Very Computer

We have several (250 to 300) Sun Sparcs on our network, mostly Sparc 5, 10
and 20 systems.  We are slowly moving into Ultra based systems.  I have
noticed something VERY strange with the ultra's we have thus far deployed.

#1, collisions..  We get many thousands of network collisions when we
transfer large amounts of data from the box (http, ftp, and nfs) reported
on the le0 ethernet port.  Transfering files to the box is no problem,
might come up with one or two collisions total, no errors, and very fast.

#2, Transfer speeds are fairly normal when one or two people are hitting
the box, but get 5 or 6 large ftp transfer going at once and the ultra
drops to pushing files at 30 and 40 kbps (Very slow)

#3, http transfers just stop half way, and never finish on small or large
documents.

Our network is very segmented, with several 10bt and 100bt switches. I
have taken the ultra's off line, and set up a simple two or three computer
network using a hub and it does NOT display the same problem at all.

Has anyone else run into this problem?  Can Sun Ultra's work on large
networks??  Seems very odd that this problem is not present on any other
Sparcs on the net running the same exact version of Solaris.

The Hardware I am refering to are Ultra 1/140 and Ultra 1/170 based
systems.  All have at least 128M (most 256M) of ram, dual 2.1 or 4.3 gig
sun HD's, and all are set up server config, Solaris 2.5.1 with
2.5.1_Recommended patch clusters.

I have tried replacing hardware, fresh unpatched 2.5.1, using FSBE's,
using Fast Etherenet (hme)  In fact the HME cards display the exact same
problem, only they send out stuff MUCH slower than the built in le0. (And
yes they receive stuff much faster than the le0 {shrug}).  

This problem seems directly related to TCP/IP because  spray'ing the
machines show that they get and send out a lot of udp traffic and dont
hiccup at all.

Open to any and all suggestions

--
Sun/Networking/ And ISP Stuff                  http://www.ezy.net/gstek
GSTek-Ezy.net Corporation       Sun Computers and Internet Access Sold Here!

comments in-line ...

Quote:> We have several (250 to 300) Sun Sparcs on our network, mostly Sparc 5, 10
> and 20 systems.  We are slowly moving into Ultra based systems.  I have
> noticed something VERY strange with the ultra's we have thus far deployed.

> #1, collisions..  We get many thousands of network collisions when we
> transfer large amounts of data from the box (http, ftp, and nfs) reported
> on the le0 ethernet port.  Transfering files to the box is no problem,
> might come up with one or two collisions total, no errors, and very fast.

> #2, Transfer speeds are fairly normal when one or two people are hitting
> the box, but get 5 or 6 large ftp transfer going at once and the ultra
> drops to pushing files at 30 and 40 kbps (Very slow)

> #3, http transfers just stop half way, and never finish on small or large
> documents.

> Our network is very segmented, with several 10bt and 100bt switches. I
> have taken the ultra's off line, and set up a simple two or three computer
> network using a hub and it does NOT display the same problem at all.

Moving your network to a hub (repeater) will eliminate the possibility
of packets being dropped in the network since a repeater does not buffer
packets, and routers/bridges/switches do and will generally drop packets when
out of buffer space.

Quote:> Has anyone else run into this problem?  Can Sun Ultra's work on large
> networks??  Seems very odd that this problem is not present on any other
> Sparcs on the net running the same exact version of Solaris.

> The Hardware I am refering to are Ultra 1/140 and Ultra 1/170 based
> systems.  All have at least 128M (most 256M) of ram, dual 2.1 or 4.3 gig
> sun HD's, and all are set up server config, Solaris 2.5.1 with
> 2.5.1_Recommended patch clusters.

> I have tried replacing hardware, fresh unpatched 2.5.1, using FSBE's,
> using Fast Etherenet (hme)  In fact the HME cards display the exact same
> problem, only they send out stuff MUCH slower than the built in le0. (And
> yes they receive stuff much faster than the le0 {shrug}).  

> This problem seems directly related to TCP/IP because  spray'ing the
> machines show that they get and send out a lot of udp traffic and dont
> hiccup at all.