Brief original question: "Recently we've noticed that out web server is
running unordinarily slow."
The whole thing is at the end:
I think someone needs to send me the "jacka*s of all times" award
Apparently, all of the offices that we were noticing this "slowness"
have their PC's hooked up to 100TX Hub, then into a switched port that
was set for 100TX/FD. I tried out another Sun in my office that is
hooked up to another port on the switch and web pages popped up with no
problem at all. Duh... either the100TX hubs or the Dell PC's don't do
100TX/FD properly. This was causing multiple retransmissions and taking
forever for the web pages to load.
That being said, I learned a few things that may be of use to others.
Thanks to:
Mark Zander mzander@newbridge.com
Randy J. Parker randy@mobiledyne.com
Timothy P. Peterman timothy.p.peterman@lmco.com
Summaries of what I was told / What I learned...
ndd -set /dev/tcp tcp_slow_start_initial 2
This is a tunable ndd paramet in 2.6 and 2.5.1 with patch 103582-12 or
later
It fixes problems that occur with Solaris web servers and Microsofts'
implementation of TCP/IP
check out http://www.mobiledyne.com/pub/tcpinc.html for more info.
Sun also has a bugid about it: BugID: 4069902
Full Duplex or Not??
Make sure that all of the duplex settings are "hard" set and not auto
detecting. This is bad.
Check the settings on the Sun by doing the following:
ndd -get /dev/hme link_status (0=down, 1=Up)
ndd -get /dev/hme link_speed (0=10Mbps, 1=100Mbps)
ndd -get /dev/hme link_mode (0=Half Duplex, 1=Full Duplex)
Check the setting on you switch.
Check your PC. Check everything else. Then check it again, before
sending email about it....
-------------------------------------------
It might seem a little off topic, but I've narrowed it down to the
system (i think)
Here's what is happening:
Recently we've noticed that out web server is running unordinarily slow.
This has probably been happening for the last 2 - 3 weeks.
During this time period, the following things have happened:
a disk died and was replaced
patches were added and the machine was rebooted
no changed to the web server software were made
In addition, every now and then my telnet or ssh session becomes
unresponsive, while work on other remote, machines is unaffected. (this
just started happening as well)
The hardware / software:
Netra 1 (Ultra II 200 Mhz)
512MB Ram
Sun Multipack with 4GB drives (Targets 2,3,4,5,8)
Solaris 2.5.1 (patched)
Netscape Enterprise 3.6
(also tried Apache 1.3.4)
FastEthernet - Full Duplex (hard set in /etc/system and on Catalyst
switch)
Network Appliance NFS server for web docs (the problem occurs even
with the docs stored locally)
Enterprise Volume Manger 2.5 (Target 2 and 3 are mirrored... so are
4 and 5)
additonal studd:
i've tunded the ndd (tcp) stuff acoring to all of the technotes i've
read and they have worked fine up until now.
patches are current as of about 2 weeks ago
What I've checked (and 2 others have verified):
there is little or no paging or swaping going on.
iostat looks fine
netstat -a shows alot of TIME_WAIT and FIN_WAIT_2 's (but that looks
normal)
the load on the system is pretty constant - around .1 to .2 (it spikes
to .3 every now and then)
top shows the cpu as being ~ 90-95% idle
As i mentioned above, we use netscape Enterprise, bbut i installed
Apache and got the same results. I tried the same web server configs
(and versions) on my workstation and work fine. (oh yeah, the machince
takes about 250,000 hits to the web server a day - but this has been
constant for the last 6 months.)
Does anyone have any ideas on how to fix this? I've run out of ideas.
Thanks in advance.
-Mike Ghicas
mghicas@stern.nyu.edu
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:13:17 CDT