is a gigabit switch designed to handle all its devices talking at a giabit? - Networking

This is a discussion on is a gigabit switch designed to handle all its devices talking at a giabit? - Networking ; The new Dell switch I bought has 48 ports, each gigabit rated. And each one of Ethernet cards connected are also gigabit. Cables too. Does that mean that all 48 machines can theoretically saturate each cable at gigabit and the ...

+ Reply to Thread
Results 1 to 20 of 20

Thread: is a gigabit switch designed to handle all its devices talking at a giabit?

  1. is a gigabit switch designed to handle all its devices talking at a giabit?

    The new Dell switch I bought has 48 ports, each gigabit rated. And each one
    of Ethernet cards connected are also gigabit. Cables too.

    Does that mean that all 48 machines can theoretically saturate each cable
    at gigabit and the switch is rated to be able to keep up with it in the
    speed of its switching circuitry?

    Or is it a "best effort basis" and the max each port can get to is a
    gigabit only on a truly unloaded switch. Just curious...

    --
    Rahul

  2. Re: is a gigabit switch designed to handle all its devices talkingat a giabit?

    Rahul wrote:
    > The new Dell switch I bought has 48 ports, each gigabit rated. And each one
    > of Ethernet cards connected are also gigabit. Cables too.
    >
    > Does that mean that all 48 machines can theoretically saturate each cable
    > at gigabit and the switch is rated to be able to keep up with it in the
    > speed of its switching circuitry?
    >
    > Or is it a "best effort basis" and the max each port can get to is a
    > gigabit only on a truly unloaded switch. Just curious...
    >



    This is one thing that separates better vs. worse switches, so it all
    depends on the model. See if you can find any specs about the
    "backplane" in your switch.

  3. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Rahul writes:

    >The new Dell switch I bought has 48 ports, each gigabit rated. And each one
    >of Ethernet cards connected are also gigabit. Cables too.


    >Does that mean that all 48 machines can theoretically saturate each cable
    >at gigabit and the switch is rated to be able to keep up with it in the
    >speed of its switching circuitry?


    >Or is it a "best effort basis" and the max each port can get to is a
    >gigabit only on a truly unloaded switch. Just curious...


    Read the specs as to wha tthe backplane can handle. It is almost certainly
    not able to saturate all 48 outlets. That would be a backplane rate of 25
    or 50 Gb.




  4. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    timeOday wrote in
    news:l7adnY6YlN_V_yXVnZ2dnUVZ_hGdnZ2d@comcast.com:

    > This is one thing that separates better vs. worse switches, so it all
    > depends on the model. See if you can find any specs about the
    > "backplane" in your switch.
    >


    lol. Its a "Dell" (better or worse? ;-) Anyways, for starters "backplane"
    doesn't register any hits on a search. Any synonyms? Any other sugesstions?
    Maybe I will have to bug the Dell Engineers before they take a stand on
    their product performance.

    --
    Rahul

  5. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Unruh wrote in
    news:zc1uk.11241$%b7.10737@edtnps82:

    > Read the specs as to wha tthe backplane can handle. It is almost
    > certainly not able to saturate all 48 outlets. That would be a
    > backplane rate of 25 or 50 Gb.
    >


    Thanks Unruh. So what are ballpark "backplane" figures that people have for
    the routers they use?


    --
    Rahul

  6. Re: is a gigabit switch designed to handle all its devices talkingat a giabit?

    Only a very expensive switch will be able to handle 48 Gigabits( think a
    Cisco 3700 series is rated at under 40). A typical Dell 24 port has a
    backplane rating of 12 - 14 gigabits, I'd expect that a 48 port has more
    capacity. If you actually saturate the bandwidth to beyond the capacity
    of the backplane, you'll get buffering (store and forward),the resultant
    latency, and dropped frames if it us sustained beyond the capacity of
    the buffers. On the other hand, what are you going to be doing that will
    demand that much bandwidth? I wouldn't even know how to generate traffic
    at that volume except to hook up a random frame generator. You're
    certainly not get anywhere near that with the typical workstation - gig
    NIC or otherwise


    Rahul wrote:
    > The new Dell switch I bought has 48 ports, each gigabit rated. And each one
    > of Ethernet cards connected are also gigabit. Cables too.
    >
    > Does that mean that all 48 machines can theoretically saturate each cable
    > at gigabit and the switch is rated to be able to keep up with it in the
    > speed of its switching circuitry?
    >
    > Or is it a "best effort basis" and the max each port can get to is a
    > gigabit only on a truly unloaded switch. Just curious...
    >


  7. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Em Sábado, 30 de Agosto de 2008 04:46, Rahul escreveu:

    > sugesstions? Maybe I will have to bug the Dell Engineers before they take
    > a stand on their product performance.
    >

    Ohhh, they will say it is the best there is... north, south, east and west
    of Pecos

  8. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    kla wrote in
    news:ncGdnbnFpt4wHCHVnZ2dnUVZ_oPinZ2d@posted.olype ninternet:

    > Only a very expensive switch will be able to handle 48 Gigabits( think
    > a Cisco 3700 series is rated at under 40). A typical Dell 24 port has
    > a backplane rating of 12 - 14 gigabits, I'd expect that a 48 port has
    > more capacity. If you actually saturate the bandwidth to beyond the
    > capacity of the backplane, you'll get buffering (store and
    > forward),the resultant latency, and dropped frames if it us sustained
    > beyond the capacity of the buffers.


