networking question

Discussion:

networking question

William Fragakis

2014-02-17 15:30:09 UTC

Our k12linux installation is out-growing our present, very primitive,
networking structure.

Currently, our ltsp server - which also behaves as the firewall for a
number of kvm servers and their vms - connects to 16 or so thin clients.
Between the clients, internal servers and printers, the switch to which
the ltsp server is connected to is at capacity (It's a basic 24 port
gigabit).

We now need to add the n+1 client that exceeds switch capacity. Do we
buy
a) a larger switch to replace the current one
b) a second switch daisy chained to the current one
c) a second switch connected to a second NIC using the same internal
ip/dhcp range (and if so, the recommended manner)
d) a more obvious, elegant, simple, cheaper method that I haven't
considered because I don't know squat (American slang for "not much")
about this stuff.

thanks to all,
William Fragakis

R. Scott Belford

2014-02-17 17:55:50 UTC

Permalink

Post by William Fragakis
Our k12linux installation is out-growing our present, very primitive,
networking structure.
Currently, our ltsp server - which also behaves as the firewall for a
number of kvm servers and their vms - connects to 16 or so thin clients.
Between the clients, internal servers and printers, the switch to which
the ltsp server is connected to is at capacity (It's a basic 24 port
gigabit).
We now need to add the n+1 client that exceeds switch capacity. Do we
buy
a) a larger switch to replace the current one
b) a second switch daisy chained to the current one
c) a second switch connected to a second NIC using the same internal
ip/dhcp range (and if so, the recommended manner)
d) a more obvious, elegant, simple, cheaper method that I haven't
considered because I don't know squat (American slang for "not much")
about this stuff.

If the printers or VM's do not need dedicated gigabit connectivity to the
network, and it is geographically possible, then you could subgroup these
devices with a smaller, less expensive switch thus freeing ports on your
24-port switch. If all devices need dedicated gigabit connectivity, then
you can add a second NIC to the server and either a. expand nfs/dhcp/tftp
ranges (complicated) or b. bond the NICs (less complicated) and add a
second gigabit switch for future expansion.

For network hygiene, I'd ponder re-purposing a box, installing pfsense, and
letting it become your firewall/gateway/dansguardian/squidguard box.

--scott

Jim Kinney

2014-02-17 18:04:01 UTC

Permalink

my ordering for solution:

1) bigger switch (big assumption is existing network is not overloaded now)
2) add a secondary "smart" switch above the existing switch if the existing
switch supports "upstream" port
3) add a second switch downstream with a bonded or aggregate connection for
more bandwidth (if "smart" primary switch)
4) add a secondary switch with Gbit upstream and 100M ports and set primary
switch to 100M except for new switch port
5) add a secondary switch upstream with bonded/aggregate connections to
multiple nics on server
6) add a second switch off second nic on separate subnet and split the
load. This requires a split subnet with dual gateways (each nic) (no smart
switch required)

--
--
James P. Kinney III

Every time you stop a school, you will have to build a jail. What you gain
at one end you lose at the other. It's like feeding a dog on his own tail.
It won't fatten the dog.
- Speech 11/23/1900 Mark Twain

*http://heretothereideas.blogspot.com/
<http://heretothereideas.blogspot.com/>*

William Fragakis

2014-02-19 15:03:25 UTC

Permalink

Jim,
(btw, hope all is well) big thanks to both you and Scott for your
exhaustive answers.

We seem to be doing okay on the bandwidth internally, nothing is hanging
or lagging even with a few clients playing Pandora and a bunch of
internal rdesktop sessions. I think I'll downstream lower bandwidth
devices like printers, secondary vms, etc to a second inexpensive
switch.

Off topic - heard on NPR how Los Angeles schools thought it would be a
good idea to get each student an iPad at about $700/device so they could
"all have computers". That was about the price of my latest kvm server
with an AMD 8 core CPU, 24GB RAM and 240GB SSDs in a RAID 1. Not to
mention the iPads are getting broken/stolen which makes the whole
process even more expensive. Not that we don't have iPads in our own
househould. Just that LTSP continues to make a ton of sense.

again, my thanks and best regards to all,
William Fragakis

Message: 2
Date: Mon, 17 Feb 2014 13:04:01 -0500
To: "Support list for open source software in schools."
Subject: Re: [K12OSN] networking question
Content-Type: text/plain; charset="utf-8"
1) bigger switch (big assumption is existing network is not overloaded now)
2) add a secondary "smart" switch above the existing switch if the existing
switch supports "upstream" port
3) add a second switch downstream with a bonded or aggregate connection for
more bandwidth (if "smart" primary switch)
4) add a secondary switch with Gbit upstream and 100M ports and set primary
switch to 100M except for new switch port
5) add a secondary switch upstream with bonded/aggregate connections to
multiple nics on server
6) add a second switch off second nic on separate subnet and split the
load. This requires a split subnet with dual gateways (each nic) (no smart
switch required)

Jim Kinney

2014-02-19 15:31:15 UTC

Permalink

I heard that story too and thought of Cobb county schools' aborted "Give
'em all a MAC" debacle.

