Howdy, Stranger!

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


Pulsed Media MiniServers // Transcoding Servers // Streaming Servers!! NVMe+1G Unmetered FROM 29.99€ - Page 3
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.

Pulsed Media MiniServers // Transcoding Servers // Streaming Servers!! NVMe+1G Unmetered FROM 29.99€

13567

Comments

  • I've never said you could replace your routers with VPP but the video I linked is from a guy with decades of experience in the field and one of the users and developers of VPP are the partying basement dwellers form Cisco. And the same no money and no experience guys at Cisco say they can do 480Gbit/s (12x40GE) with zero-packet-loss on VPP with 32 Haswell cores and an average latency of <23 usec with a 2 million FIB. While FIB size has zero impact on performance. https://www.cisco.com/c/dam/m/en_us/service-provider/ciscoknowledgenetwork/files/592_05_25-16-fdio_is_the_future_of_software_dataplanes-v2.pdf and this was back in 2016.

    Again I'm not saying this is feasible for you and the big carrier grade routers definitely have their place especially regarding reliability and if you have to do multiple Tbit/s but just laughing at software solutions isn't appropriate anymore IMHO.

  • PulsedMediaPulsedMedia Member, Patron Provider

    @babuum my beginning reply was more towards the others.

    There's been great progress on software routing indeed, there is no denying it indeed, i would just not trust our business to it.

    Regardless, as for the basic routing capacity -> Our gear already is orders of magnitude better, it's just the small FIB which is on everything this grade which is the problematic bit.

    It's small circle of people who understands this level of gear, and info is sometimes hard to come by, sometimes even simple specs because the configs are infinite, and slightest mistake and you got it wrong.

    This is all besides the point tho, we will add IPv6 only when we are good and ready, HW costs are not the issue, it's the management overhead with very very very little demand for IPv6. So business case has to be there 1st. It is not there yet, and might not even ever get there before something saner comes along.

    At the end of the day, these dedis are excellent and very very good, and you can use IPv6 no probs if you need to, quite trivial infact.

  • PulsedMediaPulsedMedia Member, Patron Provider

    on another note, just single LACP group for us is showing 12+Bpps 1day averages lasting for weeks.

    So 300Mpps @ 13k € (did not watch video fully) @ '16, could probably do 3Bpps these days, so theoretically to get to the 12+Bpps (which we did not even notice!) would need about 52k € investment (if hardware existed for that! It does not). Also i think our machine rating is per port since this level gear advertise worst of the worst of the worst case. Or perhaps per "port to port" stream. Who knows how they did the ratings, since according to spec sheet it should not be able to do 12+Bpps, yet it regularly does.

    All this thread did was new found appreciation for our current gear and highend enterprise routing HW.

    Might actually buy another one for the next DC, but ... slightly larger, 1 rack sized unit as DC core :) empty router is ~250kg, and specced for 14kw utilization, will probably windup using about 8-12kw. So needs probably 10-20KWh battery + 15KW peak power rating by itself by itself.

    This would have 100+ GiB of packet buffer alone and be capable of hosting full (IPv4) routing table in FIB by itself, but not large enough FIB to justify as edge investment at this point. Will have to do 10year TCO analysis if this model or something more in the line of 500k € (so first cheaper temporary and upgrade later on)

    This stuff is bought to be utilized for decades, not for a few years :)

  • Why do you need to do ipv6 full table when you only have a single upstream?

  • crunchbitscrunchbits Member, Patron Provider, Top Host

    The derailment was a fun read.

    It does seem like software routing has come quite a long way, but still a ways to go when we're talking MX-tier (asic) routing. Did some experimenting with TNSR and a bunch of 100G cards. It seemed solidly in "campus/small or medium office" tier. Same as other software options, they're claiming they can push pps in excess of ~10Gbps per Xeon core. That might be true, but it didn't scale all the way up to 100G (and beyond) despite making sure all gear was on their compatibility list. Hard to say exactly, but I think you can probably hit and sustain ~40G line rates with TNSR.

