Howdy, Stranger!

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


Contabo introducing VPS with NVMe drives - Page 7
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Contabo introducing VPS with NVMe drives

123457»

Comments

  • jsgjsg Member, Resident Benchmarker
    edited September 2021

    @cybertech said:
    he's a server review king. where are your creds?

    BS, I'm not a king. Someone gave me that tag, doubtlessly well meaning and with friendly intentions, so I didn't and don't complain, but putting a "king" tag on someone doesn't make him a king.
    I'm a Let user and if and insofar as I feel occasionally superior to some other users here that's only due to a difference in knowledge and hands-on experience and in terms of politeness, culture and manners and attitude. But generally I'm just yet another LET user and definitely not a king or even just a count/earl.


    Back to the matter ...

    I just had a somewhat long phone conversation with someone in Contabo's middle management who knows their operations quite well from an earlier tech position there. That guy clearly knows how many NVMes are in a node, how many vCores are on an Epyc HWT, etc. and, just like my other contacts there so far was very open and frank. While he answered all my technical questions, albeit on the perfectly understandable condition to keep internal numbers confidential, really openly and without hesitation, he also was very frank, almost a bit rude in stating unmistakably and clearly that at the end of the day Contabo is a company and needs and wants to make a profit (which didn't feel nice and cozy but in a way confirms his being really open and frank).

    I can tell you that the internal numbers he provided me clearly place Contabo in the best third of providers. IMO certainly not at the top but clearly solidly above average both wrt vCores per HWT and disk IO.
    Based on the information I got I can tell you that the allegations of the (allegedly) oh so low IO rates are BS. The actual rates are way above what is alleged, plus the allowed burst rates are about three times higher. But yes, of bloody course there are limits, as are with every other providers VPSs, but the Contabo limits are by no means tight and certainly not encountered by 99% of their customers. As for "how often and how long can one burst?" he had no concrete data available but just something to the effect of "that's just the values of the hypervisor" and, based on practical experience "most normal customers are limited very rarely or basically never and how often and long a user crosses the fair share limit plays a major role". Or in other words: just how it's done pretty much everywhere by pretty much every provider; the differentiating factors are hardware (e.g. slower or faster drives) and a providers policy, and Contabo's numbers strongly suggest that they are among the more generous players on the field.

    As I like to check providers statements by getting and looking at cold data I ran a simple test using fio which most here seem to consider "the reference" (and which is used by yabs too). Here you go:

    fio (v. 3.28) disk test with 64k block size, total 8G
    
    wr_seq_64k IOPS: min 354, max 32782, avg 10765.62 --  BW (avg)= 689.12 MiB/s
    wr_rnd_64k IOPS: min 1099, max 6436, avg 3252.45 -- BW (avg)= 208.27 MiB/s
    rd_seq_64k IOPS: min 18636, max 50264, avg 37837.93 -- BW (avg)= 2365.00 MiB/s
    rd_rnd_64k IOPS: min 11438, max 25387, avg 18744.02 -- BW (avg)= 1171.63 MiB/s
    

    First note that fio says that it actually wrote 32 GB IO, highly likely due to running 4 threads in parallel, each doing the 8 GB requested.
    Now, look at the avg IOps! Up to over 35k IOps. So kindly stop telling me BS about oh so brutal 3k IOps limits. In fact, I can tell you that some of those tests even crossed Contabo's burst limit and all but one were over the fair share limit.
    For the sake of completeness: I did another run right after the first one and I was indeed write IO limited - to an IO rate that didn't allow me to go beyond their burst rate, duh. If each and every VPS I happen to have at diverse providers were limited like that I'd still be a very happy customer because most providers would classify that limit as clear abuse and and, yes, I consider that abuse too (under normal conditions, not when benchmarking with a providers permission).
    "But you have a special test VPS!!!" you say? Nope, I don't. And in fact I've seen proof in the form of a complete hypervisor VM list on that node (obviously with potentially sensitive customer data except mine blacked out).

