RE: FTP performance

[Stefan.Tageson@xxxxxxxxxxxxx]

Bob, from V4R4 Performance Capabiblities Reference
 (and sorry for the length of the reply, I'm on vacation right now so I
don't have the time to edit)

TCP/IP and Sockets Performance Information:
. In V4R3, TCP/IP and APPC generally provided similar levels of application
performance.
. In V4R4, the performance of TCP/IP is superior to APPC. Significant
network infrastructure
software performance enhancements were implemented including: Optimization
of Sockets APIs
and re-implementing Sockets in SLIC, optimization of TCP/IP to improve
performance and scalability,
and a more efficient scheme for the Ethernet device driver to interface
with the IOP/IOA. These
changes improved performance by significantly reducing the CPU time
required to process the
communcations software.
. Measurements with the NetPerf workload demonstrates the significant
performance improvement with
V4R4 enhancements: The CPU time for the Request/Response scenario (client
server like) was
reduced by 40-50%. The CPU time for the Connect/Request/Response scenario
(web server like) was
reduced by 40-50%. The maximum transfer rate for the Streaming (large
transfer) scenario using 100
Mbps Ethernet increased from 40 Mbps up to 90 Mbps. The NetPerf workload is
defined later in this
section.

. Always ensure that the entire communications network is configured
optimally. The maximum frame
size parameter (MAXFRAME on LIND) should be maximized. The maximum
transmission unit
(MTU) size parameter (CFGTCP command) for both the route and interface
affect the actual size of
the line flows and should be configured to *LIND. This means that there
will be a one-to-one match
between frames and MTUs.
. When transferring large amounts maximize the size of the application's
send and receive size. This is
the amount of data that the application transmits with a single sockets
API. Being sockets does not
block up multiple applications sends, it is important to limit the number
of interactions. Block in the
application if possible.
. With V4R4, TCP/IP can take advantage of larger buffers. Prior to V4R4,
the TCP/IP buffer size
(TCPRCVBUF and TCPSNDBUF on the CHGTCPA or CFGTCP command) was recommended
to be
increased from 8K bytes to 64K bytes to maximize data rates. When
transferring large amounts of
data with V4R4, you may receive higher throughput by increasing these
buffer sizes up to 8MB.
. To receive the full benefit of the performance improvements in V4R4, it
is essential that Ethernet is
used and that the TCPONLY parameter in the LIND have a value of *YES. This
allows the IOP to
only have the TCP/IP version of its microcode active. This allows the IOP
and device driver running
on the AS/400 CPU to run in optimized mode. If other high-level protocols
are active on that line, like
APPC, then this parameter must be set to *NO for functional reasons.
. If Ethernet is used with TCPONLY(*NO) or if TRLAN is used, then you will
realize only a portion of
the V4R4 performance improvements in terms of CPU time reduction.
. The TCP/IP communications support also had significant potential capacity
increase because of better
scalability in V4R4 due to software contention reduction. NetPerf
measurements executed in the
Rochester performance lab on N-way processors indicate significant overall
capacity increases over
V4R3.

. For optimum performance with FTP, make sure the MTU size is as large as
possible. ASCII transfers
are slower because of character conversion from EBCDIC to ASCII prior to
the data being sent. For
PUTs and GETs with FTP from IFS, the best possible transfer rate usually
ranges from 20-25 Mbps.

Regards
// Stefan


                                                                                
                                         
                    "Marion, Bob"                                               
                                         
                    <Bob.Marion@ps.net>        To:     
"'MIDRANGE-L@midrange.com'" <MIDRANGE-L@midrange.com>             
                    Sent by:                   cc:                              
                                         
                    owner-midrange-l@mi        Subject:     RE: FTP performance 
                                         
                    drange.com                                                  
                                         
                                                                                
                                         
                                                                                
                                         
                    2000-07-21 14:11                                            
                                         
                    Please respond to                                           
                                         
                    MIDRANGE-L                                                  
                                         
                                                                                
                                         
                                                                                
                                         





Stefan,

Do you have an estimate on how much improvement will be realized by
changing
the TCPONLY parm to *yes.  On my V4R4 systems I have this parm, however on
the V4R3 systems I do not have it, so I missed the announcement on the
anticipated improvement on the parm change when I upgraded to V4R4.  Line
speed is set to 100M at *FULL DUPLEX.


Bob Marion
Phone:  972/461-3186
Fax:  972/461-3010
EMail:  bob.marion@ps.net





+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---