    Anyways back on track: QuickSync really is amazing. It sounds like even on some much older E3's that we swapped over to iGPU (12*5 variants) for QuickSync the results were extremely favorable.

  • PulsedMediaPulsedMedia Member, Patron Provider

    @crunchbits said: but I think you can probably hit and sustain ~40G line rates with TNSR.

    and then you get a tiny DDOS as packet flood and your legit traffic throughput plummets :(

    @crunchbits said: Anyways back on track: QuickSync really is amazing. It sounds like even on some much older E3's that we swapped over to iGPU (12*5 variants) for QuickSync the results were extremely favorable.

    easily underappreciated. I know initially we did too :)

    Infact, these have been all around very surprisingly high performance. So way more is en route :)

  • With "latest Ubuntu" I hope that is the latest LTS right?

    @andrewnyr said:
    Why do you need to do ipv6 full table when you only have a single upstream?

    There is also a middle ground if routing size is a problem. You can combine a partial table (the direct routes of your peers) and default routes like described here https://aboutnetworks.net/bgp-load-sharing/ to reduce the routing table size considerably. I know an ISP that did that when the v4 table reached 512K and some of the routers they used couldn't hold the full table anymore. I can't find the presentation anymore but it only had minimal impact on AS_PATH length, latency and throughput.

  • PulsedMediaPulsedMedia Member, Patron Provider

    @babuum yea there are many things you can do to lessen the FIB burden.

    This is not the limiting factor for us once more, and never was. I guess trying to teach people about the reality of things is what these zealots hate.

    We will not add something that adds 2-6% mgmt penalty across the board for the benefit of 0.1-0.5% without a solid business case, period. No matter the amount of zealotry. We are data driven business, not emotion.

    Meanwhile, these nodes are kickass and in short supply.

    Next batch of ~60 nodes is expected to arrive in like 3 months, meanwhile there might only be 8 or 16 added. Occupancy rate is already quite high without any marketing push what-so-ever apart from this thread and giving LET users essentially first chance on these.

  • PulsedMediaPulsedMedia Member, Patron Provider

    Our current edge router maxes out at 2.4M in FIB. This will be exhausted very soon, within 2 years, if not 1 year if using full table for IPv4 + IPv6.

    IPv6 table already takes as much as IPv4, so we are right up at about 2M routes in FIB total (depends a bit of your peers, so +/- 15% i would guess depending on your network)

    New router has already been ordered Q4/22, only info about that in regards of FIB is "Enough", at least 6.5M in FIB. At current growth pace that's about 5 years only when full IPv6 routing table would exhaust that as well.

    If you only do IPv4 stack, the 2.4M FIB would be enough for another 10 years.
    6.5M FIB would be enough for ~22 years ... probably more.

    Even worse thing for IPv6:
    Routing is different between the two, so you will have very hard to debug performance issues with IPv6 compared to IPv4. With IPv4 sometimes debugging can be ridiculously difficult, most often is.

    But now you have 2 completely different routing going on, 100% guarantee people will not care or know if packets go through IPv4 or IPv6. Adding extra debugging time by highly skilled (==expensive) people wasting time debugging for IPv4 only for customer having issue with IPv6. Even more time waste; IPv4 routes are ridiculously well tested since has 100% usage and IPv6 has barely any use and completely unreachable sections of internet.

    Funny things noticed today

    • It looks like Starlink uses IPv6 for their routers / satellites, but does not use IPv6 for end users and IPv6 is not available to end users, it's tunneled IPv4.
    • Yet again getting reports people wanting to remove IPv6 from their services as it causes nothing but trouble with no benefits.
    • Telia (Yes, THAT Telia) does not support IPv6 to their broadband or mobile customers

    IPv6 Future outlook?

    I do not expect this situation to change, and do not expect IPv6 to be fully supported ever. Infact, my bet is that IPv6 will "die off" eventually, remain in the margins forever, it will not outright be removed after the hundreds of billions invested in this faulty protocol, but it will never be 100% mainstream neither.

