Coffeehouse Thread

10 posts

Forum Read Only

This forum has been made read only by the site admins. No new threads or comments can be added.

Network cards need cooling now, apparently...

Back to Forum: Coffeehouse
  • User profile image
    Tom Servo

    I just got my new gigabit NIC, but the PCIe variant. Having had my share of different gigabit NICs, I'm kind of surprised that this thing needs cooling, though the feature set is more or less the same as my old one. What the hell?

    Image hosted by ImageSocket.com

  • User profile image
    Massif

    You should see the KillerNIC.

  • User profile image
    Tom Servo

    Well, the Klingon NIC... It's promised speedups are still snake oil, but the same time probably a gift from heaven for cheaters, since it allows to run Linux applications on it.

    Unless you can explain to me how a Linux based environment, that introduces another point of latency, can magically improve on a few kilobytes of traffic between the computer and the uplink router, this NIC is a piece of crap.

  • User profile image
    figuerres

    Tom Servo wrote:
    Well, the Klingon NIC... It's promised speedups are still snake oil, but the same time probably a gift from heaven for cheaters, since it allows to run Linux applications on it.

    Unless you can explain to me how a Linux based environment, that introduces another point of latency, can magically improve on a few kilobytes of traffic between the computer and the uplink router, this NIC is a piece of crap.


    Hmmm running a linux app on a card that can muck with a pc....

    I can picture an exploit that loads a "trojan" into the card and makes it do things not intened by the mfg of the card:

    log traffic and send it to a hacker.
    snoop for passwords.
    run a hidden server -- like an FTP server.

    infect the pc with a virus / trojan

    ahhh soo many fun things one might be able to do with a PCI card you can hack!

    on the other hand Imagine using them to build a "Mega Router"
    a pc running linux with a full stack of them cards and ... ?

  • User profile image
    jolson88

    Actually, I think you'll see this becoming more important in Vista and Longhorn Server (cooled network cards, that is). The complete network stack has been rewritten from the ground up, and there's some pretty cool new features that come along with that.

    One of them is network offloading. Think "graphics accelleration", only for networking. What it allows you to do is offload network processing onto the network card rather than burdening the CPU with the processing. It's actually pretty cool stuff. There's a whole slew of other improvements besides that as well. Because of this, there will be more chip activity on the network card itself (most likely there will be an actual CPU on the network card I'd imagine), which of course, in-turn, would cause more heat to be generated (hence the cooling fans).

    Just my two cents Smiley.

  • User profile image
    Angus

    Massif wrote:
    You should see the KillerNIC.


    That looks great. I'm looking for a good network interface card currently, and I have come across this one before.

    Thanks for the link.

    Angus Higgins

  • User profile image
    Tom Servo

    JasonOlson wrote:
    Actually, I think you'll see this becoming more important in Vista and Longhorn Server (cooled network cards, that is). The complete network stack has been rewritten from the ground up, and there's some pretty cool new features that come along with that.

    One of them is network offloading. Think "graphics accelleration", only for networking. What it allows you to do is offload network processing onto the network card rather than burdening the CPU with the processing. It's actually pretty cool stuff. There's a whole slew of other improvements besides that as well. Because of this, there will be more chip activity on the network card itself (most likely there will be an actual CPU on the network card I'd imagine), which of course, in-turn, would cause more heat to be generated (hence the cooling fans).

    Just my two cents .


    The NIC on top of the picture is a 1000PRO/MT, it does:

    - TCP segmentation
    - TCP/UDP checksumming
    - Large send offload (or whatever it is called)

    The PCIe card is a 1000PRO/PT, does the same stuff, plus the IPv6 variants of that. Nothing more. (Well, maybe some minor things, but I have to get home and look up the Longhorn settings panel for the card, but AFAIK it's nothing earth shattering.)

    In regards to that KillerNIC...

    I doubt that it'll bring any performance benefits. It's Linux and its TCP/IP stack running on a 400mhz CPU. It's not a specialized ASIC, it's a software stack on an underclocked CPU. Under load, it'll very likely fold compared to the Windows stack on a gigahertz CPU. And in regards to gaming, I doubt that this thing will be able to do anything at all. Game traffic happens in the range of a few kilobytes/s on average, no need for TOE or anything. Lag improvement? The only influence you have on lag is that one that builds up between the NIC and the uplink router (under 2ms, the rest past the router stacks on top of it). And lastly, this NIC requires its own TCP/IP stack installed to have any questionable benefit, surely Windows will dig that.

    Also, loading out that old PCI NIC with nearly 100mbit/s ate maybe 10% on my machine, so I doubt any TOE, no matter how it works (including the KillerNIC), will be able to boost framerates, especially not in double digit percentages.

  • User profile image
    Rossj

    Tom Servo wrote:
    
    - Large send offload (or whatever it is called)


    Yes, that's what it is called - although I think sometimes it is called TSO (TCP Segment Offload). Both that are the checksum offload make a difference in CPU usage (although I believe the checksum is minimal impact).  LSO basically means the OS doesn't have to generate MTU size chunks of data, it can be done on the card and as you can imagine is optimal with LARGE (hence the name) transfers but outbound only.  I can't tell you how this is affected by the Window size though ..

    I guess you already know this and I am teaching you to suck eggs. Sorry.


    Tom Servo wrote:
    
    I doubt that it'll bring any performance benefits. It's Linux and its TCP/IP stack running on a 400mhz CPU. It's not a specialized ASIC, it's a software stack on an underclocked CPU. Under load, it'll very likely fold compared to the Windows stack on a gigahertz CPU.


    Last I heard the Linux guys were arguing about TOE support in the 2.6 kernel because it is patent-encumbered, if they haven't integrated it yet then it won't offer any advantage at all Smiley


    Tom Servo wrote:
    
    Also, loading out that old PCI NIC with nearly 100mbit/s ate maybe 10% on my machine, so I doubt any TOE, no matter how it works (including the KillerNIC), will be able to boost framerates, especially not in double digit percentages.


    Maybe the checksum offloading will make a difference with high frequency small packets?

  • User profile image
    Tom Servo

    Until now I had no idea about the point of LSO.

    Also, I doubt that checksum offload would do lot of a difference in the sub-256kbit/s range.

  • User profile image
    Rossj

    I found why the Linux guys (I generalise) don't really want TOE

    A snippet ..

    The OSDL wiki wrote:

    If an attacker can discover the TOE NIC model in use, they can use this information to enable resource-based algorithmic attacks. For example, a SYN flood could potentially use up all TOE resources in a matter of seconds. The TOE NIC will either stop accepting connections (complete DoS), or will constantly bounce back to the software net stack.

    TOE NICs are more resource limited than your overall computer system. This is most readily apparent under load, when trying to support thousands of simultaneous connections. TOE NICs simply do not have the memory resources to buffer thousands of connections, much less have the CPU power to handle such loads. Further, each TOE NIC has different resource limitations (often unpublished, only to be discovered at the worst moments).
    Once resources are exhausted, TOE will either fall back to 100% software net stack, defeating the purpose of TOE, or will deny service to additional clients.


    Hopefully Microsoft will have reasonably good answers to some of the issues listed there*.


    * In fact if someone wanted to clue us in, that'd be *FANTASTIC*.

Conversation locked

This conversation has been locked by the site admins. No new comments can be made.