    But there actually is one "special" point with "my" node: there were problems with some drive adapter, so they moved my VM (and presumably all others on that node too) to a node that did not have that hardware issue. And that problem actually was the reason why my (personal, paid for) VPS became so lame. On the new - and (almost) fully occupied - node without the hardware issue I'm back to the old good performance numbers.

    But there still is an ugly big "BUT": what if Contabo played me and lied to me the whole time? I don't think they did, but as there are quite a few accusing Contabo of diverse evil, e.g. brutally overselling, IO rate limiting, and who knows what I'd prefer to be really sure - and I frankly told them so.

    Their reaction? -> "Hmmm, how about ordering another VPS with a different name, maybe the name of a friend, and then benchmarking the hell out of that VPS? When you are done, you lift the veil, tell us the customer number of that account and we'll balance the account".

    Wow! The mere fact alone that they are willing to do that, to allow me to really get a real "just some customer" VPS they don't know it's me, the benchmarker, behind it, quite clearly demonstrates their attitude as well as their confidence in their product! But of course, I'll do it, I'll open a "fake" customer account, benchmark the hell out of that VPS ... and then tell you about the outcome.

  • ArkasArkas Moderator
    edited September 2021

    @jsg said: wr_rnd_64k IOPS: min 1099, max 6436, avg 3252.45 -- BW (avg)= 208.27 MiB/s

    That looks a little odd, it's rather on the low side.

  • jsgjsg Member, Resident Benchmarker

    @Arkas said:

    @jsg said: wr_rnd_64k IOPS: min 1099, max 6436, avg 3252.45 -- BW (avg)= 208.27 MiB/s

    That looks a little odd, it's rather on the low side.

    I'm working on something that highly likely also will shine some light on and explain certain phenomena.
    Whatever, the numbers I have published come from an honest and fair run of fio with 4 jobs and io-depth of 64 (iirc that's very similar to the yabs way of calling fio).

    Thanked by 1Arkas
  • Germany VPS L

    # ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
    #              Yet-Another-Bench-Script              #
    #                     v2021-06-05                    #
    # https://github.com/masonr/yet-another-bench-script #
    # ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
    
    Tue 14 Sep 2021 09:37:10 PM CEST
    
    Basic System Information:
    ---------------------------------
    Processor  : AMD EPYC 7282 16-Core Processor
    CPU cores  : 8 @ 2794.748 MHz
    AES-NI     : ✔ Enabled
    VM-x/AMD-V : ❌ Disabled
    RAM        : 29.4 GiB
    Swap       : 0.0 KiB
    Disk       : 195.8 GiB
    
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 23.31 MB/s    (5.8k) | 331.27 MB/s   (5.1k)
    Write      | 23.32 MB/s    (5.8k) | 333.01 MB/s   (5.2k)
    Total      | 46.63 MB/s   (11.6k) | 664.28 MB/s  (10.3k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 3.42 GB/s     (6.6k) | 10.35 GB/s   (10.1k)
    Write      | 3.60 GB/s     (7.0k) | 11.04 GB/s   (10.7k)
    Total      | 7.03 GB/s    (13.7k) | 21.39 GB/s   (20.8k)
    
    iperf3 Network Speed Tests (IPv4):
    ---------------------------------
    Provider        | Location (Link)           | Send Speed      | Recv Speed
                    |                           |                 |
    Clouvider       | London, UK (10G)          | 550 Mbits/sec   | 597 Mbits/sec
    Online.net      | Paris, FR (10G)           | 550 Mbits/sec   | 592 Mbits/sec
    WorldStream     | The Netherlands (10G)     | 551 Mbits/sec   | 593 Mbits/sec
    Biznet          | Jakarta, Indonesia (1G)   | busy            | busy
    Clouvider       | NYC, NY, US (10G)         | 534 Mbits/sec   | 571 Mbits/sec
    Velocity Online | Tallahassee, FL, US (10G) | 499 Mbits/sec   | 525 Mbits/sec
    Clouvider       | Los Angeles, CA, US (10G) | 502 Mbits/sec   | 533 Mbits/sec
    Iveloz Telecom  | Sao Paulo, BR (2G)        | 483 Mbits/sec   | 519 Mbits/sec
    
