Performance fluctuations, exploration.

    • New

    Hello,

    I just put my hands on new RS 2000 yesterday evening. After initial configuration i started some benchmarks and I was very very happy with results. Over 2000 i single core an and almoust 12000 in multicore with 60k iops in 64K. wrk tests confirmed performance peaking at 32k RPS from caddy + frankenphp and symfony endpoints. I know, its raw test, not worth much in real life. However, good starting point to work with. Today, in peak hours my tests went down noticably, so i started yabs gb6 + fio and noticed massive drops to 1300+ SC and 7000 MC and disk fluctuate between 30 and 49k iops (average from few tests).
    I'm not here to complain (yet hehe). I undertstand how it works, usually. But all this raised question about two things - first is CPU - since it is dedicated im my case, such a massive drop means CPU clock was probably reduced (ye, I don't believe 96 core chip can hold 3.7 Ghz boost from initial 2.3 all the time). Unfortunatelly cpu clock is not passed to VM, so I cant see it directly.
    Disk operations I understand more, because I'm aware it's shared, but still dont know where is the baseline - what to expect?

    I'll be glad if someone more expierienced share some thoughts on that. Maybe I just expect too much and can't swollow reality pill.

    Edited once, last by cysiek (July 29, 2026 at 2:02 PM).

    • New

    I think it's main problem in my case also and it affects overall score. I opened ticket yesterday, because I 've made more test, including my own app which performance I know for sure. It's I/O bounded (database) and... oh boy, never saw results bad like this, maybe in debug mode.
    I dont need much for my case, but stability is most important thing, thats why decided to RS line.

    meanwhile. this is barely acceptable (7-7.5k events per second avg was inittially on this machine)
    [09:46:27] Processed: 5000 | Δt: 1.57s | Total: 1.57s | AVG EPS: ~3192, LAP EPS: ~3192
    [09:46:29] Processed: 10000 | Δt: 1.36s | Total: 2.93s | AVG EPS: ~3415, LAP EPS: ~3671
    [09:46:30] Processed: 15000 | Δt: 1.22s | Total: 4.15s | AVG EPS: ~3612, LAP EPS: ~4085
    [09:46:31] Processed: 20000 | Δt: 1.07s | Total: 5.22s | AVG EPS: ~3832, LAP EPS: ~4686
    [09:46:32] Processed: 25000 | Δt: 0.97s | Total: 6.19s | AVG EPS: ~4038, LAP EPS: ~5146
    [09:46:33] Processed: 30000 | Δt: 1.11s | Total: 7.31s | AVG EPS: ~4106, LAP EPS: ~4487
    [09:46:34] Processed: 35000 | Δt: 1.24s | Total: 8.55s | AVG EPS: ~4094, LAP EPS: ~4020
    [09:46:36] Processed: 40000 | Δt: 1.11s | Total: 9.66s | AVG EPS: ~4139, LAP EPS: ~4486
    [09:46:37] Processed: 45000 | Δt: 1.07s | Total: 10.73s | AVG EPS: ~4193, LAP EPS: ~4683
    [09:46:38] Processed: 50000 | Δt: 1.53s | Total: 12.26s | AVG EPS: ~4077, LAP EPS: ~3265

    this is not acceptable
    [09:56:11] Processed: 5000 | Δt: 2.03s | Total: 2.03s | AVG EPS: ~2469, LAP EPS: ~2469
    [09:56:12] Processed: 10000 | Δt: 1.40s | Total: 3.42s | AVG EPS: ~2921, LAP EPS: ~3576
    [09:56:14] Processed: 15000 | Δt: 1.43s | Total: 4.85s | AVG EPS: ~3093, LAP EPS: ~3507
    [09:56:15] Processed: 20000 | Δt: 1.68s | Total: 6.53s | AVG EPS: ~3063, LAP EPS: ~2974
    [09:56:17] Processed: 25000 | Δt: 1.86s | Total: 8.39s | AVG EPS: ~2978, LAP EPS: ~2683
    [09:56:19] Processed: 30000 | Δt: 1.47s | Total: 9.87s | AVG EPS: ~3041, LAP EPS: ~3398
    [09:56:20] Processed: 35000 | Δt: 1.48s | Total: 11.34s | AVG EPS: ~3086, LAP EPS: ~3386
    [09:56:22] Processed: 40000 | Δt: 1.77s | Total: 13.12s | AVG EPS: ~3050, LAP EPS: ~2818
    [09:56:24] Processed: 45000 | Δt: 2.10s | Total: 15.22s | AVG EPS: ~2957, LAP EPS: ~2379
    [09:56:26] Processed: 50000 | Δt: 1.75s | Total: 16.97s | AVG EPS: ~2946, LAP EPS: ~2855

    And this is... joke
    [10:07:51] Processed: 5000 | Δt: 3.09s | Total: 3.09s | AVG EPS: ~1620, LAP EPS: ~1620
    [10:07:54] Processed: 10000 | Δt: 2.79s | Total: 5.88s | AVG EPS: ~1701, LAP EPS: ~1791
    [10:07:57] Processed: 15000 | Δt: 2.84s | Total: 8.72s | AVG EPS: ~1721, LAP EPS: ~1761
    [10:08:00] Processed: 20000 | Δt: 3.22s | Total: 11.94s | AVG EPS: ~1675, LAP EPS: ~1553
    [10:08:04] Processed: 25000 | Δt: 4.07s | Total: 16.00s | AVG EPS: ~1562, LAP EPS: ~1229
    [10:08:07] Processed: 30000 | Δt: 3.29s | Total: 19.29s | AVG EPS: ~1555, LAP EPS: ~1522
    [10:08:11] Processed: 35000 | Δt: 3.49s | Total: 22.78s | AVG EPS: ~1537, LAP EPS: ~1434
    [10:08:14] Processed: 40000 | Δt: 3.17s | Total: 25.94s | AVG EPS: ~1542, LAP EPS: ~1579
    [10:08:17] Processed: 45000 | Δt: 3.00s | Total: 28.95s | AVG EPS: ~1555, LAP EPS: ~1665
    [10:08:20] Processed: 50000 | Δt: 2.89s | Total: 31.84s | AVG EPS: ~1570, LAP EPS: ~1729

    Database mixed insert/updates from same dataset.

    • New

    No, no e-mail response from support. Instability through whole weekend and today's average from my point of view performance (~5000 EPS in my app test while CPU stayed at ~1500 single core in geek benchmark) are somehow gone these evening.
    https://browser.geekbench.com/v6/cpu/18901069 <- was
    https://browser.geekbench.com/v6/cpu/18904577 <- currently

    For a moment I thought I broke something and I'm still in shock and not sure wtf, but looks like not only CPU is fixed, but disk performance also increased. When last top result was ~12 seconds to perform operations, moment ago:
    [18:58:29] Processed: 100000 | Δt: 0.42s | Total: 8.72s | AVG EPS: ~11471, LAP EPS: ~11927
    [19:29:03] Processed: 100000 | Δt: 0.52s | Total: 10.67s | AVG EPS: ~9368, LAP EPS: ~9594
    [19:38:20] Processed: 100000 | Δt: 0.51s | Total: 10.53s | AVG EPS: ~9499, LAP EPS: ~9812

    These numbers are crazy good. Twice as good as earlier today. I doubt it stays like this forever :(

    BTW
    I will paste fio results in Your thread. I use different settings, but I will test ones you provided.