	0
Relay ER · 01 Nov 2026 06:15–01 Nov 12:45 UTC	7,748	1,787	3,877	50	105	71	78	99	>99	0	0
Relay HR · 01 Nov 2026 05:30–01 Nov 06:00 UTC	860	202	429	50	72	73	62	97	>99	0	0
Relay JM · 01 Nov 2026 05:30–01 Nov 06:00 UTC	722	167	362	50	43	38	54	93	>99	0	0
Relay ER · 01 Nov 2026 04:15–01 Nov 06:00 UTC	5,695	1,310	2,851	50	81	59	71	99	>99	0	0
Relay GR · 01 Nov 2026 04:00–01 Nov 06:00 UTC	857	200	430	50	52	45	60	99	>99	0	0
Relay KR · 01 Nov 2026 03:45–01 Nov 05:45 UTC	619	143	310	50	40	39	57	99	>99	0	0
Relay CD · 01 Nov 2026 02:15–01 Nov 05:45 UTC	6,127	1,429	3,064	50	97	68	81	99	>99	0	0
Relay EG · 01 Nov 2026 02:00–01 Nov 06:00 UTC	7,425	1,719	3,713	50	106	63	86	99	>99	0	0
Relay HD · 01 Nov 2026 01:15–01 Nov 01:30 UTC	8,779	1,195	4,380	50	150	117	96	47	>99	0	0
Relay GO · 01 Nov 2026 01:15–01 Nov 01:30 UTC	4,302	991	2,158	50	105	90	86	51	>99	0	0
Relay EQ · 01 Nov 2026 00:45–01 Nov 01:00 UTC	14,539	1,095	7,152	50	182	138	99	96	98	<1	0
Relay GB · 01 Nov 2026 00:45–01 Nov 01:45 UTC	5,386	1,227	2,694	50	74	53	56	97	>99	<1	0
Relay GA · 01 Nov 2026 00:45–01 Nov 01:15 UTC	15,373	3,451	7,643	50	171	108	94	97	99	<1	<1
Relay BP · 01 Nov 2026 00:45–01 Nov 01:15 UTC	15,921	3,670	7,962	50	166	90	94	97	>99	<1	<1
Relay BS · 01 Nov 2026 00:45–01 Nov 02:15 UTC	6,535	1,350	3,268	50	98	78	61	98	>99	<1	<1
Relay EQ · 01 Nov 2026 00:30–01 Nov 00:45 UTC	10,960	975	5,484	50	178	145	96	90	>99	0	0
Relay JT · 01 Nov 2026 00:15–01 Nov 01:15 UTC	12,922	2,318	6,462	50	173	130	96	98	>99	0	0
Relay EQ · 01 Nov 2026 00:15–01 Nov 00:30 UTC	9,319	823	4,641	50	195	173	97	81	>99	0	0
Relay DG · 01 Nov 2026 00:15–01 Nov 00:45 UTC	9,914	1,409	4,959	50	119	83	89	97	>99	0	0
Relay DW · 01 Nov 2026 00:15–01 Nov 01:15 UTC	7,336	1,729	3,670	50	87	49	78	97	>99	0	0
Relay GX · 01 Nov 2026 00:15–01 Nov 02:00 UTC	3,099	727	1,550	50	58	47	58	99	>99	0	0
Relay DH · 01 Nov 2026 00:15–01 Nov 00:30 UTC	1,489	335	755	50	59	50	Unavailable	0	100	0	0
Relay KC · 01 Nov 2026 00:15–01 Nov 00:45 UTC	14,820	3,308	7,415	50	174	120	95	93	>99	<1	0
Relay EA · 01 Nov 2026 00:15–01 Nov 01:15 UTC	7,051	1,535	3,526	50	94	69	75	98	>99	0	0
Relay FN · 01 Nov 2026 00:15–01 Nov 01:15 UTC	6,133	1,449	3,068	50	89	59	76	98	>99	0	0
Relay BP · 01 Nov 2026 00:15–01 Nov 00:30 UTC	1,363	306	681	50	37	25	Unavailable	0	100	0	0
Relay BS · 01 Nov 2026 00:15–01 Nov 00:30 UTC	590	133	298	50	44	45	Unavailable	0	>99	0	0
Relay CG · 01 Nov 2026 00:00–01 Nov 00:45 UTC	3,304	1,138	1,414	50	54	40	45	82	86	5	<1
Relay ES · 01 Nov 2026 00:00–01 Nov 00:45 UTC	2,210	289	1,107	50	92	96	24	12	>99	<1	0
Relay KU · 01 Nov 2026 00:00–01 Nov 00:45 UTC	1,995	470	998	50	41	32	43	68	>99	<1	<1
Relay EW · 31 Oct 2026 23:45–01 Nov 00:45 UTC	4,103	880	2,055	50	71	52	56	97	>99	<1	0
Relay ES · 31 Oct 2026 23:30–31 Oct 23:45 UTC	13,672	1,454	6,824	50	187	140	Unavailable	0	>99	<1	0
Relay IY · 31 Oct 2026 21:15–31 Oct 23:45 UTC	5,233	1,158	2,617	50	83	62	73	99	>99	0	0
Relay ES · 31 Oct 2026 20:45–31 Oct 22:45 UTC	2,202	506	1,099	50	115	122	Unavailable	0	>99	0	0