    Thanks kla!

    I spoke with a Dell Engineer today and this is what he gave me:

    "Total Switch Capacity PowerConnect 6248 - 232 Gbps"
    "Maximum Forwarding Rate PowerConnect 6248 - 131 Mpps" [sic]

    Now I'm confused; these are numbers much higher than what I expected
    from this thread. Is the Dell Engineer just making a mistake or did I
    just end up with a "high-end" switch! I don't think it was that
    "expensive"!

    232 Gbps would be an overkill anyways even with all 48 ports chattering
    away at 1 Gbps.


    >On the other hand, what are you going to be doing that will
    > demand that much bandwidth? I wouldn't even know how to generate
    > >traffic at that volume except to hook up a random frame generator.

    > You're certainly not get anywhere near that with the typical
    > workstation - gig NIC or otherwise
    >


    I'm not sure at al whether we will actually use all the capacity but
    this is for a computational cluster running a bunch of highly parallel
    MPI codes. In the past we have had opinions that our interconnect was
    the limiting factor. But again I'm not sure. Since we have already
    invested in blades with twin eth ports and have the requisite switches
    too it made sense to bond them together to attain the fastest backbone
    possible.

    Incidently before I open it up for a production runs I'd like to tax the
    network to know what's the max. performance we can get out of it. I've
    been recommended netperf as a tool ; but if people have any other
    sugesstions or comments I'd be very much happy to hear them.


    --
    Rahul

  9. Re: is a gigabit switch designed to handle all its devices talking ata giabit?

    On Sep 2, 8:13*pm, Rahul wrote:

    > I spoke with a Dell Engineer today and this is what he gave me:
    >
    > *"Total Switch Capacity *PowerConnect 6248 - 232 Gbps"
    > "Maximum Forwarding Rate *PowerConnect 6248 - 131 Mpps" [sic]
    >
    > Now I'm confused; these are numbers *much higher than what I expected
    > from this thread. Is the Dell Engineer just making a mistake or did I
    > just end up with a "high-end" switch! I don't think it was that
    > "expensive"!
    >
    > 232 Gbps would be an overkill anyways even with all 48 ports chattering
    > away at 1 Gbps.


    Neither of those pieces of information are what you wanted to know,
    which is what the backplane speed is.

    However, it may be that not all traffic has to traverse the backplane.
    For example, a 48-port switch might have 6 8-port modules. Each 8-port
    module can handle full line rate between ports, but may be connected
    to an 18-Gbps backplane. So each 8-port module can handle 2Gbps per
    port plus 16Gb/s to the backplane, (8*2+18)*6 for a meaningless "total
    switch capacity" of 204 Gbps.

    DS

  10. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    On Wed, 3 Sep 2008 03:13:03 +0000 (UTC), Rahul wrote:
    > kla wrote in
    > news:ncGdnbnFpt4wHCHVnZ2dnUVZ_oPinZ2d@posted.olype ninternet:


    > > Only a very expensive switch will be able to handle 48 Gigabits( think
    > > a Cisco 3700 series is rated at under 40). A typical Dell 24 port has
    > > a backplane rating of 12 - 14 gigabits, I'd expect that a 48 port has
    > > more capacity. If you actually saturate the bandwidth to beyond the
    > > capacity of the backplane, you'll get buffering (store and
    > > forward),the resultant latency, and dropped frames if it us sustained
    > > beyond the capacity of the buffers.


    > Thanks kla!


    > I spoke with a Dell Engineer today and this is what he gave me:


    > "Total Switch Capacity PowerConnect 6248 - 232 Gbps"
    > "Maximum Forwarding Rate PowerConnect 6248 - 131 Mpps" [sic]


    Switch performance is available on the Dell website. For the 6248:

    http://www.dell.com/content/products...=04&l=en&s=bsd

    then click on the "Tech Specs" tab, scroll down to "performance".
    The website disagrees with the engineer you talked to.

    However, as another poster said, do you really think you can
    saturate the switch?