    There will be some cases where adding IPv6 as a compliment will make perfect sense, but those are few and far between. Example: Starlink routing, and NAT'd servers.

    Something better than IPv6?

    Something simpler will replace it and smaller than /24 announcements (perhaps all the way down to /28) will become the norm. Most likely a form of IPv4e // IPv4+.

    I've been advocating for a decade now a simple solution; Just add either 2 static octets, or always allow the final /32 to decide if another octet behind it is available and packet should be forwarded there. If you allow this dynamically, the only question then remains how much space in the IPv4 packet header there is to add more routing octets ...

    Only end devices and network segments where you want to support it needs router upgrades. ZERO changes to core routers, ZERO changes to routing tables (apart from larger size if smaller than /24 starts to be accepted).

    Only support critically needed is client devices, these will automatically update over time, question is how many years backwards compatibility should be given, 10 years? 15years? When we can say that 100% coverage for IPv4e (someone will forever use Win3.11, Win95 or WinXP somewhere, that super marginal last 0,25% simply does not matter, they can just use plain IPv4 if they insist on using multidecade old OS's it's their headache).

    After client devices have support, you can end service side support on a as needed basis, without zero difference to rest of your network.

    How to use IPv6 right now on Pulsed Media Dedis?

    But Do not attempt to force Your zealotry to others, be kind to everyone. Not everyone else has to suffer because you think you need, or simply have an emotional need for something overtly complex, nearly completely unsupported most complicated thingy which adds very little to average user apart from cost, overhead and headache. (AKA service value just decreased)

    Sidenote; Running other distros on these Dedis before we add baremetal images

    You can install Proxmox in a few quick sets, then setup NAT and run a KVM image with anything you want. Very easy and fast.

    Something like this gets you started, run as root:

    echo "deb [arch=amd64] http://download.proxmox.com/debian/pve bullseye pve-no-subscription" > /etc/apt/sources.list.d/pve-install-repo.list
    wget https://enterprise.proxmox.com/debian/proxmox-release-bullseye.gpg -O /etc/apt/trusted.gpg.d/proxmox-release-bullseye.gpg 
    apt update && apt full-upgrade
    apt install pve-kernel-6.1 proxmox-ve postfix open-iscsi whois inetutils-tools vnstat net-tools moreutils  ifenslave ifupdown2  -y 
    

    NAT setup:
    https://pve.proxmox.com/wiki/Network_Configuration
    https://forum.proxmox.com/threads/pve-6-2-private-vm-nat-network-configuration-setup.71038/

    BE VERY VERY CAREFUL, preferrably do a test config, ie. run ifdown ifup in screen, add 5 minute sleep and after the sleep copy back old config rerun ifdown ifup and reboot (in case ifdown ifup fails). If you press ctrl+c during the sleep the last portion will not be executed and config is retained.

    then download your distro / OS ISO to /var/lib/vz/template/iso and just add your VM :)

    Takes about 10-15minutes do first time i think.

    Sidenote for a sidenote, on topic of IPv6: See https://github.com/bruj0/ProxmoxIPv6 for fully routed config and ironically next to that search result was: https://forum.proxmox.com/threads/what-do-i-need-to-do-to-disable-ipv6.42466/

    There you go, several methods to use IPv6 without making everyone suffer and method of running any OS you'd like to have and even extend the capabilities of your server. Infact, it is far more likely we do ready Proxmox + NAT premade setup before any other distro

    now you just have to:

    Thanked by 1maverick
  • bakliakihrbakliakihr Member
    edited March 2023

    Hey, since I've received the mini dedi I've enjoyed it so far. I had a couple questions that I didn't feel were important enough to bother with a ticket though and may be of public benefit to ask here so others can know the answers too:

    Is there any sort of ETA on automated reboot and rescue functionality? It seems at the moment I must ticket in if I need even a hard reboot, let alone to rescue my OS.

