Re: Beowulf Cluster is very slow. Suggestions needed to increase the speed.

Tomcio <[email protected]> Tue, 14 Oct 2014 19:13:22 +0200
Newsgroups gmane.linux.debian.devel.beowulf
Message-ID <[email protected]>
Well..

It depends.
Based on what You're writting I assumed gateway (router) is routing 
traffic between Your machines or You're using 100Mbit switch for Your 
Gigabit-available nodes or it's just too slow switch.

When this situation occurs:
1 system - 3 months
4 systems 100Mbit - 1 year (4 times longer than single node)
4 systems 1000Mbit - 4 months (1,33 times longer than single node)
Then there is still something misconfigured or Your nodes aren't 
optimized to work in cluster or Your switch is overloaded.

The task should take less than 3 months, let's say 1 month or less in 
optimized cluster.

You could check network load on Your switch, but this model isn't even 
web managed. Also, it's still relatively low performance (16 Gbps 
switching capacity, 1,48Mpps per port, 15us switching delay).
So on Your servers there are available programs to measure network usage 
like: ntop, dstat or iptraf (the last one not always show the correct 
results) to help troubleshooting network performance.

Regards,
TooMeeK

W dniu 2014-10-14 16:42, suresh kannan pisze:
> Hi,
>
>  >>>First of all, why do you think network is the bottleneck?
>
> Thank you for your concern.
>
> If i run a specific job in a single system (without parallel) it showed
> three months (aprox) time in a quad core processor.  When i did parallel
> using four systems it showed October 2015. I was using 100 Mpbs (ip time
> router). I have conformed that all the systems uses the processors
> [using "top" command]. After that TooMeeK point out i have to use
> "Gigabit routing switch supporting layer 3". I dont have that switch in
> our lab. However, we had Netgear GS608 1000 Mpbs switch it reduced the
> time (Feb 2015). Moreover, I read the link
> http://cs.boisestate.edu/~amit/research/beowulf/beowulf-setup.pdf
> <http://cs.boisestate.edu/%7Eamit/research/beowulf/beowulf-setup.pdf>
> suggested by TooMeek. They have also suggested that Network switch is
> important.  Also, I saw few videos in youtube about layer 3 switch
> capabilities. Therefore, I am convinced that "Gigabit routing switch
> supporting layer 3" will solve this issue.
>
> regards
> Suresh
>
>
>