    --
    Dale Dellutri (lose the Q's)

  11. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Rahul writes:

    >kla wrote in
    >news:ncGdnbnFpt4wHCHVnZ2dnUVZ_oPinZ2d@posted.olype ninternet:


    >> Only a very expensive switch will be able to handle 48 Gigabits( think
    >> a Cisco 3700 series is rated at under 40). A typical Dell 24 port has
    >> a backplane rating of 12 - 14 gigabits, I'd expect that a 48 port has
    >> more capacity. If you actually saturate the bandwidth to beyond the
    >> capacity of the backplane, you'll get buffering (store and
    >> forward),the resultant latency, and dropped frames if it us sustained
    >> beyond the capacity of the buffers.


    >Thanks kla!


    >I spoke with a Dell Engineer today and this is what he gave me:


    > "Total Switch Capacity PowerConnect 6248 - 232 Gbps"
    >"Maximum Forwarding Rate PowerConnect 6248 - 131 Mpps" [sic]


    Yes, these are the kind of numbers they throw around on their web page,
    which seem to mean nothing whatsoever, That 232 Gbps seems to be just
    multipling the number of ports full duplex by their rate. I do not think it
    means anything. The maximum forwarding rate might be the backplane rate but
    it is absurdly low. and I do not know what a pps is.

    I think you need to talk the the engineer and get figures from him that
    make sense or at the least to define what they mean by their figures.



    >Now I'm confused; these are numbers much higher than what I expected
    >from this thread. Is the Dell Engineer just making a mistake or did I
    >just end up with a "high-end" switch! I don't think it was that
    >"expensive"!


    >232 Gbps would be an overkill anyways even with all 48 ports chattering
    >away at 1 Gbps.



    >>On the other hand, what are you going to be doing that will
    >> demand that much bandwidth? I wouldn't even know how to generate
    >> >traffic at that volume except to hook up a random frame generator.

    >> You're certainly not get anywhere near that with the typical
    >> workstation - gig NIC or otherwise
    >>


    >I'm not sure at al whether we will actually use all the capacity but
    >this is for a computational cluster running a bunch of highly parallel
    >MPI codes. In the past we have had opinions that our interconnect was
    >the limiting factor. But again I'm not sure. Since we have already
    >invested in blades with twin eth ports and have the requisite switches
    >too it made sense to bond them together to attain the fastest backbone
    >possible.


    >Incidently before I open it up for a production runs I'd like to tax the
    >network to know what's the max. performance we can get out of it. I've
    >been recommended netperf as a tool ; but if people have any other
    >sugesstions or comments I'd be very much happy to hear them.



    >--
    >Rahul


  12. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Dale Dellutri writes:

    >On Wed, 3 Sep 2008 03:13:03 +0000 (UTC), Rahul wrote:
    >> kla wrote in
    >> news:ncGdnbnFpt4wHCHVnZ2dnUVZ_oPinZ2d@posted.olype ninternet:


    >> > Only a very expensive switch will be able to handle 48 Gigabits( think
    >> > a Cisco 3700 series is rated at under 40). A typical Dell 24 port has
    >> > a backplane rating of 12 - 14 gigabits, I'd expect that a 48 port has
    >> > more capacity. If you actually saturate the bandwidth to beyond the
    >> > capacity of the backplane, you'll get buffering (store and
    >> > forward),the resultant latency, and dropped frames if it us sustained
    >> > beyond the capacity of the buffers.


    >> Thanks kla!


    >> I spoke with a Dell Engineer today and this is what he gave me:


    >> "Total Switch Capacity PowerConnect 6248 - 232 Gbps"
    >> "Maximum Forwarding Rate PowerConnect 6248 - 131 Mpps" [sic]


    >Switch performance is available on the Dell website. For the 6248:


    >http://www.dell.com/content/products...=04&l=en&s=bsd


    >then click on the "Tech Specs" tab, scroll down to "performance".
    >The website disagrees with the engineer you talked to.


    Which gives the same meaningless figures. What the hell is "total switch
    capacity"? Or "Maximum forwarding rate" If that is the backplane speed,
    then these switches are junk.


    >However, as another poster said, do you really think you can
    >saturate the switch?


    On a full scale MPI computation they might well be able to, especially if
    the backplane is only 100Mbps.


    >--
    >Dale Dellutri (lose the Q's)


  13. Re: is a gigabit switch designed to handle all its devices talking ata giabit?

    On Sep 3, 8:51*am, Unruh wrote:

    > On a full scale MPI computation they might well be able to, especially if
    > the backplane is only 100Mbps.


    The backplane wire speed is certainly 10Gb/s or above. I have never
    heard of a 48-port Gigabit switch any slower than that.