    I know you said Ubuntu is the next OS to be added, but is there any timeline for that?

    Another thing is unless I'm mistaken you may be manually installing the OS(saw some commands run in history on first login), is it not therefore possible to request Ubuntu in the short term while we await some automation solution? It isn't any harder to do than Debian IMO if you're manually installing that.

    I don't mean to be rude or anything by any of these questions, I really like the price point and hardware, but for my uses Debian actually has too old of packages and unless it is possible to install Ubuntu or some other OS I may not renew next month as Debian only with no automated reboot/rescue is problematic for me. (I'm aware earlier you suggested using Proxmox as a "workaround" for other OSes, but for me it won't work)

    :)

  • PulsedMediaPulsedMedia Member, Patron Provider

    @bakliakihr said: Is there any sort of ETA on automated reboot and rescue functionality? It seems at the moment I must ticket in if I need even a hard reboot, let alone to rescue my OS.

    Planned, no ETA.

    @bakliakihr said: I know you said Ubuntu is the next OS to be added, but is there any timeline for that?

    No ETA. Proxmox might be actually before that.

    @bakliakihr said: Another thing is unless I'm mistaken you may be manually installing the OS(saw some commands run in history on first login), is it not therefore possible to request Ubuntu in the short term while we await some automation solution? It isn't any harder to do than Debian IMO if you're manually installing that.

    It's automated install, we must manually ensure latest patches prior to delivery and swap the PW, so nodes are all preinstalled and online before delivery.

    @bakliakihr said: I don't mean to be rude or anything by any of these questions, I really like the price point and hardware, but for my uses Debian actually has too old of packages and unless it is possible to install Ubuntu or some other OS I may not renew next month as Debian only with no automated reboot/rescue is problematic for me. (I'm aware earlier you suggested using Proxmox as a "workaround" for other OSes, but for me it won't work)

    you can dist-upgrade to Deb12 (actually adding Deb12 is trivial).
    We actually have rescue mode too but it first needs the remote reboots.

    Speaking of which, a module, backend code etc. has to be done to add that, so what sounds like "just 30-60mins of configuring" is actually "30-60hrs of high intensity, undisturbed development hours" -- if doing haphazardly and quickly.

  • MumblyMumbly Member
    edited March 2023

    @PulsedMedia said: But Do not attempt to force Your zealotry to others, be kind to everyone. Not everyone else has to suffer because you think you need, or simply have an emotional need for something overtly complex, nearly completely unsupported most complicated thingy which adds very little to average user apart from cost, overhead and headache. (AKA service value just decreased)


    ..and yes, IPv6 requirement is completely legit requirement. If you can't/don't want to offer it for some reason that's solely on you. It's your personal decision how to run business and that's fine, but stop acting like you're a victim or something. It's open market.

    Thanked by 1babuum
  • PulsedMediaPulsedMedia Member, Patron Provider

    remember to grab your dedi before these get sold out :)

  • PulsedMediaPulsedMedia Member, Patron Provider
    edited March 2023

    long term discounts for these has been added, -5% annually, -10% 2 years, -15% 3 years.

    By end of summer we will TRIPLE the number of these. Next platform design is done, and nodes are already in freight.

    On other news, Storage Seedbox solutions has been updated, 4-16TB RAID0, SSD Cached. Storage dial turned to 11

  • PulsedMediaPulsedMedia Member, Patron Provider

    New model added, MD4: i5 7500t/16G/500-512GB NVMe.
    Current price 29.99€ see https://pulsedmedia.com/minidedi-dedicated-servers-finland.php

  • PulsedMediaPulsedMedia Member, Patron Provider

    New model added, MD5: i5-7500t / 32G / 4000GB NVMe
    1Gbps Unmetered.

    Current price 39.99€ see https://pulsedmedia.com/minidedi-dedicated-servers-finland.php

  • emghemgh Member

    My phone carrier is called HI3G and offers me 5G gigabit IPv6 unlimited data

    Thanked by 1inthecloudblog
  • emghemgh Member
    edited July 2023

    Also, isn’t this all not that much of a technical disagreement but more that @ralf & @yoursunny are fine with 0.1-1 ms (wide range to not upset anyone) higher response time for the few route lookups not in SRAM while @PulsedMedia isn’t?