Relay GY · 31 Oct 2026 20:30–31 Oct 22:00 UTC	3,451	777	1,726	50	82	73	67	99	>99	0	0
Relay BO · 31 Oct 2026 20:30–31 Oct 21:45 UTC	5,689	1,113	2,844	50	89	71	73	99	>99	<1	0
Relay DP · 31 Oct 2026 19:15–31 Oct 20:00 UTC	4,079	850	2,039	50	83	68	67	99	>99	0	0
Relay KD · 31 Oct 2026 19:15–31 Oct 23:15 UTC	6,820	1,499	3,412	50	118	90	81	>99	>99	0	0
Relay FS · 31 Oct 2026 19:15–31 Oct 23:45 UTC	6,540	1,482	3,270	50	91	60	73	>99	>99	<1	0
Relay KG · 31 Oct 2026 19:00–01 Nov 00:00 UTC	5,547	1,259	2,774	50	84	60	71	99	>99	<1	0
Relay JQ · 31 Oct 2026 18:15–31 Oct 18:30 UTC	4,445	986	2,224	50	75	59	65	96	>99	0	0
Relay JL · 31 Oct 2026 17:15–31 Oct 23:15 UTC	3,773	865	1,887	50	60	42	63	>99	>99	0	0
Relay EL · 31 Oct 2026 17:00–31 Oct 18:00 UTC	8,076	1,440	4,022	50	116	88	85	99	>99	0	0
Relay KG · 31 Oct 2026 15:15–31 Oct 17:15 UTC	3,875	890	1,938	50	66	46	66	99	>99	0	0
Relay IV · 31 Oct 2026 14:30–31 Oct 16:00 UTC	3,261	745	1,632	50	78	64	68	>99	>99	0	0
Relay FC · 31 Oct 2026 14:15–31 Oct 22:00 UTC	3,781	867	1,892	50	78	64	68	>99	>99	<1	0
Relay BN · 31 Oct 2026 13:00–01 Nov 13:30 UTC	1,683	393	841	50	50	46	47	>99	>99	<1	<1
Relay CJ · 31 Oct 2026 13:00–01 Nov 13:30 UTC	3,553	807	1,776	50	60	47	50	>99	>99	<1	<1
Relay HI · 31 Oct 2026 13:00–01 Nov 13:30 UTC	62	19	25	50	126	147	78	>99	89	4	<1
Relay BW · 31 Oct 2026 13:00–01 Nov 12:15 UTC	2,608	621	1,284	50	58	49	56	>99	99	<1	<1
Relay JR · 31 Oct 2026 13:00–01 Nov 13:30 UTC	1,169	274	585	50	33	30	39	>99	>99	<1	0
Relay JB · 31 Oct 2026 13:00–01 Nov 12:30 UTC	2,711	615	1,356	50	63	55	52	98	>99	<1	0
Relay HW · 31 Oct 2026 13:00–31 Oct 13:15 UTC	4,161	964	2,087	50	62	40	70	51	100	0	0
Relay HS · 31 Oct 2026 12:30–31 Oct 13:00 UTC	9,725	2,236	4,862	50	114	74	86	97	>99	0	0
Relay BZ · 31 Oct 2026 12:15–31 Oct 13:15 UTC	6,049	1,315	3,006	50	88	65	70	99	99	<1	0
Relay AK · 31 Oct 2026 12:15–31 Oct 13:00 UTC	7,054	1,363	3,526	50	213	205	98	99	>99	0	0
Relay GH · 31 Oct 2026 12:15–31 Oct 23:15 UTC	4,469	882	2,233	50	186	186	93	>99	>99	<1	0
Relay HS · 31 Oct 2026 12:15–31 Oct 12:30 UTC	9,137	1,129	4,570	50	112	81	88	81	>99	0	0
Relay HI · 31 Oct 2026 12:15–31 Oct 12:30 UTC	10,326	291	5,140	50	170	149	94	81	>99	0	0
Relay DJ · 31 Oct 2026 11:00–31 Oct 12:45 UTC	7,329	1,416	3,698	49	108	77	71	99	>99	0	0
Relay CL · 31 Oct 2026 09:15–31 Oct 10:45 UTC	4,182	891	2,094	50	71	56	66	99	>99	0	0
Relay CL · 31 Oct 2026 06:30–31 Oct 08:30 UTC	2,293	541	1,149	50	52	45	55	99	>99	0	0
Relay IE · 31 Oct 2026 06:30–31 Oct 11:00 UTC	3,242	758	1,621	50	55	41	60	>99	>99	0	0
Relay FS · 31 Oct 2026 04:15–31 Oct 12:15 UTC	3,128	734	1,566	50	61	48	63	>99	>99	0	0
Relay BX · 31 Oct 2026 03:15–31 Oct 13:45 UTC	5,292	1,223	2,646	50	138	130	86	99	>99	0	0
Relay KG · 31 Oct 2026 02:30–31 Oct 12:15 UTC	7,406	1,722	3,705	50	95	57	82	>99	>99	0	0
Relay DR · 31 Oct 2026 01:45–31 Oct 13:30 UTC	5,902	1,380	2,952	50	89	62	77	99	>99	<1	0
Relay DM · 31 Oct 2026 00:30–31 Oct 00:45 UTC	10,936	707	5,467	50	121	89	87	93	>99	<1	0
Relay DJ · 31 Oct 2026 00:15–31 Oct 07:30 UTC	3,307	756	1,685	48	62	49	55	>99	>99	<1	0
