But the wifi in RPi is connected, is this reason it fail to get ip from osx?
i tried to ifdown and up the eth0, i saw it requesting ip through 255.255.255.255, is this normal? because the OSX self assigned ip (on OSX) is something like 169.254.XXX.XXX

try disconnecting the wifi and see? for me,169.254.xxx.xxx only appear when the OS X cannot get an IP,hence self assign

Tried before...and it is no problem to get ip when connect it to router, so maybe osx side need to set something?but so far i only can check the internet sharing, nothing much i can do.

Your internet sharing setting?can print screen?

WTS Raspberry Pi 2 complete with heatsinks, noobs m/c, cover, wifi adapter and Pi NoIR camera module for RM250. the whole thing just use for a few times then kept in box... Price is negotiable. TQ

Sold to forumer Mousqy.

This post has been edited by MrLoo: Aug 5 2015, 11:22 PM

Sorry for OT but, any of you know where to get odroid C1?

OSX should have an Internet sharing feature like Windows and Linux w/ GUI where the DHCP Server is automatically run when u enable internet sharing. Otherwise, u can install a 3rd party dhcp server and give ip to ur raspi

Try and run your parallel program and run top. Look for wa and if it is high then IO is the issue in your setup(I guess IO should be the main issue for distributed or parallel for nodes in the same network)

Do you have a thread on garage sales? Wish to have a look on details

erm i dunno to open in which thread on garage sales so try my luck here. btw if u wan more details and photo just whatsapp me. i will send thru there TQ

I don't have access to the cluster currently,on holiday

but i made a no. thread vs time taken graph for two program


- MPI Prime number program


-MPI 2D Poisson equation using Jacobi algorithm

by right the performance graph should decrease exponentially,but the Prime number program doesn't perform as expected,as u mentioned,most probably caused by high IO

for the 2D Poisson equation using Jacobi algorithm,initially i set the no. of interior vertices and fractional accuracy for the program to complete faster,which almost the same graph as the prime program.
so i change the no. of interior vertices and fractional accuracy to be higher,which increase the required computing power for each Pi to get the exponential decrease graph,which indicates that the IO latency have become insignificant

anyone made a rp2+openelec+usb dvb-t tuner setup yet?
wanna know which tv tuner works

here is the setting.... just check the sharing option only..
[attachmentid=4696740]

Any suggested good 3rd party dhcp server?

seem like your setting from OS X is ok,did u set static IP or DHCP assigned for your Pi?

le me not OSX user.

Anyway, plug ur raspi back to your router and edit this file: /etc/network/interfaces



At your mac, set the ip of your ethernet interface(lan) to
ip: 10.10.10.11
netmask: 255.255.255.0

Plug your raspi to ur mac and access ur raspi via 10.10.10.10

the eth0 for rpi is set to dhcp.. i hvae tried to plug it in router, it can get ip form router... no sure why OSX is not working.

it that any special by setting to 10.10.10.10?
i have tried to set static on Rpi and OSX before, it worked, can ssh into RPi form OSX, but the problems was: Rpi cannot access internet via wifi any more.

My purpose connecting OSX to Rpi via LAN is for the case whne i cant access the internet ( when i bring it to some place which i do not have the wifi password), i got an other way to connect it, which is LAN for this case.

no, as well as you put a private ip that is not used by any devices at your network then it will be okay (i.e 192.168.x.x)
replace x with any number from 2 - 253.

Anyway, the above is just for additional knowledge

Do the "2. Set up the client" where you will set Using DHCP for configuring IPv4 and it should work. Set back your raspi to use dhcp if you've changed it to static ip previously.
http://www.maclife.com/article/howtos/how_...ection_ethernet

Try and see if it works. I don't have a Mac so can't try it out personally

This post has been edited by wcypierre: Aug 5 2015, 08:28 PM

for your 2nd one, when you wanted more accuracy, everything takes a longer time to complete. but eventually as you add more nodes your issue should still be on IO, because as you can see from the prime num test, when the requests is faster, the possibility of each node sending data back to the server at the same time is higher, so if your server node still runs on sd card, or even worse usb thumb drive(for storage of data), you should consider using a normal computer(desktop/laptop) or at least run raspi with HDD as the server node to solve the io issue and you should see a better result.

btw, you should plot it with nodes/time taken, because let's say you have 6 threads, I don't know if you're giving node 1 and spawn 5 threads and node 2 spawn 1 thread or node 1 2 threads, node 2 2 threads, node 3 2 threads, and the combination etc.

I believe this is some academic research stuff for uni/company?

This post has been edited by wcypierre: Aug 5 2015, 08:44 PM

the Pi cluster was just a proof of concept not performance crunching and yea,its really cool to have a cluster i drawn to a conclusion that it all depends on how you split the task between all the nodes,good programming is required (some take decade just to write an optimised program).

the threads are actually simple,each Pi have 4 cores,so 4 threads,its balanced out (if u try to assign 5 process into one Pi,you would end up slowing down the whole program)

i did it just for fun since my lecturer had so many Pi 2 and didn't know what to do with it,but of course,academic ($$$++)