Howdy, Stranger!

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


Question about network speed conversion (MB/s to mbps)
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.

Question about network speed conversion (MB/s to mbps)

fanfan Veteran
edited January 2023 in General

Recently I encountered a provider that advertises 1Gbps port speed but caps the network port at 100MB/s (equivalent to 800Mbps in real world), who also insisted there is nothing wrong to do such a conversion.

Is this the VPS's version of "It depends on what the meaning of the word 'is' is.", or just common practice?

Comments

  • TrKTrK Member

    Port speed != Bandwidth speed
    or did i miss something?

  • It is MB and Mb. Not MB/s and mbps or mb/s.

    Please use correct SI prefix.
    M = mega (MB = megabyte)
    m = mili (ml = milliliters)
    MB = megabyte
    Mb = megabit

    Mbps = Megabits per second
    Mb/s = Megabits per second

    MBps = Megabytes per second
    MB/s = Megabytes per second

    There is no difference between "/s" and "ps" too, they mean the same thing - per second.

    So getting to your question - if provider says 1Gbps it means 1000Mbps, because in SI prefixes 1 giga = 1000 mega.

    Although in that instance I think that there was misunderstanding with provider and provider stated that 1Gbps is burst speed and 800Mbps guaranteed. Could you provide screenshots for clarification?

  • fanfan Veteran

    @AXYZE said: So getting to your question - if provider says 1Gbps it means 1000Mbps, because in SI prefixes 1 giga = 1000 mega.

    Although in that instance I think that there was misunderstanding with provider and provider stated that 1Gbps is burst speed and 800Mbps guaranteed. Could you provide screenshots for clarification?

    Edited, thanks for the correction.

    Here's a direct quote from the provider:

    We allocate each VPS with 1 TB bandwidth on 1 Gbps port.

    and they've confirmed that the actual cap is 100MB/s.

  • stefemanstefeman Member
    edited January 2023

    The issue here is that 1Gbps translates to 1000 Mbps which is fine, but you should still get 125 MB/s from the server itself. Otherwise its just 838 Mbps which means you are missing some bandwidth that should be there.

    Seems like the provider does not understand conversion too well between bits and bytes or just lazily set the limit to "100" in the management panel.

    @fan name and shame please.

    edit: looks like its https://v.ps/faq/

    @xTom explain please :D

  • fanfan Veteran

    @stefeman said: name and shame please.

    Too early to shame I think, but indeed it's the first time I see such conversion among providers.

  • AXYZEAXYZE Member
    edited January 2023

    @fan said:

    @AXYZE said: So getting to your question - if provider says 1Gbps it means 1000Mbps, because in SI prefixes 1 giga = 1000 mega.

    Although in that instance I think that there was misunderstanding with provider and provider stated that 1Gbps is burst speed and 800Mbps guaranteed. Could you provide screenshots for clarification?

    Edited, thanks for the correction.

    Here's a direct quote from the provider:

    We allocate each VPS with 1 TB bandwidth on 1 Gbps port.

    and they've confirmed that the actual cap is 100MB/s.

    So the cap should be 1Gbps = 1000 Mbps. You should expect 1000Mbps cap, their calculations don't matter.

    Be aware that getting full 1000Mbps is best case scenario as speed depends on protocol, protocol overhead, routing, how many parallel connections etc.
    For example in TCP single stream I get just 50Mbps to Hetzner Finland (1Gbps dedi), but with multistream it scales up to ~980Mbps.

    Do speedtest-cli test in order to check if there is 800Mbps cap. If there is then provider lies about 1Gbps port per each VPS.

  • @stefeman said:
    The issue here is that 1Gbps translates to 1000 Mbps which is fine, but you should still get 125 MB/s from the server itself. Otherwise its just 838 Mbps which means you are missing some bandwidth that should be there.

    I guess it's less true nowadays, but when I mentally perform this calculation myself, I also do the same, dividing by 10 instead of 8.

