Chapter 3
System Preparation
gather performance data using a variety of settings to determine the optimal combinations.
The primary values that govern performance include send/receive buffers, size of reads/
writes, and rfc1323 value for high performance networks (HIPPI, G-Enet). Create a table
showing these values. An example table can be found below:
To test the receiving machines performance issue:
% ttcp -r -s -b<bsize> -l<lsize>
To test the sending machines performance issue:
% ttcp -t -s -b<bsize> -l<lsize> <hostname/interfacename>
The following is an example of the information to be gathered from the above commands
(assumes en0 is 10 MB Ethernet, if en0 is Fast or G-Enet, rfc1323 should be on:
Receiver Table
Interface
en0
en0
en0
en0
...
ccs0
ccs0
...
Sender Table
Interface
en0
en0
en0
en0
...
ccs0
ccs0
...
What you are looking for are the best values possible for each network connection. These
values will be used in turn by HPSS to optimize its data transfers. By no means is this a
complete picture of what controls network performance. In fact, the assumption is made
that the customer already has the networks optimized by their local network group or has
contacted outside assistance to perform this task. The process described here is to
determine the best user-level values to optimize HPSS performance on an already tuned
network, rather then trying to fix underlying network problems.
To test the TCP socket performance over a network connection, issue the following on the
receiving node:
% ttcp -r -s -p<port>
where a typical port is 4321. Then issue the following on the transmitting node:
% ttcp -t -s -p<port> <hostname>
160
Bsize Lsize RFC1323
16k
16k
Off
16k
8k
Off
64k
64k
Off
64k
32k
Off
64k
64k
On
64k
32k
On
Bsize Lsize RFC1323
16k
16k
Off
16k
8k
Off
64k
64k
Off
64k
32k
Off
64k
64k
On
64k
32k
On
September 2002
Performance CPU utilization
--
--
--
--
--
--
--
--
--
--
--
--
Performance CPU utilization
--
--
--
--
--
--
--
--
--
--
--
--
HPSS Installation Guide
Release 4.5, Revision 2