RE: TCP speed

["Y P Cheng" <ycheng@advansys.com>]

Spencer Wrote:
>And without additionly significant work, NFS over UDP does not have
>any form of congestion control and running NFS/UDP cross country should
>never be done.  It bad news. :-)
>The NFSv4 specification requires the use of a transport that provides
>congestion control.  NFSv4 does not mandate the use of TCP but does
>mandate the use of congestion control for comforming implementations.

Yes, flow control is important.  When hundreds of NFS clients on Internet
write to a NFS server, it could be very bad.

On SCSI the savor is RTT.  While on the initiator side, the resource is
pre-allocated before a SCSI request is generated, the target must manage its
resource using RTT.  The difference between TCP and UDP is the
connection-oriented acknowledges.  On long latency between two endpoints,
waiting acknowledges can be very detrimental to performance.

The ultimate solution to flow control in iSCSI with gigabit media is to move
the datagrams to application software quickly.  We have absolutely no
control when the thousands of nodes should send datagrams to a server at
gigabit speed.  We can slow the flow down by using the concept of EE-credit.
But, this would be a wrong solution when latency time is in milliseconds or
even seconds.

Y.P. Cheng, CTO, ConnectCom Solutions Corp.


-----Original Message-----
From: Spencer.Shepler@eng.sun.com [mailto:Spencer.Shepler@eng.sun.com]On
Behalf Of Spencer Shepler
Sent: Monday, September 11, 2000 9:04 AM
To: Y P Cheng
Cc: ips@ece.cmu.edu
Subject: Re: TCP speed


Y P Cheng wrote:
>
> Julo wrote:
> >Our experience is the same. TCP is FAST.
> >The only remaining trouble is memory copy from TCP buffers
> >to application buffers. Unless
> >handled properly this may slow you down considerably.
>
> The issue is not the TCP memory to memory copy speed, it is the latency
time
> of receiving TCP acknowledges.  Between two endpoints of New York and Los
> Angeles, latency is in milliseconds if not in seconds.  On a one-gigabit
> network, for each millisecond there are 100K of data, or 66 1.5K datagrams
> being transferred.  In fibre channel, there is this EE-credit, End-to-End.
> If the sending party has 10 EE credits, it can't send more than 10
> datagrams.  EE-credit manages the TCP sliding window currently discussed
in
> iSCSI.  After sending 10 datagrams, one must wait for acknowledges that
may
> take several hundred milliseconds to come.
>
> I do believe TCP is a wrong protocol for iSCSI.  A SCSI request from an
> initiator is inherently acknowledged by its response from a target.
> Therefore, UDP for iSCSI is a better choice.  NFS is implemented on UDP.

And without additionly significant work, NFS over UDP does not have
any form of congestion control and running NFS/UDP cross country should
never be done.  It bad news. :-)

The NFSv4 specification requires the use of a transport that provides
congestion control.  NFSv4 does not mandate the use of TCP but does
mandate the use of congestion control for comforming implementations.

Spencer