Given the poor fiscal support of schools in general, LTSP (and a next gen
thing I'm still poking at) are the most cost effective ways to put students
in front of computers for any reason.

Next Gen Thing: server(s) runs Ovirt and many generic VMs. Student uses
wireless android pad/laptop and connects using spice to get full desktop
with sound and video.

Memory usage is similar using memory ballooning (shared, read-only memory)
as all VMs are identical. Drive space is similar (only /home is different
for each) and it works over wireless and from home. Big advantage is 3-fold
A) GUI for admin of VM environment (and user authentication using freeIPA
has gui as well) B) better stability and security - student can only crash
their VM, not the classroom or half the school. C) can use BYOD (mostly) as
there is a spice client for windows, Linux, Mac and Android.

I've found cheap android laptops for $100. tiny screen and keyboard but
good size for little hands. Can still use this stuff for stand-alone
machine lab setup by mounting /home from nfsv4 server. Good for video work.
Most stuff needs minimal horsepower. Bigger kids can get bigger android
laptops for $170. If schools could figure out how to use freetextbooks, the
backpack load for the kids drops to 3-5 lbs. down from 20-40 lbs.

Flash is still a hog. Wish it would go the way of the dodo bird.

Wouldn't is be cool for kids to be able to say they use a supercomputer
every day at school? !!!

Post by William Fragakis
Jim,
(btw, hope all is well) big thanks to both you and Scott for your
exhaustive answers.
We seem to be doing okay on the bandwidth internally, nothing is hanging
or lagging even with a few clients playing Pandora and a bunch of
internal rdesktop sessions. I think I'll downstream lower bandwidth
devices like printers, secondary vms, etc to a second inexpensive
switch.
Off topic - heard on NPR how Los Angeles schools thought it would be a
good idea to get each student an iPad at about $700/device so they could
"all have computers". That was about the price of my latest kvm server
with an AMD 8 core CPU, 24GB RAM and 240GB SSDs in a RAID 1. Not to
mention the iPads are getting broken/stolen which makes the whole
process even more expensive. Not that we don't have iPads in our own
househould. Just that LTSP continues to make a ton of sense.
again, my thanks and best regards to all,
William Fragakis

Message: 2
Date: Mon, 17 Feb 2014 13:04:01 -0500
To: "Support list for open source software in schools."
Subject: Re: [K12OSN] networking question
<CAEo=
Content-Type: text/plain; charset="utf-8"
1) bigger switch (big assumption is existing network is not overloaded

now)

2) add a secondary "smart" switch above the existing switch if the

existing

switch supports "upstream" port
3) add a second switch downstream with a bonded or aggregate connection

for

more bandwidth (if "smart" primary switch)
4) add a secondary switch with Gbit upstream and 100M ports and set

primary

switch to 100M except for new switch port
5) add a secondary switch upstream with bonded/aggregate connections to
multiple nics on server
6) add a second switch off second nic on separate subnet and split the
load. This requires a split subnet with dual gateways (each nic) (no

smart

switch required)

clients.

Post by William Fragakis
Between the clients, internal servers and printers, the switch to which
the ltsp server is connected to is at capacity (It's a basic 24 port
gigabit).
We now need to add the n+1 client that exceeds switch capacity. Do we
buy
a) a larger switch to replace the current one
b) a second switch daisy chained to the current one
c) a second switch connected to a second NIC using the same internal
ip/dhcp range (and if so, the recommended manner)
d) a more obvious, elegant, simple, cheaper method that I haven't
considered because I don't know squat (American slang for "not much")
about this stuff.
thanks to all,
William Fragakis
_______________________________________________
K12OSN mailing list
https://www.redhat.com/mailman/listinfo/k12osn
For more info see <http://www.k12os.org>

_______________________________________________
K12OSN mailing list
https://www.redhat.com/mailman/listinfo/k12osn
For more info see <http://www.k12os.org>