Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Advertise on LowEndTalk.com
Is this NVME Normal?
New on LowEndTalk? Please Register and read our Community Rules.

Is this NVME Normal?

Hi, LET!

So I have a Rise-1 Dedicated Server with OVH. And when I run some yabs.sh, I got these results.

Is this normal?
It was NVMe with Soft RAID 1

YABS.SH Results

NVMe Smartcl Results


Is this normal?
I mean, I have never seen NVMe this low on disk test.

Thanks, LET!

Freelance Web Developer & Web Designer - Available for hire | DT Server Status - My server uptime status

Comments

  • As per my knowledge this is pretty lame speed to a NVMe disk. :/

    Thanked by 1DianTama

    Kimsufi ;) | HostSolutions o:) | Nexusbytes B) | Racknerd <3 | OVH :* |
    Letbox :love: | Hotlineservers :flushed:

  • verovero Member, Provider

    Here's NVMe storage benchmark from another active thread, 4k speeds are quite low, but the rest should be similar for general NVMes:

    @marvel said:

    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 93.50 MB/s   (23.3k) | 804.48 MB/s  (12.5k)
    Write      | 93.75 MB/s   (23.4k) | 808.71 MB/s  (12.6k)
    Total      | 187.25 MB/s  (46.8k) | 1.61 GB/s    (25.2k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 1.11 GB/s     (2.1k) | 1.23 GB/s     (1.2k)
    Write      | 1.17 GB/s     (2.2k) | 1.31 GB/s     (1.2k)
    Total      | 2.29 GB/s     (4.4k) | 2.54 GB/s     (2.4k)
    
  • marvelmarvel Member, Provider
    edited February 21

    What OS (distro) are you running?

    Thanked by 1DataWagon
  • marvelmarvel Member, Provider

    @vero said:
    Here's NVMe storage benchmark from another active thread, 4k speeds are quite low, but the rest should be similar for general NVMes:

    @marvel said:

    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 93.50 MB/s   (23.3k) | 804.48 MB/s  (12.5k)
    Write      | 93.75 MB/s   (23.4k) | 808.71 MB/s  (12.6k)
    Total      | 187.25 MB/s  (46.8k) | 1.61 GB/s    (25.2k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 1.11 GB/s     (2.1k) | 1.23 GB/s     (1.2k)
    Write      | 1.17 GB/s     (2.2k) | 1.31 GB/s     (1.2k)
    Total      | 2.29 GB/s     (4.4k) | 2.54 GB/s     (2.4k)
    

    That's networked storage ;)

  • Definitely not okay , but it depend on the brand you have is it Samsung or something else ?

  • spectraipspectraip Member, Provider
    edited February 21

    If you don't have anything important running on the server, reinstall it and install Linux on one disk only and leave the other NVMe empty. Software RAID can slow down SSD disks and NVMe disks.

    Then you can see if it's (one of) the NVMe drives or a performance loss due to software RAID.

    Thanked by 1ariq01
  • DianTamaDianTama Member
    edited February 21

    @marvel said:
    What OS (distro) are you running?

    It was running on Centos 7 with DirectAdmin in it.

    @momkin said:
    Definitely not okay , but it depend on the brand you have is it Samsung or something else ?

    I don't know the exact thing since I rented it from OVH.
    But i think it was Intel.

    @spectraip said:
    If you don't have anything important running on the server, reinstall it and install Linux on one disk only and leave the other NVMe empty. Software RAID can slow down SSD disks and NVMe disks.

    Then you can see if it's (one of) the NVMe drives or a performance loss due to software RAID.

    Too bad currently, I'm running a production site in this Dedi.

    In OVH,
    does anyone have experience with requesting NVMe Replacement?

    Freelance Web Developer & Web Designer - Available for hire | DT Server Status - My server uptime status

  • HosterlabsHosterlabs Member, Provider

    @DianTama said:

    @marvel said:
    What OS (distro) are you running?

