SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


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

    RE: iSCSI/VI/TCP



    Costa,
    
    Thanks for your comments on the value of using VI as part of the
    iSCSI solution.  I was pretty much in agreement with you, until 
    I got to your last paragraph. 
    
    > 
    > At the end of the day, this argument comes down to whether folks think
    > that an RDMA mechanism is the right way to go and what RDMA scheme
    > over IP is best. 
    
    I agree. If you are transporting data between application memory and  
    a remote periperal device, a scheme that permits direct DMA transfer
    makes sense.  And, if we agree to that, it also makes sense to 
    investigate a range of DMA transfer solutions, to see which provides 
    the most appropriate capabilities, allows broad implementation, etc.
    
    > A suitably generic RDMA scheme allows innovation in
    > protocol design, without worrying that the new protocol will be
    > hopelessly slow in legacy hardware.  It decreases the pressure to load
    > features into old, hardware-supported protocols instead of introducing
    > new ones. 
    
    Here's where I must ask for a clarification. Are you saying that 
    performance, or for that matter the ability to implement these 
    solutions on existing hardware, is a "don't care"?  Or, are you
    saying that a RDMA scheme is inherently efficient enough that
    even legacy hardware will work just fine? 
    
    I hope you meant the latter, but admit that my first 
    impression was that you'd give up "legacy hardware" support 
    for a fast protocol that could only run in a new hardware 
    environment.
    
    > It destroy an artificial barrier (host overhead) that
    > prevents IP from scaling to fit all our needs.
    >
    
    Allowing IP to be used across a wider range of solutions is 
    a good thing. Removing the perception that IP stacks are 
    "cycle hogs" is a good thing, too.
    
    But, I suggest that its history of maintaining compatibility 
    with older hardware and software implementations is also a 
    pretty good selling point for TCP/IP, and one that shouldn't 
    be lightly dismissed.  
    
    	- milan
    
    
    Milan J. Merhar,  Chief Engineer, Pirus Networks,  
    43 Nagog Park, Acton MA 01720-3425
    voice:  978-206-9124  email: milan@pirus.com
     
    


Home

Last updated: Tue Sep 04 01:08:05 2001
6315 messages in chronological order