    Geekbench 5 Benchmark Test:
    ---------------------------------
    Test            | Value
                    |
    Single Core     | 1025
    Multi Core      | 6688
    Full Test       | https://browser.geekbench.com/v5/cpu/9832600
    
  • bruh21bruh21 Member, Host Rep

    @baruna said:
    Germany VPS L

    # ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
    #              Yet-Another-Bench-Script              #
    #                     v2021-06-05                    #
    # https://github.com/masonr/yet-another-bench-script #
    # ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## ## #
    
    Tue 14 Sep 2021 09:37:10 PM CEST
    
    Basic System Information:
    ---------------------------------
    Processor  : AMD EPYC 7282 16-Core Processor
    CPU cores  : 8 @ 2794.748 MHz
    AES-NI     : ✔ Enabled
    VM-x/AMD-V : ❌ Disabled
    RAM        : 29.4 GiB
    Swap       : 0.0 KiB
    Disk       : 195.8 GiB
    
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 23.31 MB/s    (5.8k) | 331.27 MB/s   (5.1k)
    Write      | 23.32 MB/s    (5.8k) | 333.01 MB/s   (5.2k)
    Total      | 46.63 MB/s   (11.6k) | 664.28 MB/s  (10.3k)
               |                      |
    Block Size | 512k          (IOPS) | 1m            (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 3.42 GB/s     (6.6k) | 10.35 GB/s   (10.1k)
    Write      | 3.60 GB/s     (7.0k) | 11.04 GB/s   (10.7k)
    Total      | 7.03 GB/s    (13.7k) | 21.39 GB/s   (20.8k)
    
    iperf3 Network Speed Tests (IPv4):
    ---------------------------------
    Provider        | Location (Link)           | Send Speed      | Recv Speed
                    |                           |                 |
    Clouvider       | London, UK (10G)          | 550 Mbits/sec   | 597 Mbits/sec
    Online.net      | Paris, FR (10G)           | 550 Mbits/sec   | 592 Mbits/sec
    WorldStream     | The Netherlands (10G)     | 551 Mbits/sec   | 593 Mbits/sec
    Biznet          | Jakarta, Indonesia (1G)   | busy            | busy
    Clouvider       | NYC, NY, US (10G)         | 534 Mbits/sec   | 571 Mbits/sec
    Velocity Online | Tallahassee, FL, US (10G) | 499 Mbits/sec   | 525 Mbits/sec
    Clouvider       | Los Angeles, CA, US (10G) | 502 Mbits/sec   | 533 Mbits/sec
    Iveloz Telecom  | Sao Paulo, BR (2G)        | 483 Mbits/sec   | 519 Mbits/sec
    
    Geekbench 5 Benchmark Test:
    ---------------------------------
    Test            | Value
                    |
    Single Core     | 1025
    Multi Core      | 6688
    Full Test       | https://browser.geekbench.com/v5/cpu/9832600
    

    poor disk, but not bad for a contabo processor

  • jsgjsg Member, Resident Benchmarker

    @bruh21 said:
    poor disk, but not bad for a contabo processor

    Yeah right, because 300 MB/s at 64k blocks and multi GB/s at 512k and larger block sizes is sooo poor...

    Yet another case of someone having an "opinion" beforehand and stubbornly sticking to it even when ones favourite benchmark says otherwise.

    Yeah right, "yabs is such a great benchmark" - except when it doesn't confirm your biased "opinion".

  • @jsg said:
    Yeah right, "yabs is such a great benchmark" - except when it doesn't confirm your biased "opinion".

    Do you know the difference between observation and inference?

    Thanked by 2cybertech TimboJones
  • bruh21bruh21 Member, Host Rep
    edited September 2021

    @jsg said:

    @bruh21 said:
    poor disk, but not bad for a contabo processor

    Yeah right, because 300 MB/s at 64k blocks and multi GB/s at 512k and larger block sizes is sooo poor...

