🚀 go-pugleaf

In article <3t3s1k$fqn@nytimes.us.wfl.com>,
R.A. Henderson <woody_henderson@us.wfl.com> wrote:

>If you have the docs for LWP there is a section on tuning
>the TCP/IP parameters in the net.cfg. If you don't and you
>e-mail me, I'll cut of my net.cfg and mail it back with a 
>short explanation from the docs. Right at the bottom of
>your current net.cfg there should be a couple of lines for
>tuning things like window-size, buffers, etc. that relates
>to the tcp/ip. 

A Internet user sent me this:

>I don't know if you got an answer elsewhere, but I thought I'd mail you
>this README from the Lan Workplace 4.2 T3 maintenance patch.  I
>believe the section 4.5 may cover what you want.  The maintenance
>patch was downloaded from the Novell web site.
>
>Regards,
>
>                               ...Bob Ruigh
>

Newsgroups: comp.os.netware.connectivity,chico.ect.help,comp.protocols.tcp-ip,comp.protocols.tcp-ip.ibmpc
Subject: Re: Slow performance on Novell DOS TCP/IP but great with Linux.. Help!
Summary: 
Expires: 
References: <Pine.HPP.3.91.950629110004.11699A-100000@corpse.ecst.csuchico.edu> <3t3s1k$fqn@nytimes.us.wfl.com>
Sender: 
Followup-To: 
Distribution: 
Organization: California State University, Chico
Keywords: 
Cc: 

In article <3t3s1k$fqn@nytimes.us.wfl.com>,
R.A. Henderson <woody_henderson@us.wfl.com> wrote:

>If you have the docs for LWP there is a section on tuning
>the TCP/IP parameters in the net.cfg. If you don't and you
>e-mail me, I'll cut of my net.cfg and mail it back with a 
>short explanation from the docs. Right at the bottom of
>your current net.cfg there should be a couple of lines for
>tuning things like window-size, buffers, etc. that relates
>to the tcp/ip. 

Bob Ruigh replied:

>David,
>
>I don't know if you got an answer elsewhere, but I thought I'd mail you
>this README from the Lan Workplace 4.2 T3 maintenance patch.  I
>believe the section 4.5 may cover what you want.  The maintenance
>patch was downloaded from the Novell web site.
>
>Regards,
>
>                               ...Bob Ruigh
>

I looked at the new patch and this was the relivant code I needed to fix 
my throughput problem

----

4.5 - NET.CFG: Tuning the TCP Protocol Receive Window Size.

Establishes the TCP Protocol's initial maximum receive window size.  This
controls how many back-to-back packets a remote host will transmit. 

Since the TCP window size is a function of the ability to receive the data,
adjusting the LSL BUFFERS is necessary to accommodate larger
TCP_WINDOW
settings.  A simplistic relation (formula) which describes the
TCP_WINDOW size
setting is:

window LESS THAN OR EQUAL TO (number_buffers *
(packet_frame_size -
packet_header_size))

To simplify this calculation, use zero for packet_header_size. 

Ethernet has 16 bytes of header overhead.  FDDI has approximately 58
symbols
of header overhead (approximately 36 bytes).  ARCNet has 9 bytes of
header
overhead.  Token Ring has 21 bytes of overhead. 

If other protocol stacks are utilizing the ODI BUFFERS for incoming data,
the
dynamic TCP_WINDOW size will be adjusted accordingly.  A
TCP_WINDOW size of 0
(only found in LAN traces) indicates that the sending host should wait
until
the receiving host's protocol stack has available buffers to receive the
data.

The syntax is:

LINK SUPPORT
     BUFFERS     {buffers} {frame_size}

PROTOCOL TCPIP
     TCP_WINDOW     {window_size}

The minimum {window_size} is 512 and maximum is 32767.  The default
{window_size} is 2880.

The {frame_size} for ETHERNET_II is 1518.  The {frame_size} for FDDI is
4558. 
The {frame_size} for ARCNet is 517.  The value of {frame_size} is the
packet
headers plus the packet data for a given media type. 

Ethernet Example:

LINK SUPPORT
     BUFFERS      16   1518     ## Allow 16 inbound frames.

PROTOCOL TCPIP
     TCP_WINDOW     17856     ## 12 packets sans headers.

This example arbitrarily leaves 4 packets for use by other protocol
stacks. 

Token Ring Example:

LINK SUPPORT
     BUFFERS     8   4202     ## Allow 8 inbound frames.

PROTOCOL TCPIP
     TCP_WINDOW     20905     ## 5 packets sans headers.

This example arbitrarily leaves 3 packets for use by other protocol
stacks. 

FDDI Example:

LINK SUPPORT
     BUFFERS     10   4558     ## Allow 10 inbound frames.

PROTOCOL TCPIP
     TCP_WINDOW     22500     ## 5 packets sans headers.

This example arbitrarily leaves 5 packets for use by other protocol
stacks. 

ARCNet Example:

LINK SUPPORT
     BUFFERS     12   517     ## Allow 12 inbound frames

PROTOCOL TCPIP
     TCP_WINDOW     6096     ## 12 packets sans headers.

This example arbitrarily leaves 0 packets for use by other protocol
stacks. 







-- 
 +----------------------------------------------------------------------------+
 |    David A. Ranch  [Computer Engineer]         dranch@ecst.csuchico.edu    |
 +----                                                                    ----+
 +------- Linux 1.1.x Administrator ---- dranch@rocko.lab.csuchico.edu -------+