    It was running on Centos 7 with DirectAdmin in it.

    @momkin said:
    Definitely not okay , but it depend on the brand you have is it Samsung or something else ?

    I don't know the exact thing since I rented it from OVH.
    But i think it was Intel.

    @spectraip said:
    If you don't have anything important running on the server, reinstall it and install Linux on one disk only and leave the other NVMe empty. Software RAID can slow down SSD disks and NVMe disks.

    Then you can see if it's (one of) the NVMe drives or a performance loss due to software RAID.

    Too bad currently, I'm running a production site in this Dedi.

    In OVH,
    does anyone have experience with requesting NVMe Replacement?

    Just use the API on OVH or open a ticket, they'll replace it probably.

    [Signature violates rules]

  • @Hosterlabs said: Just use the API on OVH or open a ticket, they'll replace it probably.

    Ill try :D
    But, does the data will be gone?

    Freelance Web Developer & Web Designer - Available for hire | DT Server Status - My server uptime status

  • HosterlabsHosterlabs Member, Provider

    @DianTama said:

    @Hosterlabs said: Just use the API on OVH or open a ticket, they'll replace it probably.

    Ill try :D
    But, does the data will be gone?

    Probably. Make sure to have backups and if it is a RAID Array you could probably rebuild it.

    I am not sure if they require additional info for troubleshooting. They always give me new disks for some reason.

    Thanked by 1DianTama

    [Signature violates rules]

  • WebProjectWebProject Member, Provider

    @DianTama said: But, does the data will be gone?

    as OVH is unmanaged service, do you think they will migrate the data for you?

    VPS Price Match Guarantee on: All our range of DDOS protected VPS Plans
    Are you looking for best price for self-managed VPS? See WebProVPS website for more details.
  • @WebProject said:

    @DianTama said: But, does the data will be gone?

    as OVH is unmanaged service, do you think they will migrate the data for you?

    I don't know,
    I expect they will clone my drive maybe :/

    Freelance Web Developer & Web Designer - Available for hire | DT Server Status - My server uptime status

  • seriesnseriesn Member, Top Provider

    @DianTama said:

    @WebProject said:

    @DianTama said: But, does the data will be gone?

    as OVH is unmanaged service, do you think they will migrate the data for you?

    I don't know,
    I expect they will clone my drive maybe :/

    No sir.

    Also check your raid health just to be safe :)

  • marvelmarvel Member, Provider
    edited February 21

    Try a more recent OS, like Ubuntu 20.04. Perhaps you can try a live CD and run the test again. You know CentOS 7 is quite old and I've seen some performance related issues with it on newer hardware.

    Also like said above me, type 'cat /proc/mdstat' to see if no rebuilt/repair is in progress.

  • MarcoooMarcooo Member, Provider
    edited February 21

    you can try this command to see exactly what model the drives are lshw -class disk or fdisk -l maybe you can see here more information about the disks.

    Thanked by 1Mahfuz_SS_EHL

    | TurboHost| turbohost.nl | DDoS Protected |

  • WebProjectWebProject Member, Provider

    @DianTama said: I expect they will clone my drive maybe

    with unmanaged service, you don't get such privileges. Unfortunately, you will require to do it yourself.

    Thanked by 2webcraft DianTama
    VPS Price Match Guarantee on: All our range of DDOS protected VPS Plans
    Are you looking for best price for self-managed VPS? See WebProVPS website for more details.
  • momkinmomkin Member
    edited February 21

    Also if this doesn't affect the performance of your websites , no need to do anything just leave it as is or you will end up break your OS , unless if you are benchmarks geek :smile:

    Thanked by 2Falzo DianTama
  • darkimmortaldarkimmortal Member
    edited February 22

    Assuming you haven’t just cropped off the bit of smartctl that identifies the disks, try nvme-cli or lsscsi to identify them

    Thanked by 1DianTama
  • Mahfuz_SS_EHLMahfuz_SS_EHL Member, Provider

    Check the NVMe Model, then something could be decided.

