SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


    [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

    Re: your query



    TCP works with some other structures. It is a sliding window protocol. And
    yes for the links you mentioned the window has to be sizable. And the
    number-of-SCSI commans/R2Ts/Data in flow has to be also large to fill the
    pipe.
    
    Julo
    
    "Satish Menon" <satish_s_menon@hotmail.com> on 08-08-2001 00:55:41
    
    Please respond to "Satish Menon" <satish_s_menon@hotmail.com>
    
    To:   Julian Satran/Haifa/IBM@IBMIL
    cc:
    Subject:  your query
    
    
    
    
    Here is how the  calculation was done. Also, some folks told me that N-port
    HBAs don't reserve  end-to-end credits by exchange, they just have a common
    pool of
    credits  for the entire HBA (around 16), so the calculations factor in
    this
    new
    data point.
    
    Assume a 1 Gbps link.
    Assume a 2K FC  frame.
    
    So it takes 16us to transmit a frame. For 200ms RTT, that means  that
    200ms/
    16us ~ 10,000 credits.
    
    Case A. For a 1Gbps FCIP link with  100 N-port HBAs, each HBA needs to
    advertise 100 credits (10,000/100) to  fully utilize the link. The
    utilization with 16 credits/ HBA will be  16%.
    
    Case B. For a 1 Gbps FCIP link with 5 N-port HBAs, each HBA needs  to
    advertise 2,000 credits (10,000/5) to fully utilize the link.  The
    utilization with 16 credits/ HBA will be 0.8%.
    
    Case C. For a 10  Gbps FCIP link with 100 N-port HBAs, each HBA needs to
    advertise 1,000  credits to fully utilize the link. The utilization with 16
    credits/ HBA will  be 1.6%.
    
    Case D. For a 10 Gbps FCIP link with 5 N-port HBAs, each HBA  needs to
    advertise 20,000 credits to fully utilize the link. The utilization  with
    16
    credits/ HBA will be 0.08%.
    ------------
    
    > -----  Original Message -----
    >
    > From: "Julian Satran" <Julian_Satran@il.ibm.com>
    > To: <ips@ece.cmu.edu>
    > Sent: Monday, August 06, 2001 11:01  PM
    > Subject: Re:
    >
    >
    > >
    > > How did you  come to this number?  What did you assume for the exchange
    > >  length?
    > >
    > > Julo
    > >
    > > "Satish Menon"  <satish_s_menon@hotmail.com>@ece.cmu.edu on 06-08-2001
    > > 23:44:05
    >  >
    > > Please respond to "Satish Menon" <satish_s_menon@hotmail.com>
    > >
    > > Sent by:  owner-ips@ece.cmu.edu
    > >
    > >
    > > To:   <ips@ece.cmu.edu>
    > > cc:
    > > Subject:
    > >
    >  >
    > >
    > >
    > >
    > > Does anyone know how  many end-to-end credits a typical HBA (N-port)
    > issues
    > > per  exchange? For a long distance link over FCIP, clearly this number
    > >  needs to be large. If not, then the link utilization will be extremely
    >  > poor. For e.g. a 200ms RTT link at 1 Gbps assuming 1 exchange  through
    it,
    > > it will need 10,000 credits to keep the link fully  utilized. Using a
    > > different  number of exchanges say 2000, the  number of credits per
    > exchange
    > > drop to  5.
    >  >
    > >
    > >
    >
    
    
    
    
    


Home

Last updated: Tue Sep 04 01:04:04 2001
6315 messages in chronological order