[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Globus-discuss] Globus and IPv6

To: Tim Chown <tjc@ecs.soton.ac.uk>
Subject: Re: [Globus-discuss] Globus and IPv6
From: Joseph M Link <link@mcs.anl.gov>
Date: Wed, 2 Oct 2002 11:05:20 -0500
Cc: "Nicholas T. Karonis" <karonis@niu.edu>, Azmi Bin Mohamed Ridwan <azmi@bii-sg.org>, Shyamal Mitra <mitra@agni.tacc.utexas.edu>, discuss@globus.org, David G Mills <dgm300@ecs.soton.ac.uk>, Michael Saywell <ms@ecs.soton.ac.uk>, Peter Kirstein <P.Kirstein@cs.ucl.ac.uk>, Sheng JIANG <s.jiang@cs.ucl.ac.uk>
In-Reply-To: <20021002131215.GF13594@starling.ecs.soton.ac.uk>
References: <Pine.GSO.4.43L0.0210021903280.20351-100000@rebecca.krdl-core.bii> <200210021254.g92CsJCa023566@olympus.cs.niu.edu> <20021002131215.GF13594@starling.ecs.soton.ac.uk>
Sender: owner-discuss@globus.org
User-Agent: Internet Messaging Program (IMP) 3.2-cvs

Quoting Tim Chown <tjc@ecs.soton.ac.uk>:

> There is some Globus 2 porting work being done by UCL (and to some extent
> us).  There was a posting to the list about 5-6 week ago by one of
> the Argonne guys saying they were going to produce a protocol independent
> Globus, but it has gone very quiet since then... :(

I have just started back on this project this week.  I do plan to make it 
protocol independent from the start.  However, my work will be biased towards 
the GridFTP side of things in the beginning and I cant give an estimate on how 
long it will take to push those changes through the entire toolkit.

Some of the major issues I have seen at a glance are openssl's, openldap's and 
GSI's dependance on ipv4.  I have little control over these libraries and will 
have to work with others on porting them.

btw, my idea of protocol independent is still IP specific.  If you think we 
should take this a step further, please, plant some thoughts into my head now 
before I get to far in.  At first thought, though, I dont see an need for any 
other protocols in the toolkit.  Also, we actually have another library in the 
works that will bring protocol independence to the table for free.