    AlphaSSL Revocation Issue is being investigated.

  • @momkin said:
    Also if this doesn't affect the performance of your websites , no need to do anything just leave it as is or you will end up break your OS , unless if you are benchmarks geek :smile:

    I think its affecting the production website its load slower than before.
    No changes made on the code itself.

    And
    This is the NVMe Model.

    Freelance Web Developer & Web Designer - Available for hire | DT Server Status - My server uptime status

  • K4Y5K4Y5 Member
    edited February 23

    @DianTama I have the exact server. You certainly have some issue with your NVMe drives (Nothing crazy though). Here's my YABS (NVMe drives in RAID1):

    Basic System Information:
    ---------------------------------
    Processor  : Intel(R) Xeon(R) CPU E3-1270 v6 @ 3.80GHz
    CPU cores  : 8 @ 1195.385 MHz
    AES-NI     : ✔ Enabled
    VM-x/AMD-V : ✔ Enabled
    RAM        : 30.9 GiB
    Swap       : 1022.0 MiB
    Disk       : 29.9 GiB
    
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 527.95 MB/s (131.9k) | 582.30 MB/s   (9.0k)
    Write      | 529.34 MB/s (132.3k) | 585.37 MB/s   (9.1k)
    Total      | 1.05 GB/s   (264.3k) | 1.16 GB/s    (18.2k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 583.68 MB/s   (1.1k) | 587.76 MB/s    (573)
    Write      | 614.69 MB/s   (1.2k) | 626.91 MB/s    (612)
    Total      | 1.19 GB/s     (2.3k) | 1.21 GB/s     (1.1k)
    
    
    Drives:
    Model Number:                       INTEL SSDPE2MX450G7
    Data Units Read:                    35,746,170 [18.3 TB]
    Data Units Written:                 88,295,440 [45.2 TB]
    
    Model Number:                       INTEL SSDPE2MX450G7
    Data Units Read:                    36,979,436 [18.9 TB]
    Data Units Written:                 90,101,944 [46.1 TB]
    
    Thanked by 1DianTama
  • jsgjsg Member
    edited February 23

    @spectraip said:
    Then you can see if it's (one of) the NVMe drives or a performance loss due to software RAID.

    Insignificant

    @DianTama

    That's a weird disk anyway. On the one hand it's MLC, which is much better and faster than TLC or QLC, but on the other hand its write IOPs are quite low (< 20k). And the performance profile is really perverse; 4k ("sector") performance way better than 64k and even better than 1M?
    All in all that thing is worse than a halfway decent SSD drive.

    Something is seriously wrong. Either with the hardware/disk or with your OS or with your benchmarking.

    Thanked by 1DianTama

    The problem with democracy is that by definition > 85% of the voters are not particularly intelligent.

  • MarcoooMarcooo Member, Provider

    That SSD is not that fast rated speeds are
    1200 MBps (read) / 600 MBps (write) maybe the raid slow down the speeds

    Thanked by 1DianTama

    | TurboHost| turbohost.nl | DDoS Protected |

  • FalzoFalzo Member
    edited February 23

    @jsg said: And the performance profile is really perverse; 4k ("sector") performance way better than 64k and even better than 1M?

    yes! maybe this shows that there is simply a lot going on on that node already? bandwidth seems limited or eaten by something else and therefore the limiting factor.

    obviously one can't expect to see raw performance numbers that express an idle disk in system running in full production...

    @jsg said: ... or with your benchmarking.

    this. @DianTama overhink if you taken the current load on the system into account while you ran the benchmark and if this can even properly reflect what you expected to see (raw nvme performance)

    UltraVPS.eu KVM in US/UK/NL/DE: 15% off first 6 month | Netcup VPS/rootDS - 5€ off: 36nc15279180197 (ref)

  • @jsg said: Something is seriously wrong. Either with the hardware/disk or with your OS or with your benchmarking.