Relay KQ · 30 Oct 2026 23:45–31 Oct 00:15 UTC	5,310	756	2,652	50	182	175	92	95	>99	<1	0
Relay IH · 30 Oct 2026 22:30–31 Oct 12:15 UTC	8,105	1,887	4,054	50	112	75	81	>99	>99	0	0
Relay BX · 30 Oct 2026 22:15–31 Oct 03:00 UTC	11,274	2,166	5,629	50	196	173	95	99	>99	<1	0
Relay DJ · 30 Oct 2026 21:30–31 Oct 00:00 UTC	5,958	1,250	3,010	49	94	71	76	99	>99	0	0
Relay DM · 30 Oct 2026 21:15–31 Oct 00:00 UTC	3,354	767	1,678	50	61	46	61	99	>99	0	0
Relay JI · 30 Oct 2026 17:00–31 Oct 12:15 UTC	4,066	933	2,032	50	96	85	75	>99	>99	<1	0
Relay HJ · 30 Oct 2026 16:00–31 Oct 13:30 UTC	7,886	1,694	3,911	50	105	71	78	98	99	<1	<1
Relay CT · 30 Oct 2026 16:00–31 Oct 12:15 UTC	6,639	1,462	3,318	50	128	114	89	>99	>99	<1	0
Relay DR · 30 Oct 2026 14:30–31 Oct 01:30 UTC	5,777	1,271	2,889	50	89	67	77	>99	>99	<1	0
Relay DZ · 30 Oct 2026 14:30–31 Oct 04:30 UTC	2,309	528	1,153	50	128	134	80	>99	>99	<1	0
Relay HJ · 30 Oct 2026 14:15–30 Oct 15:45 UTC	6,811	1,392	3,407	50	102	75	72	>99	>99	0	0
Relay DI · 30 Oct 2026 14:00–30 Oct 21:45 UTC	4,003	865	1,998	50	73	56	66	>99	>99	<1	0
Relay JK · 30 Oct 2026 13:45–30 Oct 20:00 UTC	3,219	615	1,609	50	85	77	65	>99	>99	<1	0
Relay IR · 30 Oct 2026 13:30–31 Oct 12:15 UTC	5,929	1,353	2,965	50	93	68	76	>99	>99	<1	0
Relay DR · 30 Oct 2026 13:15–30 Oct 14:15 UTC	5,839	1,232	2,920	50	94	73	79	99	>99	0	0
Relay HR · 30 Oct 2026 12:15–31 Oct 12:15 UTC	6,193	1,439	3,096	50	113	88	79	>99	>99	0	0
Relay KP · 30 Oct 2026 12:15–31 Oct 12:15 UTC	3,957	868	1,978	50	90	78	71	>99	>99	<1	0
Relay CR · 30 Oct 2026 12:15–31 Oct 23:15 UTC	2,960	663	1,479	50	141	138	86	>99	>99	<1	0
Relay CM · 30 Oct 2026 12:15–30 Oct 12:30 UTC	7,234	1,608	3,627	50	109	82	Unavailable	0	>99	0	0
Relay JF · 30 Oct 2026 12:15–30 Oct 12:30 UTC	4,282	979	2,152	50	95	82	Unavailable	0	100	0	0
Relay JV · 30 Oct 2026 12:15–31 Oct 12:15 UTC	6,809	1,539	3,404	50	103	77	77	>99	>99	<1	0
Relay BX · 30 Oct 2026 12:15–30 Oct 22:00 UTC	5,256	1,187	2,628	50	153	146	89	>99	>99	<1	0
Relay HE · 30 Oct 2026 12:15–31 Oct 04:00 UTC	5,805	1,352	2,902	50	92	71	75	>99	>99	0	0
Relay JZ · 30 Oct 2026 12:15–31 Oct 06:15 UTC	8,236	1,756	4,118	50	115	81	81	>99	>99	<1	0
Relay FU · 30 Oct 2026 12:15–30 Oct 22:00 UTC	4,863	1,123	2,432	50	77	54	69	>99	>99	0	0
Relay CH · 30 Oct 2026 12:15–30 Oct 12:30 UTC	2,532	585	1,270	50	57	50	38	51	100	0	0
Relay GF · 30 Oct 2026 10:30–30 Oct 12:30 UTC	2,143	491	1,073	50	44	33	52	99	>99	0	0
Relay FI · 30 Oct 2026 09:30–30 Oct 10:45 UTC	6,857	1,508	3,431	50	93	58	80	98	>99	<1	0
Relay FO · 30 Oct 2026 09:00–30 Oct 09:15 UTC	6,778	1,407	3,397	50	89	63	78	76	>99	0	0
Relay IL · 30 Oct 2026 04:30–30 Oct 05:15 UTC	11,486	2,073	5,742	50	138	89	89	98	>99	0	0
Relay CW · 30 Oct 2026 02:45–30 Oct 03:15 UTC	9,679	1,378	4,837	50	152	132	95	97	>99	0	0

Resource changes before, during and after BEGIN_DIR surges

