Beiträge von davorzec

Yes, it seems fine now.

Zitat von cysiek

So i performed same fio as davorzec on my RS 2000 (iddling)

Code

fio --name=mixed_workload --filename=~/fio_test.file --size=10G \
--rw=randrw --rwmixread=75 --bs=4k --ioengine=libaio --direct=1 \
--numjobs=4 --iodepth=16 --runtime=60 --time_based --group_reporting
mixed_workload: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=16
...
fio-3.39
Starting 4 processes
mixed_workload: Laying out IO file (1 file / 10240MiB)
Jobs: 4 (f=0): [f(4)][100.0%][r=284MiB/s,w=93.6MiB/s][r=72.7k,w=23.9k IOPS][eta 00m:00s]
mixed_workload: (groupid=0, jobs=4): err= 0: pid=500344: Mon Aug  3 21:52:35 2026
 read: IOPS=75.1k, BW=293MiB/s (308MB/s)(17.2GiB/60001msec)
   slat (nsec): min=761, max=1884.1k, avg=2652.56, stdev=1984.20
   clat (usec): min=55, max=13371, avg=782.53, stdev=128.74
    lat (usec): min=77, max=13377, avg=785.18, stdev=128.77
   clat percentiles (usec):
    |  1.00th=[  474],  5.00th=[  685], 10.00th=[  709], 20.00th=[  734],
    | 30.00th=[  758], 40.00th=[  766], 50.00th=[  783], 60.00th=[  799],
    | 70.00th=[  816], 80.00th=[  832], 90.00th=[  857], 95.00th=[  881],
    | 99.00th=[  938], 99.50th=[  971], 99.90th=[ 1139], 99.95th=[ 1336],
    | 99.99th=[ 6587]
  bw (  KiB/s): min=280752, max=358266, per=100.00%, avg=300339.15, stdev=1602.27, samples=480
  iops        : min=70188, max=89566, avg=75084.82, stdev=400.56, samples=480
 write: IOPS=25.1k, BW=98.0MiB/s (103MB/s)(5878MiB/60001msec); 0 zone resets
   slat (nsec): min=912, max=525562, avg=2967.64, stdev=1934.44
   clat (usec): min=35, max=18793, avg=196.68, stdev=139.03
    lat (usec): min=39, max=18796, avg=199.65, stdev=139.03
   clat percentiles (usec):
    |  1.00th=[  114],  5.00th=[  127], 10.00th=[  135], 20.00th=[  147],
    | 30.00th=[  157], 40.00th=[  167], 50.00th=[  180], 60.00th=[  190],
    | 70.00th=[  204], 80.00th=[  221], 90.00th=[  260], 95.00th=[  330],
    | 99.00th=[  537], 99.50th=[  635], 99.90th=[  996], 99.95th=[ 1352],
    | 99.99th=[ 6783]
  bw (  KiB/s): min=92336, max=121717, per=100.00%, avg=100324.64, stdev=729.36, samples=480
  iops        : min=23084, max=30429, avg=25081.16, stdev=182.34, samples=480
 lat (usec)   : 50=0.01%, 100=0.06%, 250=22.27%, 500=3.17%, 750=20.00%
 lat (usec)   : 1000=54.24%
 lat (msec)   : 2=0.22%, 4=0.01%, 10=0.02%, 20=0.01%
 cpu          : usr=2.30%, sys=8.87%, ctx=4464948, majf=0, minf=46
 IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=100.0%, 32=0.0%, >=64=0.0%
    submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
    complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.1%, 32=0.0%, 64=0.0%, >=64=0.0%
    issued rwts: total=4505067,1504862,0,0 short=0,0,0,0 dropped=0,0,0,0
    latency   : target=0, window=0, percentile=100.00%, depth=16
Run status group 0 (all jobs):
  READ: bw=293MiB/s (308MB/s), 293MiB/s-293MiB/s (308MB/s-308MB/s), io=17.2GiB (18.5GB), run=60001-60001msec
 WRITE: bw=98.0MiB/s (103MB/s), 98.0MiB/s-98.0MiB/s (103MB/s-103MB/s), io=5878MiB (6164MB), run=60001-60001msec
Disk stats (read/write):
 vda: ios=4493392/1501032, sectors=35947560/12009064, merge=30/98, ticks=3513055/290992, in_queue=3804050, util=83.93%

Alles anzeigen

Those are good results.

Where is your server located?

Zitat von meti

I think the main point is being misunderstood here. The fio benchmark was not the reason to assume there is a problem. The problem was noticed first in real-world usage, and the benchmark was only used afterwards to investigate the possible cause.