    Yet another case of someone having an "opinion" beforehand and stubbornly sticking to it even when ones favourite benchmark says otherwise.

    Yeah right, "yabs is such a great benchmark" - except when it doesn't confirm your biased "opinion".

    yes but the small block sizes are slower than they should be for an nvme with only 5.8k IOPS
    even SATA SSDs can beat that
    in fact, this HDD VPS i had a while back beat it in 4k:

    Fri Sep  3 23:08:13 EDT 2021
    
    Basic System Information:
    ---------------------------------
    Processor  : QEMU Virtual CPU version 2.5+
    CPU cores  : 8 @ 2799.998 MHz
    AES-NI     : ❌ Disabled
    VM-x/AMD-V : ❌ Disabled
    RAM        : 7.6 GiB
    Swap       : 1024.0 MiB
    Disk       : 983.3 GiB
    
    fio Disk Speed Tests (Mixed R/W 50/50):
    ---------------------------------
    Block Size | 4k            (IOPS) | 64k           (IOPS)
      ------   | ---            ----  | ----           ----
    Read       | 59.26 MB/s   (14.8k) | 306.60 MB/s   (4.7k)
    Write      | 59.39 MB/s   (14.8k) | 308.21 MB/s   (4.8k)
    Total      | 118.65 MB/s  (29.6k) | 614.81 MB/s   (9.6k)
    
  • jsgjsg Member, Resident Benchmarker
    edited September 2021

    @bruh21 said:

    @jsg said:

    @bruh21 said:
    poor disk, but not bad for a contabo processor

    Yeah right, because 300 MB/s at 64k blocks and multi GB/s at 512k and larger block sizes is sooo poor...

    Yet another case of someone having an "opinion" beforehand and stubbornly sticking to it even when ones favourite benchmark says otherwise.

    Yeah right, "yabs is such a great benchmark" - except when it doesn't confirm your biased "opinion".

    yes but the small block sizes are slower than they should be for an nvme with only 5.8k IOPS
    even SATA SSDs can beat that
    in fact, this HDD VPS i had a while back beat it in 4k:
    ```
    Fri Sep 3 23:08:13 EDT 2021

    Basic System Information:

    [yabs results]

    That means pretty much nothing. You need to run at least 20 (50 would be better) tests spread over min a day (incl night) and then compute min, max, avg. A single benchmark run on a VPS doesn't tell you the truth about performance.
    Also note that, as I've written, some Contabo nodes seem to have (had?) a hardware problem with disk related stuff. I myself saw a dramatic difference between a concerned node and a fixed one. Maybe your VPS is on a node they didn't fix yet (it seems to take quite some time to move all node inhabitants to another node). Maybe check with their support to be sure.

  • Contabo has IOPS limit - it’s a known fact. It won’t magically increase whether you run the benchmark during the day or the night. Contabo lied, you got a special one for reviewer - it was quite obvious. You just don’t want to accept that, and take back your words. It’s the cociu shenanigans again.

  • edited September 2021

    @jsg said:

    @bruh21 said:

    @jsg said:

    @bruh21 said:
    poor disk, but not bad for a contabo processor

    Yeah right, because 300 MB/s at 64k blocks and multi GB/s at 512k and larger block sizes is sooo poor...

    Yet another case of someone having an "opinion" beforehand and stubbornly sticking to it even when ones favourite benchmark says otherwise.

    Yeah right, "yabs is such a great benchmark" - except when it doesn't confirm your biased "opinion".

    yes but the small block sizes are slower than they should be for an nvme with only 5.8k IOPS
    even SATA SSDs can beat that
    in fact, this HDD VPS i had a while back beat it in 4k:
    ```
    Fri Sep 3 23:08:13 EDT 2021

    Basic System Information:

    [yabs results]

    That means pretty much nothing. You need to run at least 20 (50 would be better) tests spread over min a day (incl night) and then compute min, max, avg. A single benchmark run on a VPS doesn't tell you the truth about performance.

