Networking Q

I've got a question at work that I've been asked and I'm looking for ideas.

We have 8 computers located in a somewhat remote area - about 600' (200m)
from the servers. These computers are engineering workstations; they hit
the server hard for short amounts of time. It's not unusual for them to
read/write up to 500MB of data at a time.

There are also about 6 admin-type workstations located adjacent to the
servers.

We have 2 fiber optic lines available to connect the servers to the remote
site.

The 'traditionalists' want to put a switch in at the servers, connect both
servers to them, then run a single fiber optic connection to another
switch, and connect the engineering workstations to the second switch.

Server1 has a gigabit network card, server2 has a 100Mbit card.

server1 | | = admin workstations
|switch| = | |=
server2 | | ~~~~ fiber optic ~~~~~~~|switch|=engineering workstations
| |=

I favor putting a second gigabit card into server1, a second Gb/100Mb card
into server2, and running this:


server1 ~~~~~~~~ fiber optic ~~~~~~~~~~~~~| |=
\ |switch|=eng workstations
server2\\ ~~~~~~ fiber optic ~~~~~~~~~~~~~| |=
\\| |=
\|switch|=admin workstations
| |=

IMO, that separates the traffic, and provides better bandwith all around.

Server1 (with the Gb card) is the main file server, and will take all of
the heavy file server loads. Server2 (the 100Mb one) is an application
server, and will see relatively little network traffic (it's mostly an SQL
server).

Thoughts, comments, suggestions? Especially as pertains to routing 1Gb
traffic through 2 switches? Any performance issues there?

--

use munged address above to email me
SpamTrap (E-Mail Removed)

Captain Dondo <(E-Mail Removed)> writes:
> I've got a question at work that I've been asked and I'm looking for ideas.
>
> We have 8 computers located in a somewhat remote area - about 600' (200m)
> from the servers. These computers are engineering workstations; they hit
> the server hard for short amounts of time. It's not unusual for them to
> read/write up to 500MB of data at a time.
>
> There are also about 6 admin-type workstations located adjacent to the
> servers.
>
> We have 2 fiber optic lines available to connect the servers to the remote
> site.
>
> The 'traditionalists' want to put a switch in at the servers, connect both
> servers to them, then run a single fiber optic connection to another
> switch, and connect the engineering workstations to the second switch.
>
> Server1 has a gigabit network card, server2 has a 100Mbit card.
>
> server1 | | = admin workstations
> |switch| = | |=
> server2 | | ~~~~ fiber optic ~~~~~~~|switch|=engineering workstations
> | |=
>
> I favor putting a second gigabit card into server1, a second Gb/100Mb card
> into server2, and running this:
>
>
> server1 ~~~~~~~~ fiber optic ~~~~~~~~~~~~~| |=
> \ |switch|=eng workstations
> server2\\ ~~~~~~ fiber optic ~~~~~~~~~~~~~| |=
> \\| |=
> \|switch|=admin workstations
> | |=
>
> IMO, that separates the traffic, and provides better bandwith all around.
>
> Server1 (with the Gb card) is the main file server, and will take all of
> the heavy file server loads. Server2 (the 100Mb one) is an application
> server, and will see relatively little network traffic (it's mostly an SQL
> server).
>
> Thoughts, comments, suggestions? Especially as pertains to routing 1Gb
> traffic through 2 switches? Any performance issues there?

If you've got good 1Gb/s switches, and good 1Gb/s cards in the
servers, then the second solution will be better, for the admin
workstation should not notice the engineering trafic.

Note that with a normal PCI bus 133 MHz, 32-bit, you get only 133*4 =
532 MB/s of internal bandwidth, (~= 4 Gb/s) and most 1Gb/s cards won't
give you 1Gb/s streams, but only 500 Mb/s or 600 Mb/s, so it may be
worthwhile to add network cards in the servers (you could add up to 4
of 6 cards before hiting the PCI bottleneck). Then the bottleneck
will be the disk system in the server, with normal disks running at
about 40 MB/s (even with faster buses!), you'd need a good RAID setup
to be able to read and write 1Gb/s, or big buffers in the servers, but
they would help only in writting, not in reading.



--
__Pascal Bourguignon__ http://www.informatimago.com/

The world will now reboot. don't bother saving your artefacts.

Captain Dondo wrote:
> I've got a question at work that I've been asked and I'm looking for
ideas.
>
> We have 8 computers located in a somewhat remote area - about 600'
(200m)
> from the servers. These computers are engineering workstations; they
hit
> the server hard for short amounts of time. It's not unusual for them
to
> read/write up to 500MB of data at a time.
>
> There are also about 6 admin-type workstations located adjacent to
the
> servers.
>
> We have 2 fiber optic lines available to connect the servers to the
remote
> site.
>
> The 'traditionalists' want to put a switch in at the servers, connect
both
> servers to them, then run a single fiber optic connection to another
> switch, and connect the engineering workstations to the second
switch.
>
> Server1 has a gigabit network card, server2 has a 100Mbit card.
>
> server1 | | = admin workstations
> |switch| = | |=
> server2 | | ~~~~ fiber optic ~~~~~~~|switch|=engineering
workstations
> | |=
>
> I favor putting a second gigabit card into server1, a second Gb/100Mb
card
> into server2, and running this:
>
>
> server1 ~~~~~~~~ fiber optic ~~~~~~~~~~~~~| |=
> \ |switch|=eng workstations
> server2\\ ~~~~~~ fiber optic ~~~~~~~~~~~~~| |=
> \\| |=
> \|switch|=admin workstations
> | |=
>
> IMO, that separates the traffic, and provides better bandwith all
around.
>
> Server1 (with the Gb card) is the main file server, and will take all
of
> the heavy file server loads. Server2 (the 100Mb one) is an
application
> server, and will see relatively little network traffic (it's mostly
an SQL
> server).
>
> Thoughts, comments, suggestions? Especially as pertains to routing
1Gb
> traffic through 2 switches? Any performance issues there?

Much may depend on just what the admin hosts are meant to administer.

Off hand impressions...

Eg., if admin hosts are _just_ for the servers, then #2 is a
no-brainer.

If, however, they are to used to keep tabs on the engineering hosts as
well, then #2 requires the servers to route traffic from the admin to
the engineering hosts. Not so optimal in this situation. Your call as
to what is required of admin hosts.

On general principal, I like #2 as you could also use them as syslog
servers and "on-line" monitors for the servers. But if you need the
same for the engineering hosts, the need to route _through_ the servers
would be something to be avoided, IMHO. Possible to provide admin
functions to engineering hosts with a machine on their switch?

BTW, I can't see the advantage to a second Gb nic in server 1 just for
the admins unless you need them for doing backups/transfers, etc.

good luck,
prg
email above disabled