    Thanks for the explanation, ill try to check on all of those and hope that I found the issues. And maybe try to get disk replacement with the provider when I'm ready, since replacing a disk took a lot of effort.

    @Falzo said: yes! maybe this shows that there is simply a lot going on on that node already? bandwidth seems limited or eaten by something else and therefore the limiting factor.

    This could be true, right now I'm analyzing all possible causes.

    Freelance Web Developer & Web Designer - Available for hire | DT Server Status - My server uptime status

  • ProwHostProwHost Member, No Sales

    Hello there,

    Their speed seems a bit slow compared to NVMe Disk. With the Intel P4618, I get about 1500 mb / s speed.

  • @Falzo said:
    yes! maybe this shows that there is simply a lot going on on that node already? bandwidth seems limited or eaten by something else and therefore the limiting factor.

    Isn't this a dedicated server not a virtual one ?

    So what does has to do with the node ?

  • jsgjsg Member

    @momkin said:

    @Falzo said:
    yes! maybe this shows that there is simply a lot going on on that node already? bandwidth seems limited or eaten by something else and therefore the limiting factor.

    Isn't this a dedicated server not a virtual one ?

    So what does has to do with the node ?

    It's a production machine.

    The problem with democracy is that by definition > 85% of the voters are not particularly intelligent.

  • @momkin said:

    @Falzo said:
    yes! maybe this shows that there is simply a lot going on on that node already? bandwidth seems limited or eaten by something else and therefore the limiting factor.

    Isn't this a dedicated server not a virtual one ?

    So what does has to do with the node ?

    maybe wrong choice of word, sorry. I meant 'a lot going on on that whole server already'...

    @DianTama mentioned that it runs a website in production, that is busy and he even think he noticed that it is getting slower. so there probably is quite some load already and therefore a benchmark in parallel of course can only measure what's left to use and not what could be, if the whole thing was fresh/idle otherwise.

    without knowing what exactly the server is doing that's just speculation though - however the numbers in the benchmark show enough breathing room left.

    TL;DR; if you want to be worried about those numbers, then probably only if you get the same result on an idle box/NVMe.

    UltraVPS.eu KVM in US/UK/NL/DE: 15% off first 6 month | Netcup VPS/rootDS - 5€ off: 36nc15279180197 (ref)

  • darkimmortaldarkimmortal Member
    edited February 23

    @DianTama said:

    @momkin said:
    Also if this doesn't affect the performance of your websites , no need to do anything just leave it as is or you will end up break your OS , unless if you are benchmarks geek :smile:

    I think its affecting the production website its load slower than before.
    No changes made on the code itself.

    And
    This is the NVMe Model.

    They are in 512 byte sector mode so check your partitions are aligned correctly

    Edit: just seen it’s an old pci card style nvme, probably working as expected then. You might be able to squeeze more out of it in 4k sector mode, but needs a reinstall

  • coolicecoolice Member
    edited February 23
    1. this is Intel DC p3520 according to serial number it is not a speedster (how other mentioned too)

    https://ark.intel.com/content/www/us/en/ark/products/93188/intel-ssd-dc-p3520-series-450gb-2-5in-pcie-3-0-x4-3d1-mlc.html

    It allready exceed the 4k random write specifics at IOPS (19000)

    600MB Write and 1200MB Reads speed are for sequential - click on question mark at Intel Site

    Sequential Bandwidth - 100% Read (up to)
    Speed with which the device is able to retrieve data that forms one contiguous, ordered block of data. Measured in MB/s (MegaBytes per Second)

    Sequential Bandwidth - 100% Write (up to)
    Speed with which the device is able to record data into one contiguous, ordered block of data. Measured in MB/s (MegaBytes per Second)

    You can test sequential speed with dd

    (Home PC drives are faster as they use large ram buffers and do not have power loss protection) ...

    2 Centos is not Ubuntu and do not execute trim weekly... with great probability drives are never trimmed

    Future-Proof Yourself! Buy! Buy!

Sign In or Register to comment.