    I agree with multiple time running benchmark to get more accurate reading. But, doing it more than 3 times per day on a VPS is insane. You can compare the result with other user who already posted it.

    The difference between tests should not be significant. If you have significant difference then I suggest you to move away from your VPS povider because they unable to provide you a stable VPS.

    You can do it once per day to get rough performance. When you do it daily you can get daily time series of your VPS rough performance.

    BTW there is good quote:

    Insanity is doing the same thing, over and over again and expecting different results.
    
  • jsgjsg Member, Resident Benchmarker

    @muffin said:
    Contabo has IOPS limit - it’s a known fact. It won’t magically increase whether you run the benchmark during the day or the night. Contabo lied, you got a special one for reviewer - it was quite obvious. You just don’t want to accept that, and take back your words. It’s the cociu shenanigans again.

    Uhm, I myself openly wrote that, yes, they do have IO limits - as do btw. virtually all VPS providers and even the devices themselves! And in contrast to you I actually know their fair share and their burst limit and I can tell that all the alleged oh so evil low, low, low IO limits of Contabo are BS. The real limits are quite (I'd even say very) generous and far above what some allege.

    Plus, yes, maybe Contabo lied/gave me a special test VPS. I don't think they did but I can't exclude it. But now I'm buying a VPS with a wrong name and data and they definitely don't know it's me. In other words it's a "just some customer" VPS.

  • @jsg said:
    Uhm, I myself openly wrote that, yes, they do have IO limits - as do btw. virtually all VPS providers and even the devices themselves! And in contrast to you I actually know their fair share and their burst limit and I can tell that all the alleged oh so evil low, low, low IO limits of Contabo are BS. The real limits are quite (I'd even say very) generous and far above what some allege.

    Plus, yes, maybe Contabo lied/gave me a special test VPS. I don't think they did but I can't exclude it. But now I'm buying a VPS with a wrong name and data and they definitely don't know it's me. In other words it's a "just some customer" VPS.

    Maybe not you, but I wouldn’t call 5.1k limit for 4k GREAT and FAST for a NVMe. I would expect more, especially after reading their blog post https://contabo.com/blog/introducing-high-performance-vps-with-nvme-drives/

    @AXYZE said:
    22MB/s in 4K can be saturated even in Minecraft server with mods and in databases that have a lot of relational queries. If you put Minecraft and databases into one server (hey, 4+ cores!) then good luck xD They will fight for IOPS and Contabo knows this. By limiting 4K IO nobody will use CPU that much so people wont notice that they are heavily oversold.

    Kinda funny - in minecraft forums people are complaining that 10 cores from Contabo isnt giving them lag-free experiemce. Then they switch to CHEAPER 2/3/4core Hetzner/BuyVM/Nexusbytes/netcup plan and problems go away :)

    The limit is low, so much so that there are people who actually experience better performance when they move to the cheaper competitors, the competitors with less core, less ram, and have “slower disk” than Contabo. This more explains the vps performance in a real world scenario than a benchmark.

    To summarize Contabo, they are not that great, but are relatively cheap. Having said that, better yet similarly priced providers like netcup and Hetzner do exist in Europe, maybe not SG. I’m not against you, I think you are smart, but just too stubborn and kind - you gave cociu too much benefit of the doubt, and now you are doing the same thing to Contabo. I think you are the victim here. I’m just against Contabo who, in my, and probably many other people’s view, is lying about not giving you a special vps, and making up excuses about the node blah blah blah when accused. That alone makes me want to stay away from them no matter the price. They didn’t even bother responding to the thread when they could have to clear up any misunderstanding!

  • jsgjsg Member, Resident Benchmarker
    edited September 2021

    @muffin

    Let me start like this: I'm a NexusBytes fan, not a Contabo fan! Whenever I need a reliably good and fast VPS NexusBytes is the provider where my first look goes.
    I've come to like a few people at Contabo who've always been perfectly straight, honest, and constructive with me - even when I told them that I do not trust them enough to just blindly believe them and that my first loyalty is to LET.
    Their reaction? An offer to create an account with fake data and to buy a "Joe Anybody, just some customer" VPS and to later, after completing my test, tell them it was me and to get the account balanced. Any which way one turns that, that's definitely not the behaviour of a lying, tricky, shady provider.
    And liking some Contabo people != liking Contabo.