    DS

  14. Re: is a gigabit switch designed to handle all its devices talking ata giabit?

    On Sep 3, 8:48*am, Unruh wrote:

    > means anything. The maximum forwarding rate might be the backplane rate but
    > it is absurdly low. and I do not know what a pps is.


    pps = packets per second

    Sometimes a switch may be able to forward jumbo packets at a high
    speed but bogs down with large numbers of small packets. It's
    essentially a measure of how quickly the switch can figure out where a
    packet needs to go.

    DS

  15. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    David Schwartz writes:

    >On Sep 3, 8:51=A0am, Unruh wrote:


    >> On a full scale MPI computation they might well be able to, especially if
    >> the backplane is only 100Mbps.


    >The backplane wire speed is certainly 10Gb/s or above. I have never
    >heard of a 48-port Gigabit switch any slower than that.


    I agree. Do you have any idea what their claim of 100Mpps is all about?


    >DS


  16. Re: is a gigabit switch designed to handle all its devices talking ata giabit?

    On Sep 3, 1:58*pm, Unruh wrote:

    > I agree. Do you have any idea what their claim of 100Mpps is all about?


    It means the switch can handle 100 million packets per second.

    DS

  17. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Unruh wrote:
    > Which gives the same meaningless figures. What the hell is "total
    > switch capacity"? Or "Maximum forwarding rate" If that is the
    > backplane speed, then these switches are junk.


    Think of it as being rather like STREAMS results for some systems. As
    STREAMS can be the sum of the individual memory subsystems, those
    numbers are probably sums of what the individual modules can do in
    isolation from the others in the switch.

    Given the domain from which I post I will mention:

    Near as I can tell, the HP ProCurve 3500-yl-48G (J8693A) is the HP
    ProCurve competitor. (take that with a grain of salt, for while I may
    be "Mr. Netperf" I am not in the ProCurve side of the house and not
    fully "up" on their product line)



    The performance specs seem to include a few more details. Perhaps not
    as many as some might want, but they seem to include latency.

    rick jones
    --
    denial, anger, bargaining, depression, acceptance, rebirth...
    where do you want to be today?
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

  18. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Rahul wrote:
    > Incidently before I open it up for a production runs I'd like to tax
    > the network to know what's the max. performance we can get out of
    > it. I've been recommended netperf as a tool ; but if people have any
    > other sugesstions or comments I'd be very much happy to hear them.


    Certainly some variation on a netperf test could be used. My initial
    inclination would be to go with many concurrent netperf TCP_RR tests
    configured for single-connection bidirectional bandwidth:

    ../configure --enable-burst ...

    The "fun" part (unless you want the even greater fun of netperf4's XML
    config files is getting all the runs going. They won't be
    synchronized - netperf2 isn't setup for synchronized multiple stream
    testing. The "confidence intevals" hack doesn't apply because this
    will be many to many systems. So, short of going down the netperf4
    path (which I wouldn't mind mind you but may be more than you want)
    I'd probably enable "demo mode" and have each netperf instance report
    the bandwidth it was achieving as it went:

    ../configure --enable-burst --enable-demo

    and then on your systems it would be a command along the lines of:

    netperf -H -t TCP_RR -l 900 -D 1 -- -s 1M -S 1M -r 64K -b 12

    and then once all of them were running (hopefully within 900 seconds)
    you would look at what each was reporting.

    [Although taking a quick look suggests that demo mode with a
    bidirectional TCP_RR test may not be working quite right, in which
    case, some combination of TCP_STREAM and/or TCP_MAERTS tests might be
    in order]

    You can "apt-get install netperf" from non-free (I never bothered to
    get the netperf2 license open source certified) and use that netserver
    on the "target" systems, but you will want to compile from source to
    get the netperf binaries that can do the burst-mode and demo stuff.

    For netperf things, the netperf-talk mailing list on netperf.org
    (subscription required to post to try to deal with spam) is useful.

    happy benchmarking,

    rick jones
    http://www.netperf.org/svn/netperf2/...c/netperf.html

    --
    No need to believe in either side, or any side. There is no cause.
    There's only yourself. The belief is in your own precision. - Jobert
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

  19. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    David Schwartz writes:

    >On Sep 3, 1:58=A0pm, Unruh wrote:


    >> I agree. Do you have any idea what their claim of 100Mpps is all about?


    >It means the switch can handle 100 million packets per second.


    Thanks.

  20. Re: is a gigabit switch designed to handle all its devices talking at a giabit?

    Rick Jones wrote in news:g9n81u$lsq$1
    @usenet01.boi.hp.com:

    > Certainly some variation on a netperf test could be used. My initial
    > inclination would be to go with many concurrent netperf TCP_RR tests
    > configured for single-connection bidirectional bandwidth:
    >


    Thanks Rick. netperf it is for me! I am trying to get that ompiled and
    running now.


    --
    Rahul

+ Reply to Thread