    (Don’t wanna start anything again hehe, just noticed that this seems to be the actual curlpit in the disagreements)

  • PulsedMediaPulsedMedia Member, Patron Provider

    @emgh said: Also, isn’t this all not that much of a technical disagreement but more that @ralf & @yoursunny are fine with 0.1-1 ms (wide range to not upset anyone) higher response time for the few route lookups not in SRAM while @PulsedMedia isn’t?

    Assuming back to the sheer idiocracy prior about software routing, and how routers are easy to build.

    Latency is part of it. If 1ms is added on the whole route to china for each route lookup that will add probably something like 50ms to RTT. Also the sheer amount of data and packets which is even bigger issue.

    Current dual CPU highest end setup, everything maxed out, nitrogen cooled and overclocked the snot out could not hold a candle for decade old mid-range switch.

    I think only now 400Gbps is starting to become reality per CPU.

    Still waiting for the magical weekend router build which outperforms anything on the market at 1/10 000th the cost. Guys, if you can build it and it's so easy you can do it in college weekend booze fest haze, you'll be billionaires. So go do it, since it's so easy ^_^

    Thanked by 1emgh
  • j3rj3r Member

    I want to need this. Cool deal.

    Thanked by 1PulsedMedia
  • @PulsedMedia said: 1Gbps Unmetered

    bandwide unmetered or have fair usage...?

  • PulsedMediaPulsedMedia Member, Patron Provider

    @horasjey said:

    @PulsedMedia said: 1Gbps Unmetered

    bandwide unmetered or have fair usage...?

    Yes, unmetered. Use as much as you want to :)

    Detail:

    Right now still we have to watch out if there is serious abuse among multiple nodes. Bandwidth share is 20Gbps for 48 units, so in practice it is fully unmetered, just not fully dedicated 1Gbps.

    For all practical purposes this is fully unmetered, only abuse will get curtailed. Say full 1Gbps 24/7, or other very high abnormal BW usage, and even then only if it causes issues for other users.

    Once we hit scaling the bandwidth will be even further increased to practically dedicated, guaranteed, business grade 1Gbps. We do not ever expect to run out of bandwidth, or even come near to the limits.

  • PulsedMediaPulsedMedia Member, Patron Provider

    @j3r said:
    I want to need this. Cool deal.

    Time for self hosting stuff? ;)

    Install proxmox, docker and go wild! :)


    COMMENT YOUR INVOICE # BELOW AND WE WILL EXTEND 1 FREE MONTH TO YOUR SERVICE!

    Offer valid until 14th of August OR to maximum of 30 free months given out

    Thanked by 1ehab
  • ShadosShados Member

    Invoice #178589. Also, just opened a ticket asking for a manual power cycle of the server :sweat_smile:.

    Thanked by 1PulsedMedia
  • PulsedMediaPulsedMedia Member, Patron Provider

    @Shados said:
    Invoice #178589. Also, just opened a ticket asking for a manual power cycle of the server :sweat_smile:.

    +1 month added to your service.
    See ticket reply.

    Thanked by 1Shados
  • artxsartxs Member

    @PulsedMedia

    Invoice #178693

  • PulsedMediaPulsedMedia Member, Patron Provider
    edited July 2023

    @artxs said:
    @PulsedMedia

    Invoice #178693

    Thank You.
    +1 Month added.

    EVERYONE ELSE Even if you ordered prior to posting that above, just comment invoice # here and you get +1 month, if order was after 1st of July. No need to be shy :)

  • Invoice #178451

  • SamidareSamidare Member
    edited July 2023

    Invoice #178114

    I'm not sure if it also applies to July 4th deal but thanks anyway.

  • Possible to get Windows installed on these machines?

Sign In or Register to comment.