🚀 go-pugleaf

Single server + many users + Switch = worse performance!!!!
Multipe servers + many users + Switch = better performance!!!!

--
Dave Mills
Associated Bodywork & Massage Professionals
Systems Administrator
dave@spam.abmp.com
www.abmp.com



tbenson@dzs.com wrote in message <6m9gc1$r7j$1@nnrp1.dejanews.com>...
>  I have a small Netware 411 10BASE-T network that is growing (1 server,
CAT
>5, 4 hubs, 24 users).  I'll soon have another 10 nodes.  Are there
statistics
>in the LAN/WAN section of MONITOR that I can use to see how much bandwidth
is
>being used?  How about a freeware/demo/shareware utility?  I'll pay for
>"regular" software if I have to, but can't spend much. If I had some
network
>monitoring software it would help me decide what new hardware I'll need.
>
>  I'd like to plan ahead and get a 10/100 hub to provide a fatter
connection
>to the server and a few power users.  If it is justified, I'd also like to
>buy a switch to be the backbone of the network.
>
>Thanks for any suggestions,
> Tom Benson
>
>-----== Posted via Deja News, The Leader in Internet Discussion ==-----
>http://www.dejanews.com/   Now offering spam-free web-based newsreading

tbenson@dzs.com wrote:

>  I have a small Netware 411 10BASE-T network that is growing (1 server, CAT
>5, 4 hubs, 24 users).  I'll soon have another 10 nodes.  Are there statistics
>in the LAN/WAN section of MONITOR that I can use to see how much bandwidth is
>being used?  How about a freeware/demo/shareware utility?  I'll pay for
>"regular" software if I have to, but can't spend much.	If I had some network
>monitoring software it would help me decide what new hardware I'll need.

I went from two 10Mbps segments (4 hubs daisy-chained on each NIC to a
tune of 50 users) to a switch that has one 100Mbps segment to the file
server & 10Mbps to the workstations. I just dropped the eight 10 Mbps
hubs I had each from the switch port; the switch I got came with 24
10/100Mbps auto-sensing ports so I used eight of those ports for the
hubs & one for the server & I still have 15 ports open on the switch
where I can either hang hubs or connect a power hungry user direct to
the port @ 10Mbps or 100Mbps.

I think what I heard from the gurus was in a single server environment
with the same pipe width all around one won't see a difference (or
even a worse performance like Dave has suggested) but if the pipe to
the server is bigger than pipe to the workstations & the box can
handle the requests (I upgraded my CPU from 486/66 to PPro200; also
from NW312 to NW411) then it should be at least the same as or better
performance.

For me the performance was not a concern even before the switch but
the introduction of a switch had the added benefits like no need for
subnetting IP, easier to expand (instead of adding segments each time
I hit the 4 hub daisy-chain limit I can just hang another hub from a
switch port), etc.

Also, go look @ www.wg.com; there's a demo of their s/w called
"LinkView" ... I tried that & it was neat. Although it is crippled in
the sense that it's limited to collecting packets for only 5 minutes &
is limited to only 5 workstations I had fun playing w/it. 

HTH. Robert



>
>Thanks for any suggestions,
> Tom Benson
>
>-----== Posted via Deja News, The Leader in Internet Discussion ==-----
>http://www.dejanews.com/   Now offering spam-free web-based newsreading

On Thu, 18 Jun 1998 08:55:31 -0600, "Dave Mills" <dave@spam.abmp.com>
wrote:

>Single server + many users + Switch = worse performance!!!!
>Multipe servers + many users + Switch = better performance!!!!

I don't understand the logic behind the above.  Connected to a 100
Mbps (probably full duplex) port with clients accessing at 10 Mbps
(half duplex) could accept what, to the clients, would appear to be 10
simultaneous inbound packets while sending 10 simultaneous outbound
packets.  And if you switch segments instead of individual stations,
the load can be shared for multiple users per switch port.

The critical factor is that the uplink port is capable of
significantly outpacing the client ports.

How do you do your math to prove otherwise?

Ken Agress/CNE

In article <6mb9rh$7o8$1@chile.it.earthlink.net>,
  "Dave Mills" <dave@spam.abmp.com> wrote:
>
> Single server + many users + Switch = worse performance!!!!
> Multipe servers + many users + Switch = better performance!!!!
>


Is that due to the delay as packets are checked and routed to the appropriate
port?  Is that server dependant, or traffic dependant?  I'm a newbie, but I
assumed that as network traffic increases, the added delay would be
insignificant compared to the delay caused by collisions/retransmissions on a
shared network.

(The first half of my original post, which started this thread, was to
inquire about software that monitors network traffic.  If I knew the
utilization level, it might help me decide whether to buy a switch or just
another hub)

Thanks for your input,
  Tom


-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/   Now offering spam-free web-based newsreading

On Fri, 19 Jun 1998 13:05:16 GMT, tbenson@dzs.com wrote:

<snip>

>Is that due to the delay as packets are checked and routed to the appropriate
>port?  Is that server dependant, or traffic dependant?  I'm a newbie, but I
>assumed that as network traffic increases, the added delay would be
>insignificant compared to the delay caused by collisions/retransmissions on a
>shared network.

All the issues you raise are entirely dependent on the switch selected
and the network design.  A modern cut-through switch should be able to
switch 20 10Mbps ports to two 100Mbps ports at full wire rate with no
delay other than buffering delay as the packets are queued/dequeued
from the high speed port.  If the load to that port is high, there
will be minimal lost time here.
>
>(The first half of my original post, which started this thread, was to
>inquire about software that monitors network traffic.  If I knew the
>utilization level, it might help me decide whether to buy a switch or just
>another hub)

Novell produces the LANalyzer package, which is quite good, and
NetWork Associates (I believe) produces Net Xray, which is also
supposed to be quite good.  I'd lean towards renting a sniffer from
Network Associates, as I've never used anything better for traffic
analysis (we own two of their box/console combinations now, and I'm
angling for a third).

Ken Agress/CNE

This is very logical.

Having a multiple conversations to a single port is not switching ethernet
beneficial
whereas balancing the load on more then one server aka creating a multiple
to multiple
ethernet conversations allows for a much better switch fabric utilization.

Shaul Efraim
Tel-Aviv, Israel.


tbenson@dzs.com wrote in message <6mdnmb$tm8$1@nnrp1.dejanews.com>...
>In article <6mb9rh$7o8$1@chile.it.earthlink.net>,
>  "Dave Mills" <dave@spam.abmp.com> wrote:
>>
>> Single server + many users + Switch = worse performance!!!!
>> Multipe servers + many users + Switch = better performance!!!!
>>
>
>
>Is that due to the delay as packets are checked and routed to the
appropriate
>port?  Is that server dependant, or traffic dependant?  I'm a newbie, but I
>assumed that as network traffic increases, the added delay would be
>insignificant compared to the delay caused by collisions/retransmissions on
a
>shared network.
>
>(The first half of my original post, which started this thread, was to
>inquire about software that monitors network traffic.  If I knew the
>utilization level, it might help me decide whether to buy a switch or just
>another hub)
>
>Thanks for your input,
>  Tom
>
>
>-----== Posted via Deja News, The Leader in Internet Discussion ==-----
>http://www.dejanews.com/   Now offering spam-free web-based newsreading
>