Joe

> The problem is that the way the code is written it is possible to port it
> to work over IPv6 (which a Japanese group did for Globus 1) but to make
> it protocol independent is harder.
> 
> Of course it would be nice if Globus 3 shipped out of the box with IPv6
> support...
> 
> Tim
> 
> On Wed, Oct 02, 2002 at 07:54:19AM -0500, Nicholas T. Karonis wrote:
> > Azmi,
> > 
> > You bring up a good point, but like everything else, this is a problem
> > of limited resources.  We want to solve this problem but we simply
> > haven't had the time to address it.
> > 
> > Thanks for your input.
> > 
> > Nick
> > 
> > Azmi Bin Mohamed Ridwan writes:
> > >Hi there,
> > >
> > >Are u guys working on this problem now? If so, how close are u to a
> > >possible solution?
> > >
> > >I believe clusters with only the head node with public IPs and the
> compute
> > >nodes which are on private IPs are the way most people are setting up
> > >their systems due to the fact that there isn't enough public IPs to go
> aroun
> > >d.
> > >(Who knows when IPv6 is fully implemented - maybe ALL machines can get
> > >their own public IPs). In my opinion, this drawback will seriously
> hamper
> > >G2's implementation and usage especially to those restricted to Class C
> > >IPs.
> > >
> > >Just putting in my 2 cents worth :)
> > >
> > >Cheers ...
> > >
> > >=====================================
> > >Azmi Bin Mohamed Ridwan
> > >Grid Computing Training Program
> > >BioInformatics Institute
> > >21 Heng Mui Keng Terrace
> > >Singapore 119613
> > >Tel: 8741273
> > >=====================================
> > >
> > >On Wed, 2 Oct 2002, Nicholas T. Karonis wrote:
> > >
> > >> Dear Shyamal,
> > >>
> > >> Unfortunately at this time there are no solutions in Globus or
> MPICH-G2
> > >> that gets around the private IP problem.
> > >>
> > >> Nick
> > >>
> > >> Shyamal Mitra writes:
> > >> >
> > >> >Hi,
> > >> >
> > >> >I have a similar situation. We have a cluster with 16 nodes. There
> > >> >is a login node and the other nodes are on a private subnet. We
> > >> >have Globus 2.0 and MPICH-G2 on the front node. There is no scheduler
> > >> >like PBS on this cluster.
> > >> >
> > >> >There is a regular mpich-1.2.3 version installed that we use to
> submit
> > >> >mpi jobs. The "machines" file has the list of nodes running from
> node0
> > >> >to node15.
> > >> >
> > >> >The "machines" file for MPICH-G2 has a single entry
> > >> >"xxxx.xxx.xx.xx" 16
> > >> >
> > >> >where the xxx's represent the fully qualified domain name of the
> > >> >login node.
> > >> >
> > >> >I use the fork job manager to launch a Globus MPICH-G2 job using
> > >> >mpirun -globusrsl file.rsl
> > >> >
> > >> >n number of processes are created as specified by the count variable.
> > >> >But these processes run on the front node!
> > >> >
> > >> >The "hello world" program from the www.globus.org/mpi page runs fine.
> > >> >There are no firewalls on this machine so the comments on how to run
> > >> >Globus jobs through a firewall do not apply.
> > >> >
> > >> >What am I missing?
> > >> >
> > >> >Does the fork job manager need to be configured to send jobs to the
> compu
> > >te
> > >> >nodes? If so, how?
> > >> >
> > >> >Does the binary image of any Globus packages need to be on each of
> the
> > >> >compute nodes?
> > >> >
> > >> >Thanks for your help and suggestions.
> > >> >
> > >> >Regards,
> > >> >
> > >> >Shyamal Mitra
> > >> >
> > >> >On Thu, 29 Aug 2002, Nicholas T. Karonis wrote:
> > >> >
> > >> >> Dear Sangmin,
> > >> >>
> > >> >> This sounds to be a firewall issue.  There is a new section on
> > >> >> the MPICH-G2 page that addresses firewalls with Globus and
> MPICH-G2.
> > >> >> Please see the section entitled "Firewalls" on the MPICH-G2 web
> > >> >> page www.globus.org/mpi.
> > >> >>
> > >> >> Nick
> > >> >>
> > >> >> "Sangmin Park" writes:
> > >> >> >We have a cluster with 8nodes in the LAB and it's connected with
> priva
> > >te
> > >> >> >ip.
> > >> >> >And I installed Globus and MPICH on the master node of cluster.
> > >> >> >
> > >> >> >And there are other workstations installed Globus and MPICH.
> > >> >> >
> > >> >> >I've tried to submit mpi jobs on the cluster and workstations.
> > >> >> >But the job was failed because of private ip problem.
> > >> >> >I know cluster's computing node cannot exchange messages with
> > >> >> >other workstations or clusters because it has a private ip.
> > >> >> >
> > >> >> >Then, is there anyway solve this problem.
> > >> >> >I know configuring the computing nodes with public ip makes it
> work.
> > >> >> >But we don't have enough public ip addresses.
> > >> >> >
> > >> >> >Please let me know about the solution.
> > >> >> >I'm always grateful for all your advices.
> > >> >> >Thank you.
> > >> >> >
> > >> >> >
> > >> >> >
> > >> >> >
> > >> >> >
> > >> >> >
> > >> >> >
> > >> >> >Graduate School of Computer Science, Ajou University
> > >> >> >Sangmin Park
> > >> >> >
> > >> >> >
> > >> >>
> > >> >
> > >>
> > >
>

References:
- Re: [Globus-discuss] private ip problems on MPI program.
  - From: Azmi Bin Mohamed Ridwan <azmi@bii-sg.org>
- Re: [Globus-discuss] private ip problems on MPI program.
  - From: "Nicholas T. Karonis" <karonis@niu.edu>
- Re: [Globus-discuss] Globus and IPv6
  - From: Tim Chown <tjc@ecs.soton.ac.uk>

Prev by Date: [Globus-discuss] Help building and installing Globus v2.1.3
Next by Date: [Globus-discuss] MDS Compatibility problems
Prev by thread: Re: [Globus-discuss] Globus and IPv6
Next by thread: Re: [Globus-discuss] private ip problems on MPI program.
Index(es):
- Date
- Thread