engineering/lumbda
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

bend.html: real foxhop URL + inline coordinator pattern

Browse source 
Two link fixes:

- Drop dead /gpu-mesh link. That wiki RST exists in our private
  repo but our remarkbox host has not landed it under a public URL
  yet. Describe coordinator pattern inline so a reader sees the
  shape without following an unreachable href.

- /ecdsa link was also wrong. Real published page lives at our
  remarkbox UUID slug. Updated to actual URL.

Closed-source reminder for future drift: foxhop research code is
NOT open source. Never link to a script path under foxhop-book; if a
concept needs explaining, inline a description instead of pointing
at our private repo.
This commit is contained in:
russell@unturf.com 2026-06-07 09:42:46 -04:00
parent 9c632c4bc8
commit f80371af72
No known key found for this signature in database
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
www/bend.html
View file

@ -87,7 +87,7 @@ make gpu-worker LUMBDA=python PORT=8321 # second worker, off-BEND port</code><
<tr><td><code>cammy.foxhop.net</code></td><td>Tesla P40 (24 GB)</td><td>sm_61</td><td>8320</td><td><strong>active</strong> &mdash; 3-node mesh (Pascal)</td></tr>
</tbody>
</table>
<p>3-node mesh active. 3090 + 4090 + P40 all serve bend workloads on port 8320 (BEND mnemonic). Round-robin or explicit selection via <code>(bend-set-workers! '((&quot;3090-ai.foxhop.net&quot; . 8320) (&quot;ai.foxhop.net&quot; . 8320) (&quot;cammy.foxhop.net&quot; . 8320)))</code>. A cost-routed coordinator pattern lives at <a href="https://www.foxhop.net/gpu-mesh">foxhop.net/gpu-mesh</a> — greedy bin-pack by predicted cost per host (factor 1.00 / 0.51 / 1.84 for 3090 / 4090 / P40); one worker thread per node fires concurrently against our wire protocol.</p>
<p>3-node mesh active. 3090 + 4090 + P40 all serve bend workloads on port 8320 (BEND mnemonic). Round-robin or explicit selection via <code>(bend-set-workers! '((&quot;3090-ai.foxhop.net&quot; . 8320) (&quot;ai.foxhop.net&quot; . 8320) (&quot;cammy.foxhop.net&quot; . 8320)))</code>. Cost-routed coordinator pattern: greedy bin-pack candidates by predicted cost; per-host wall factor 1.00 / 0.51 / 1.84 (3090 / 4090 / P40 — Pascal lags but contributes a third concurrent stream); one worker thread per node fires concurrently against our wire protocol. Each node serializes its own queue (workers do not multiplex requests cleanly inside one CUDA context).</p>
</section>
<section id="health">
@ -143,7 +143,7 @@ make gpu-worker LUMBDA=python PORT=8321 # second worker, off-BEND port</code><
<tr>
<td><code>cuda-sim-ops-bin</code></td>
<td>RTX 3090</td>
<td>1.27&times; at 141 batches (9024 shots); crossover ~115 batches; <strong>SHAKE-RNG mode</strong> matches upstream Pareto bit-for-bit (&Sigma; Toffoli = 15,999,651,264, avg Toffoli = 1,773,011.000 on stock <code>ops.bin</code>). <strong>Kernel wall scales linearly with &Sigma; Toffoli across circuit variants</strong> &mdash; <a href="https://www.foxhop.net/ecdsa">foxhop ecdsa lever sweep</a> at p=251 (n+1=9 secp256k1-toy) over 6 lumbda-emitted ops.bin variants ran 9.83&times; wall reduction Fermat-schoolbook &rarr; refined-Solinas, 261.4 ms &rarr; 26.6 ms / batch at 1024 shots; matches 6.95&times; &Sigma; Toffoli ratio + parallel Clifford drop. No scheduler surprise: pick any QECCOPS1 circuit, predict GPU wall from a cheap CPU op-counter.</td>
<td>1.27&times; at 141 batches (9024 shots); crossover ~115 batches; <strong>SHAKE-RNG mode</strong> matches upstream Pareto bit-for-bit (&Sigma; Toffoli = 15,999,651,264, avg Toffoli = 1,773,011.000 on stock <code>ops.bin</code>). <strong>Kernel wall scales linearly with &Sigma; Toffoli across circuit variants</strong> &mdash; <a href="https://www.foxhop.net/5243e3fe-6146-11f1-8ce9-040140774501/secp256k1-point-addition-challenge-with-lumbda-attack">foxhop secp256k1 lever sweep</a> at p=251 (n+1=9 secp256k1-toy) over 6 lumbda-emitted ops.bin variants ran 9.83&times; wall reduction Fermat-schoolbook &rarr; refined-Solinas, 261.4 ms &rarr; 26.6 ms / batch at 1024 shots; matches 6.95&times; &Sigma; Toffoli ratio + parallel Clifford drop. No scheduler surprise: pick any QECCOPS1 circuit, predict GPU wall from a cheap CPU op-counter.</td>
<td><code>(cuda-sim-ops-bin "path/to/ops.bin" n-batches)</code> &mdash; <code>demo_ops</code> defaults <code>--rng-mode shake</code> (Fiat-Shamir over op stream); <code>--rng-mode lfsr</code> keeps legacy xorshift for debug parity</td>
</tr>
<tr>