In my own experience, I have been running workloads on a weaker Het**zner VPS for more than a year without seeing this kind of storage-related degradation. I also had no issues with this Netcup server during the first week. The problem appeared only recently, which suggests something changed rather than simply being a case of unrealistic expectations about NVMe performance.

The fio results, along with other system information and observations, were provided through multiple support tickets over several days after no response was received. The concern is not only the performance issue itself, but also the lack of communication and feedback from support.

Even very small and less established providers usually acknowledge a support request within an hour or two. Waiting several days without even a basic response from a provider of this size is the frustrating part of this situation

Exactly.

I sent the first of three support tickets eight days ago, and I still have not received any response.

Zitat von webdesignheld

My little 1,94 Euro/Month G8 VPS

fiotest.png

perf.jpg

This is my with test as yours.

Only Falzo does not see problem here.

No, sir. This is the exact command used for my latest test:

fio --name=mixed_workload --filename=/root/fio_test.file --size=10G \
--rw=randrw --rwmixread=75 --bs=4k --ioengine=libaio --direct=1 \
--numjobs=4 --iodepth=16 --runtime=60 --time_based --group_reporting

I am not complaining blindly based on an fio benchmark. I first noticed terrible performance under real-world workloads on my 24 webistes. I then ran fio to investigate the cause, and the results pointed to storage performance as the likely bottleneck.

If you see a specific technical problem with the test command or methodology, please point it out.

Zitat von Falzo

Did the offer you bought specify any kind of IOps guarantue or did it just mention, that the underlying shared storage will consist of NVMe ?

Are you just complaining about an artificial benchmark number or is in anyway affecting your workload?

I would suggest to collect logs of high iowait impact on your real world use case, if you want to have something solid to base your complaints towards support on.

Or maybe make use of the satisfaction guarantue and quickly back out before its too late and you are stuck with a product you don't like.

This is a remarkably poor take and adds absolutely nothing useful to the discussion.

If your understanding of the subject is this limited, why bother replying at all?

Did you get any response from NetCup?

Same for me and my problem.

The worst part is that I haven't received a reply from Netcup in over a week.

No.

The worst part is that I have no idea, because I haven't received a reply from Netcup in over a week.

My latest FIO test still shows very poor storage performance:

Metric Result
Read IOPS 14,716
Write IOPS 4,922
Combined IOPS ≈19,638
Read bandwidth 57.4 MiB/s
Write bandwidth 19.2 MiB/s
Average read latency 2.99 ms
Average write latency 4.06 ms
99th percentile read latency 10.99 ms
99th percentile write latency 12.63 ms
Disk utilization 64.8%

Should netcup be concerned about potential legal consequences in Germany if its servers are advertised as using high-performance NVMe storage while customers receive substantially lower performance, especially when support tickets concerning the issue remain unanswered for several days?

My latest FIO test still shows very poor storage performance:

Metric Result
Read IOPS 14,716
Write IOPS 4,922
Combined IOPS ≈19,638
Read bandwidth 57.4 MiB/s
Write bandwidth 19.2 MiB/s
Average read latency 2.99 ms
Average write latency 4.06 ms
99th percentile read latency 10.99 ms
99th percentile write latency 12.63 ms
Disk utilization 64.8%

Should netcup be concerned about potential legal consequences in Germany if its servers are advertised as using high-performance NVMe storage while customers receive substantially lower performance, especially when support tickets concerning the issue remain unanswered for several days?

For the info server is in Wiena.

Mine is also in Vienna.

I recently purchased an RS 16000 G12 and am experiencing very poor disk performance, with only around 2,580 IOPS and very high, inconsistent latency.

Is anyone else using an RS 16000 G12 experiencing the same problem, or could this be an issue with my host node?

Netcup support has also been extremely unresponsive, so I would appreciate feedback from other users.

fio test.jpg

I share your frustration—my experience with the RS 8000 G12 has been very similar.

Hitting a ceiling of 20k IOPS on an RS 16000 G12 with 32 dedicated cores clearly points to a host-level bottleneck, whether it's storage limits, CPU pinning, or NUMA configuration.

The main issue is the wasted investment. Hiring an external team to configure the server and purchasing 3rd-party software licenses becomes a direct financial loss when the underlying hardware cannot deliver basic expected performance.

The lack of support response only makes things worse. Leaving tickets unanswered for days on production-tier servers leaves customers with no clear path forward.

I purchased an RS 16000 G12 and am experiencing terrible disk performance, with FIO benchmarks showing only around 20,000 IOPS.

The worst part is that support has not responded for five days.