Confusing data from cool new Epoptes benchmarking tool

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Confusing data from cool new Epoptes benchmarking tool

David Groos
Hi All,

The computers connected to my slowest switch always boot almost a minute earlier than those connected to my faster switch! And furthermore, the slower switch has to go through the faster switch to get to the server.

I've got a teacher-computer serving, via ltsp-pnp, a classroom network with 30 fat clients. This server is connected to a Netgear 24 port switch (model FSM 726) through 1 of its 2 gig ports: the rest of the ports are 100 MB.

The other gig port on the Netgear switch is connected to a Cisco 24 port switch (only has 100 MB ports) on the other side of the room.

Look if you will at the Epoptes benchmark tool results (http://tinyurl.com/p6ewzxa) the faster-booting computers that boot off the older Cisco switch average around 25 Mbps (7A-8B), this makes sense since they share a single 100 mb path to the server. The slower-booting computers that boot off the faster Netgear switch average around 70 Mbps. these numbers all makes sense.

So, why do those last 4 computers, the ones averaging around 25 MB/s each, boot long before the other 12 computers, even though those 12 computers are directly connected to the first gig switch and are getting faster service (apparently)? This doesn't seem a flow control issue since the benchmark clearly shows the slow-booting computers have faster communications.

Clue? the slow-booting computers take a loooong time to get an address from DHCP. Any ideas?

Thanks,
David

--
edubuntu-users mailing list
[hidden email]
Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/edubuntu-users
Reply | Threaded
Open this post in threaded view
|

Re: Confusing data from cool new Epoptes benchmarking tool

Άλκης Γεωργόπουλος
On 24/10/2015 06:06 πμ, David Groos wrote:
> Clue? the slow-booting computers take a loooong time to get an address
> from DHCP. Any ideas?


Hi David,

1) Temporarily try with ipappend 3 in pxelinux.cfg/default (google for
instructions). This will eliminate the DHCP delay and let you compare
the boot times more easily.

2) If your router has 4 LAN ports, use it as your "main switch", i.e.
connect all the other switches directly to the router, don't chain them.

3) What are the client specs? Better CPUs boot faster...

4) Bootchart can measure boot time delays.


--
edubuntu-users mailing list
[hidden email]
Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/edubuntu-users
Reply | Threaded
Open this post in threaded view
|

Re: Confusing data from cool new Epoptes benchmarking tool

David Groos
In reply to this post by David Groos
Will try this evening, thanks! David


Sent from my T-Mobile 4G LTE Device


-------- Original message --------
From: Alkis Georgopoulos
Date:10/24/2015 7:20 AM (GMT-06:00)
Subject: Re: Confusing data from cool new Epoptes benchmarking tool

On 24/10/2015 06:06 πμ, David Groos wrote:
> Clue? the slow-booting computers take a loooong time to get an address
> from DHCP. Any ideas?


Hi David,

1) Temporarily try with ipappend 3 in pxelinux.cfg/default (google for
instructions). This will eliminate the DHCP delay and let you compare
the boot times more easily.

2) If your router has 4 LAN ports, use it as your "main switch", i.e.
connect all the other switches directly to the router, don't chain them.

3) What are the client specs? Better CPUs boot faster...

4) Bootchart can measure boot time delays.


--
edubuntu-users mailing list
[hidden email]
Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/edubuntu-users

--
edubuntu-users mailing list
[hidden email]
Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/edubuntu-users
Reply | Threaded
Open this post in threaded view
|

Re: Confusing data from cool new Epoptes benchmarking tool

David Groos
Thanks Alkis! This is looking like I'll have to study/backup a bit deeper so I'll be testing this on this upcoming Thur/Fri after school. Quick question, I found this page: https://help.ubuntu.com/community/UbuntuLTSP/ProxyDHCP. Will I need to make any adjustments for 14.04.3?
David G

Adjusting pxelinux.cfg/default for Ubuntu 12.04

(cat <<EOF
ipappend 3
EOF
) | sudo tee -a /var/lib/tftpboot/ltsp/i386/pxelinux.cfg/default

Additional step for 12.04

Installing ltsp-server does not install tftpd-hpa, which is a necessary component needed in order to load tftp images to the client computers upon pxe boot. The tftpd-hpa program listens on port 69, therefore in order to transfer the ltsp image files to the clients it must be installed. There is no additional configuring to do once it is installed. APT should install and start the program, you do not have to "ltsp-update-image", the syntax:

sudo apt-get install tftpd-hpa



On Sat, Oct 24, 2015 at 11:08 AM, David <[hidden email]> wrote:
Will try this evening, thanks! David


Sent from my T-Mobile 4G LTE Device


-------- Original message --------
From: Alkis Georgopoulos
Date:10/24/2015 7:20 AM (GMT-06:00)
Subject: Re: Confusing data from cool new Epoptes benchmarking tool

On 24/10/2015 06:06 πμ, David Groos wrote:
> Clue? the slow-booting computers take a loooong time to get an address
> from DHCP. Any ideas?


Hi David,

1) Temporarily try with ipappend 3 in pxelinux.cfg/default (google for
instructions). This will eliminate the DHCP delay and let you compare
the boot times more easily.

2) If your router has 4 LAN ports, use it as your "main switch", i.e.
connect all the other switches directly to the router, don't chain them.

3) What are the client specs? Better CPUs boot faster...

4) Bootchart can measure boot time delays.


--
edubuntu-users mailing list
[hidden email]
Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/edubuntu-users


--
edubuntu-users mailing list
[hidden email]
Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/edubuntu-users