Of 97 periods of high admissions on 83 relays, 31 periods on 28 relays had sufficient measurements and stable observed flags for before/during/after comparison. CPU, read and written averages were higher during than on both sides in 26 comparisons on 24 relays. All 31 qualifying comparisons are shown, grouped by resource pattern and ordered by similarity within each group.

HSDir flag loss: During October 30–November 4, 11 observations on 11 relays—eight running stock Tor 0.4.9.15 and three with the custom patch—showed the local HSDir flag present during high admissions and absent afterward. Running and V2Dir were present in both compared samples; independent Onionoo snapshots corroborate three cases. These observations were excluded from the stable-flag resource comparison. They do not establish why the flag changed.

All 31 qualifying before, during and after comparisons across 28 relays, grouped into larger rises and falls, smaller rises and falls, and different traffic patterns. Columns show admitted BEGIN_DIR streams, whole-relay read/written bandwidth and main-thread CPU on shared scales. Repeat relay observations are shown separately. — **Footnote:** Matching means each of CPU, read and written averages was higher during the surge than before and after. October 29–November 4, 2026. All 31 qualifying comparisons on 28 relays are shown. Repeated observations of the same relay are numbered chronologically. Dates mark rounded observation starts. Points are fixed ten-minute means; spacing is categorical, not elapsed time. Read/written use Mbit/s; CPU is main-thread use as a percentage of one core. The first two groups have CPU and both traffic directions higher during than before and after. Larger changes require CPU at least 10 percentage points higher and each traffic direction at least 25% and 10 Mbit/s higher than both sides. The last group retains the five comparisons with different traffic patterns. Within groups, rows are ordered by similarity of their plotted profiles; measurements retain their original units and shared scales. Each phase requires at least 90% native-counter and CPU coverage with stable observed relay flags. The 24 matching relays comprise 11 with a custom patch and 13 running stock Tor. Instrumented examples are HSDir GET-largest overall; stock counters cannot identify HTTP-wait classifications. This sample does not describe all relays. Full-size figure.

Relay LF on November 4, 2026. Fixed ten-minute samples used for the 24-relay comparison; the during sample is the episode midpoint, separate from the late high-load period above.
Relay LF measurement	Before	During	After
Admitted BEGIN_DIR streams/s	197	7,836	6
Whole-relay read (Mbit/s)	36	119	21
Whole-relay written (Mbit/s)	33	70	21
Main-thread CPU (% of one core)	46	90	21

October 29–November 4. Each relay counts once per result even if it has several sequences. Larger changes require CPU at least 10 percentage points higher and each traffic direction at least 25% and 10 Mbit/s higher than both flanks.
Comparison result	Sequences	Distinct relays
Eligible complete comparisons	31	28
CPU, read and written higher during	26	24
Larger changes in all three	18	16