    Partially, this is because it's easier, historically it used to be accurate over point-to-point serial connections (if you consider start and stop bits), but even in real use there's quite a lot of overhead in packet headers in IP.

    Every IPv4 packet has a 20 byte header, IPv6 has a 40 byte header. UDP adds another 8 bytes, TCP adds another 20 bytes. There can be optional header data that increases both. The ethernet headers add another 42 bytes on top of that. Another 4 bytes for VLAN.

    So, a single TCP over IPv4 packet will have at least 82 bytes of overhead. If you consider that a packet is normally maximum 1500 bytes, so that's 5.5% overhead at best. For smaller packets, it's far more significant a slice.

    Of course, some networks support jumbo frames (packets larger than 1522), but many don't and so you can expect packet fragmentation (and yet more header overhead) if you try to use them.

    All in all, as a general rule-of-thumb, considering 1000Gbp/s line speed to give you approximately 100 MB/s of application throughput doesn't particularly controversial to me.

  • stefemanstefeman Member
    edited January 2023

    @ralf said:

    @stefeman said:
    The issue here is that 1Gbps translates to 1000 Mbps which is fine, but you should still get 125 MB/s from the server itself. Otherwise its just 838 Mbps which means you are missing some bandwidth that should be there.

    I guess it's less true nowadays, but when I mentally perform this calculation myself, I also do the same, dividing by 10 instead of 8.

    Partially, this is because it's easier, historically it used to be accurate over point-to-point serial connections (if you consider start and stop bits), but even in real use there's quite a lot of overhead in packet headers in IP.

    Every IPv4 packet has a 20 byte header, IPv6 has a 40 byte header. UDP adds another 8 bytes, TCP adds another 20 bytes. There can be optional header data that increases both. The ethernet headers add another 42 bytes on top of that. Another 4 bytes for VLAN.

    So, a single TCP over IPv4 packet will have at least 82 bytes of overhead. If you consider that a packet is normally maximum 1500 bytes, so that's 5.5% overhead at best. For smaller packets, it's far more significant a slice.

    Of course, some networks support jumbo frames (packets larger than 1522), but many don't and so you can expect packet fragmentation (and yet more header overhead) if you try to use them.

    All in all, as a general rule-of-thumb, considering 1000Gbp/s line speed to give you approximately 100 MB/s of application throughput doesn't particularly controversial to me.

    You can still get 116 MB/s with 1Gbps with all of the above considered. This is based on my own real world tests with Proxmox over public WAN.

  • Well

    Rules for Rounding
    Here's the general rule for rounding:

    If the number you are rounding is followed by 5, 6, 7, 8, or 9, round the number up. Example: 38 rounded to the nearest ten is 40.
    If the number you are rounding is followed by 0, 1, 2, 3, or 4, round the number down. Example: 33 rounded to the nearest ten is 30

    Maybe they rounded it to the nearest Gbps? Then it's correct even if you get only 500 Mbps, because it can be rounded up to 1 Gbps. :wink:
    Anyway, I would say rounding up to 1 Gbps from 500 Mbps and selling it as 1 Gbps is cheating.

    Just saying. This is not the case OP asked about. Just one example of extremes. :smiley:

  • crunchbitscrunchbits Member, Patron Provider, Top Host

    We sell as X Gbps since that is more commonly understood, but on the back-end (and in our virtualization platform) the limits are in MBps. These are visible to customers and we still get ~1 ticket a month asking why they have "125Mbps" instead of 1Gbps.

    Of course, it's 125MBps limit for a 1Gbps plan. The wording is definitely a little confusing to the uninitiated, though.

    Thanked by 2fan raza19
  • @ralf said:

    @stefeman said:
    The issue here is that 1Gbps translates to 1000 Mbps which is fine, but you should still get 125 MB/s from the server itself. Otherwise its just 838 Mbps which means you are missing some bandwidth that should be there.

    I guess it's less true nowadays, but when I mentally perform this calculation myself, I also do the same, dividing by 10 instead of 8.