    As a matter of fact I've scrutinized Contabo more than any other provider so far and I'm still re-checking and re-verifying their claims. Btw, I pay for my private Contabo NVMe VPS.

    I think one major point that leads to contention here is that many of us are not even within Contabo's "prey scheme" (preferred and targeted customer segment). Their whole business is quite streamlined in every regard, incl. having a clear frame of what kind of customers and market segment they target. They do not target the 'lowest price" crowd, nor the nerd crowd but rather the Joe and Mary Anybody and small businesses and they try to be the "reliable solid, not expensive german quality provider". Note: That's my impression, not some information or statement from them.

    And it works very well! They run an increasingly global operation with a 6 digit number of VPSs.

    Joe and Mary Anybody or the restaurant at the corner or the small company do not need or look for 'x thousand IOps'. They want "plenty enough IOps", they want "plenty fast enough in every regard - and reliable and stable" to run their web site, their database or whatever. Plus good and fast support. And if it grows and gets slower they simply upgrade to a larger model, done.

    That's not what I want and that's why for myself I prefer NexusBytes who (at least also) addresses nerds and techies. NexusBytes uses Ryzens and Contabo uses Epycs, that already should tell us something. 'Epyc' translates to slower (than Ryzen) but more "enterprisy" and dual sockets (running thousands of nodes in multiple data centers is a different scale from most providers here and the whole approach and thinking is different).

    And indeed Contabo does not even care much about LET. They early on told me quite frankly that they like my work and have no problem with me publishing it here but what they were after is for internal use. Maybe some kind of an outside opinion and sometimes "pre-flight" checking. Good example: They were not at all angry about me saying that my NVMe drive got crappy slow, quite the contrary, they seemed to be (kind of) happy that a potential problem could be spotted thanks to my benchmarking and feed back.

    So, we are wrong judging Contabo only by LET standards because we are not the segment they target. And as far as I can tell (and I did some checks) they always were honest and open with me - although they know that for myself (and probably many at LET) they are not a preferred provider. And they really try hard to be among the the best providers in their segment. That includes them welcoming criticism and occasionally hard questions from me; they were not looking for softball but for useful good feedback.

    Anyway, soon I'll have first "anonymous" results from my "Joe Anybody, just some customer" VPS. Then we'll see and have proof whether they lied to me or not.

  • @AXYZE said:

    @yates said:
    Thanks for the recommendations! I was actually specifically looking for the most budget option, haha (I wanted a cheap set-up to play around with). But well, you get what you pay for (or less than that)...I've had a look at netcup, I like that they're using Ryzen GPUs in their more expensive package.

    Well I dont know what specs you need, but also check Scaleway Stardust. Its 0.37euro/mo if you remove flexible ip/ipv4.
    1vCore AMD EPYC
    1GB ram
    200Mbps-1Gbps network
    10GB NVMe
    Amsterdam location

    Good enough for basic game server, web server...

    The nice trick is to remove the flexible IPv4, however the system won't be able to access public IPv4 unless you have some gateway or vpn to translate from IPv6. The cloudflare warp vpn works, but still, kind of annoy if you'd like to host something other than http over IPv4, such as game server or teamspeak server.

    It's could be interesting if we can setup a shared NAT64 at scaleway and provides IPv4 port forwarding service.

  • Don't get me started about how shit contabo is. Ordered a VPS in singapore for the first hour everything was normal ssd, bandwidth speed etc. After an hour speed was 200kbps from fast.com phoned and messaged Contabo and said the speed is crap and provided them will all traceroute etc. There response. We are not responsible for speeds that is the issue of where you're downloading from. Well from my experience fast.com and speedtest.net don't lie.

Sign In or Register to comment.