Re: speed diff ethernet half and full duplex

[DrFranken <midrange@xxxxxxxxxxxx>]

Jim,

    You will not get double the speed from *HALF to *FULL. The 
theoretical doubling is because traffic can move both ways on the 
connection. So *FULL on a 100Mb connection would allow in theory 100Mb 
inbound and 100Mb outbound simultaneously and thus you have '200Mb' 
total throughput.  So for your 'huge volume going one way' problem you 
need bandwidth. Thus getting the thing from 10Mb to 100Mb will almost 
certainly cut the time buy 80% to as much as 90%. The duplex change WILL 
still help because TCP is sending back the confirmation of delivery 
packets and they ride 'for free' with full duplex while they interrupt 
outbound connectivity with half duplex.

     Other things to look at are in CHGTCPA make sure that the send 
buffer on the V5R4 system and the receive buffer on the V5R2 system are 
set appropriately. I would suggest 64K or so for a local connection such 
as this. The system defaults of about 8K are often limiting. When the 
send buffer is full of UN-acknowledged packets it stops sending! End and 
restart the *FTP server for this change to be recognized.

    As you already have verified the frame size on the lines is huge 
and 1496 is correct here. This is the default but sometimes we see lower 
numbers. This is especially true if any Virtual address are in use where 
the default is 576 bytes! (Use NETSTAT *IFC and option 5 to check the IP 
interfaces)  If both systems had GbE capability then 8996 byte packets 
would also help.

    Realizing that you don't have switch visibility now this one is 
hard but if the switch and the systems don't both report the SAME VALUEs 
for speed and duplex then you have a problem. Often with older Cisco 
gear and some Dell gear the system will report *FULL and the switch 
*HALF. This will cut your speed by 80 to 95%, and show huge error rates 
on the switch end. Assure that both the system and the switch are 
configured for *FULL.  You can tell if this is a problem even without 
looking in the switch by logging onto the V5R2 system, do DSPPFM on some 
huge file then just stand on the page down key. If you see continuous 
fast scrolling with no pauses then you're likely OK. If you see a big 
bunch of screens then a huge pause (3 to 5 or more seconds) then the 
pattern repeats - then you likely have this problem. (Works locally not 
so much remotely.)

   Finally don't believe the byte count on the TCP screens!!! Look at 
the size of the object on the receiving system. The TCP numbers reset 
somewhere around 4G I believe......

     - Larry "DrFranken" Bolhuis.

On 11/13/2011 2:15 PM, Jim Franz wrote:
> Attempting to ftp a 32 gig savf from a v5r4 system to v5r2 sys (I know...going the wrong way). Systems have no tape compatibility, and opposite sides of the country.
>
> The v5r4 has 5706 ethernet 10/100/1000 linespeed=*auto, duplex=*half
> No weekend view of the switch&  router.
> The v5r2 has 2849 ethernet 10/100 linespeed=10M, duplex=*half
>
> I can see that if switches support it, get the v5r2 machine set to 100M or *auto linespeed, and duplex on both.
> Figuring 10 x faster xmit if linespeed increase, but any way to quantify the duplex part? What I can google (non-iSeries) says it's almost twice, but not quite..
>
> Also - framesize 1496 on each.
>
> After 12 hrs only 4 gig received (per netstat byte counts)
>
> Jim