Poor disk performance in the new servers in USA

    Hello, I have 2 root servers in USA and several in Germany. I have noticed bad performance on the USA servers and it seems to be due to disk performance. Using a simple crystaldiskmark I can see that the write performance is quite bad on both the US servers. Around 500MB/s compared to Germany based servers that give similar write performance as read, somewhere around 4000MB/s. Is this is due to a problem, or is this the expected performance? If so I will have to move back the servers.

    Quote from Zulan

    If I have 4000-7000 on one server and 500 on another, then that is the issue rather then what you find to be good performance.


    Seems something is wrong with your test. Or there is the MB/s<>MBit/s misunderstanding here.
    500MB/s is a good value.
    >1G you get probably only on real hardware with NVME.

    Quote from Zulan

    TBT your comment just makes me sick. If I have 4000-7000 on one server and 500 on another, then that is the issue rather then what you find to be good performance.

    The last generation didn't really go over 250 Mbytes/s, no matter where. So 500 Mbytes/s are already a big improvement and by no way "poor". Get other VServers and experience "poor disk performance".


    Apart from that: no shown test results, no credibility.

    RS Ostern L OST22 (~RS "3000" G9.5) (8C,24GB,960GB) | RS Cyber Quack (1C,2GB,40GB)

    My opinion on this: If you consider that the vServer RS 1000 G11 costs less than 10 euros per month as a special edition and only has a contract period of 31 days, I consider the following values to be appropriate for such a small vServer:

    If you really put a vServer RS 1000 G11 under load, you will see a significantly better IO:nc_stress_test_rs1000g11.png


    uptime

    12:19:07 up 6 days, 12:06, 2 users, load average: 1024,11, 888,34, 489,63

    Well, if the question simply is: "Is IO for big block sizes on RS G11 worse than on RS G9.5?", i would say the answer is "Yes, definitely, and by a big margin". For answering this question, prices and considerations what is "good enough" are irrelevant. The test below also shows, that more than 500 MB/s have been reality for all but the lowest block sizes.


    I agree, that the values of G11 are still ok, especially for the price. vServers from competitors in this price range are often worse. So I won't complain. But that's a completely different story.


    My old RS 2000 G9.5 showed (tested with YABS)

    Thanks for all your answers guys! It was crystaldiskmark that gave me the numbers. I am not sure how they have calculated the MB but it looks like this:


    CrystalDiskMark test US server RS2000 G11


    And my older G9 RS2000 server gives me the highest numbers of 4000 both read/write.


    Anyone has any idea how to get yabs to run on windows? Virtualization is not enabled in bios and I'm not sure how to edit bios settings. I would think crystal diskmark shows my point anyway?

    Thank you Andreas, I did that, but I get no confirmation it's running, has completed or anything. I went to All Yab Runs | VPSBenchmarks but I can't see anything from netcup there today. It's my understanding it will run the benchmark and upload the results to vpsbenchmarks? I then tried running curl -sL yabs.sh | bash to see if that changed anything but got the exactly same, nothing, prompt just goes to next level with no output.

    I also played through it again and you can get it to work like this:

    Quote from andreas.

    I also played through it again and you can get it to work like this:

    Now the tests run, but the disk tests only gives out 0:s.

    io Disk Speed Tests (Mixed R/W 50/50) (Partition overlay):

    ---------------------------------

    Block Size | 4k (IOPS) | 64k (IOPS)

    ------ | --- ---- | ---- ----

    Read | 0.00 KB/s (0) | 0.00 KB/s (0)

    Write | 0.00 KB/s (0) | 0.00 KB/s (0)

    Total | 0.00 KB/s (0) | 0.00 KB/s (0)

    | |

    Block Size | 512k (IOPS) | 1m (IOPS)

    ------ | --- ---- | ---- ----

    Read | 0.00 KB/s (0) | 0.00 KB/s (0)

    Write | 0.00 KB/s (0) | 0.00 KB/s (0)

    Total | 0.00 KB/s (0) | 0.00 KB/s (0)

    I had to mount the /dev/vdb3 disk instead and now it works. Now there is no longer any difference between read/write. I am not sure what conclusions I can actually draw from this.


    I tested it using Atto Disk benchmark as well. Here you really see the difference. The operating system is a default Windows standard 2022.


    First image is the US server.

    I have a second US RS2000 G11 server with the same operatingsystem and the same performance difference in read/write.

    I also have a small vps 1000 G10 based in nurnberg de and it also has similar disk performance both read/write.

    This is not a problem only visible in benchmarks. I can clearly see how sluggish the US servers are compared to the DE.

    So to sum up the problem. I have 1 RS2000 G9.5 based in Germany and two RS2000 G11 based in US.

    All three servers have about the same performance when benchmarking with yabs.


    But in windows the US servers have considerably lower performance tested using two different benchmarks, not only visible in benchmarks but also noticeable when the servers are under load.


    All three servers use windows 2022 standard.


    One theory I have is it might be something with caching since the write speeds I get in windows is similar to what I get with linux and the read speeds are just insanely high compared to the linux tests. So I guess linux does not use caching. I have not changed any caching settings on either of the machines so it's still confusing.


    I can now see what you meant TBT and I'm sorry barked at you.

    Have you selected the same drivers in SCP at Media > Local Disks?

    RS Brezn | VPS 500 G8 Plus | 2× VPS Karneval 2020 | VPS Pocket Admin | RS Cyber Quack | VPS 500 ARM


    Dieses Gebäude hat mir die Vorfahrt genommen! *hup*