    The math really isn't that hard to be cutting corners like that and being wrong by 20%. If you want to go the other way and give 20% more for free (ie, My Internet connection used to be advertised as 100Mbps but was ~118Mbps through over provisioning), then great. But to advertise higher and in reality offer lower, is nonsense.

    All in all, as a general rule-of-thumb, considering 1000Gbp/s line speed to give you approximately 100 MB/s of application throughput doesn't particularly controversial to me.

    No. The general rule of thumb is 8:1, not 10:1. No knowing 125MB/s = 1Gbit isn't something that someone in their line of work should fuck up. This is a huge red flag that this provider is Mickey Mouse Amateur Hour. Intentionally limiting performance? Don't care for any of that shit.

    I'd move services away if I was OP.

    Thanked by 1fan
  • fanfan Veteran

    @AXYZE said: Do speedtest-cli test in order to check if there is 800Mbps cap. If there is then provider lies about 1Gbps port per each VPS.

    Speedtest-cli says it's 800Mbps: https://www.speedtest.net/result/c/658ae6d6-aead-4b04-8c3b-01fe79549ccd

    For my usage 800Mbps is good enough, but this will surely leave a bad taste in my mouth.

  • MaouniqueMaounique Host Rep, Veteran

    @fan said: For my usage 800Mbps is good enough, but this will surely leave a bad taste in my mouth.

    It would be "good enough" for most people, so the provider is "selecting" the gullible customers against the knowledgeable ones who know what they need.

    The 1 TB is also a serious limitation so I suppose the provider is buying expensive traffic from somewhere.

  • fanfan Veteran

    Update:
    @xTom did the right thing to correct this issue, now it's full 1Gbps.
    https://www.speedtest.net/result/c/adbb33ce-2d39-4e8e-8b87-9a425c4c3ed8

    Thanked by 1xTom
  • emgemg Veteran

    @ralf said:

    I guess it's less true nowadays, but when I mentally perform this calculation myself, I also do the same, dividing by 10 instead of 8.

    Partially, this is because it's easier, historically it used to be accurate over point-to-point serial connections (if you consider start and stop bits), but even in real use there's quite a lot of overhead in packet headers in IP.
    [...]

    I was planning to write the exact same two paragraphs, but continued scrolling and saw @ralf's post.

    Like Ralf, I have always used 10 bits per second = 1 byte per second for communications data rate estimates as a general rule, for the same reason. It is a holdover from the days of working with serial interfaces.

    The funny thing is that even with Ethernet, the 10 bits / byte for network throughput works well enough. It works for estimating when a download will complete for a file whose size you know in bytes, but the download speed is shown in bits per second. Precision is less important there.

    I would be one of the "gullible" customers, mostly because I use the same estimation system and would not question it from them. That said, I expect to see network data rates shown in bits per second, not any form of "bytes per second".

    (For anything other than communications data rates, I use 8 bits = 1 byte. In case it matters, I have worked on processors with unusual word sizes, such as one type with 18 bit words.)

    Nobody has mentioned the 1000 vs 1024 debate yet.

  • raza19raza19 Veteran
    edited January 2023

    The wealth of knowledge in this thread alone is a perfect demonstration of why I have always loved the lowend-ecosystem. While many gently explain the MB to Mb conversion some have gone ahead and detailed beautifully the protocol overhead. Frankly while I have always used multiple streams to download stuff I never actually realized wt were the limiting factors of a single stream. Also, I feel ecstatic understanding why some hosts ask u to use jumbo frames when making an internal network between 2 vps ie reducing overhead.

    The point of mentioning all this is that I propose admins create a specific "KnowledgeBase" type category here at LowEndTalk where only an admin or a mod can deem a post worthy enough to be moved into the "KnowledgeBase" category. I am aware that there is a tutorials category but the knowledgebase(or wtever name admins come up) category will suit a lot of other purposes that don't really fit in the tutorials or general.

    @Arkas @FAT32 @raindog308

Sign In or Register to comment.