[prev in list] [next in list] [prev in thread] [next in thread] 

List:       npaci-rocks-discussion
Subject:    Re: [Rocks-Discuss] IP address change on Frontend
From:       Joe Kaiser <jlkaiser () gmail ! com>
Date:       2010-06-26 21:27:14
Message-ID: AANLkTikgKevwtpOoTc6mt9cnlcDRH_UCmM7dXuKMMBoF () mail ! gmail ! com
[Download RAW message or body]

Reinstall your frontend.

Thanks,

Joe

On Fri, Jun 25, 2010 at 3:40 PM, Kewei Li <ikeweili@gmail.com> wrote:

> We have just relocated our Rocks cluster (version 5.2) to a new network.
> Can someone tell me how should I make it work in the new network? Thank you!
> 
> 
> Kewei
> 
> 
> > -----Original Message-----
> > From: npaci-rocks-discussion-bounces@sdsc.edu [mailto:
> npaci-rocks-discussion-
> > bounces@sdsc.edu] On Behalf Of Anthony Power
> > Sent: Monday, June 21, 2010 3:02 PM
> > To: 'Discussion of Rocks Clusters'
> > Subject: Re: [Rocks-Discuss] IP address change on Frontend
> > 
> > I don't develop for profit.
> > 
> > Our profit is made from our hardware sales, we do not charge for O/S
> installations,
> > I don't even charge professional services to customers to help get their
> code
> > running on a new system, so your assumption in differences is far from
> correct. I
> > don't make commissions off cluster sales, ... and I don’t get any pretty
> bonuses for
> > putting in all the extra time I do working on side projects like this,
> ... so honestly
> > I'd like to think im the same boat as you, .. im just not EDU .. im a
> Vendor
> > 
> > Minus this whole Rocks thing ive been apart of the Linux community for
> over 10
> > years, and I've spent countless hours helping people on other mailing
> lists im on...
> > including this one.
> > 
> > " I'd certainly hope I have 5.3 since I develop rolls for it."
> > 
> > Well, when you get a chance, ... try your whole sed / awk , database
> change, *it’s a
> > cake walk* public ip change, .. and let me know how it works for you :)
> > 
> > I'm done with this conversation, .... I feel like im getting more
> attitude than
> > anything.
> > 
> > Anthony Power
> > Sr Systems Engineer
> > Toll Free : (877) 870-AHPC (2472)
> > Phone : (858) 268-9600
> > Fax : (858) 268-9605
> > anthony.power@advancedhpc.com
> > 
> > 
> > -----Original Message-----
> > From: npaci-rocks-discussion-bounces@sdsc.edu [mailto:
> npaci-rocks-discussion-
> > bounces@sdsc.edu] On Behalf Of Steve Jones
> > Sent: Monday, June 21, 2010 11:35 AM
> > To: Discussion of Rocks Clusters
> > Subject: Re: [Rocks-Discuss] IP address change on Frontend
> > 
> > I'd certainly hope I have 5.3 since I develop rolls for it. It's probably
> a good idea to
> > test and use what we provide to the community. I guess that's the
> difference
> > between vendors and research or academic institutions. Vendors develop
> for
> > profit, research and academic develop to use and often share. Fortunately
> the
> > Rocks group is academic and funded by the NSF, or it would no longer be
> free.
> > 
> > The issue still exists, it's near impossible to create a script for the
> unknown. It's
> > easy for a vendor to create a script for a controlled environment, say
> you build and
> > ship clusters to a client with known rolls installed, they then run the
> script to
> > change the network information. It's near impossible for anyone to write
> a script
> > to change the IP on any system in the wild as people have custom
> configurations,
> > custom rolls and site modifications that you'd have to account for.
> > 
> > Let us know when you have something to test.
> > 
> > Steve
> > 
> > ----- Original Message -----
> > > " Versions since 2.x, but I don't think you've been around since then,
> > > otherwise you wouldn't be asking me this question."
> > > 
> > > ... I've been around for 6-7 years, ... so ya you've using rocks for
> > > longer than I have, .. this isn’t a pissing contest, the reason I
> > > asked was because I was curious to see if you were even RUNNING 5.3 so
> > > maybe you could actually SEE what im talking about.
> > > 
> > > " Please, go ahead and script it up, show your support as a commercial
> > > vendor who gives back to the community. I already develop a high
> > > number of rolls for the community and have for years."
> > > 
> > > I am 100% down to do this, .. but I would need some information from
> > > the Rocks group on exactly what needs to be changed where, and the
> > > order these changes should be made in, ... im asking for SUGGESTIONS
> > > from the group and a little help, .. I never said "YOU GUYS DO IT THIS
> > > SUCKS" ... im saying it would be a valuable asset to the Rocks suite.
> > > 
> > > " I'd have to guess you're not managing systems for years, then have
> > > to move it or change networks. That's an easy route to take as a
> > > vendor building systems, but not practical in the real world."
> > > 
> > > How is it not practical? I just went through this with 3 different
> > > customers who had new data centers built and had to move their
> > > clusters and change their ip information, ... shutting it down,
> > > rolling it to a new room, and plugging it into a new network is pretty
> > > much what I do when I show up on site to install the thing in the
> > > first place. Moving it should be no different.
> > > 
> > > * blah blah blah. Write the script, give back to the community.*
> > > 
> > > Actually I have given back to the community plenty of times, .. I know
> > > who you are, and I know all about your rocks.stanford.com project,
> > > I've worked with Scott on a few things, .. he actually helped me out
> > > with an openmpi issue I had, .. and I was the one that actually
> > > recompiled your openmpi roll with infiniband support it in using the
> > > 5.1 OFED roll, so that openmpi would actually work over IB and play
> > > nice with Torque .. I give back to this community as much as I can, im
> > > in the process of building an OFED roll right now that will build its
> > > openmpi with Torque or SGE support since that’s another common issue
> > > with the current rolls out there.
> > > 
> > > Anthony Power
> > > Sr Systems Engineer
> > > Toll Free : (877) 870-AHPC (2472)
> > > Phone : (858) 268-9600
> > > Fax : (858) 268-9605
> > > anthony.power@advancedhpc.com
> > > 
> > > -----Original Message-----
> > > From: npaci-rocks-discussion-bounces@sdsc.edu
> > > [mailto:npaci-rocks-discussion-bounces@sdsc.edu] On Behalf Of Steve
> > > Jones
> > > Sent: Friday, June 18, 2010 6:16 PM
> > > To: Discussion of Rocks Clusters
> > > Subject: Re: [Rocks-Discuss] IP address change on Frontend
> > > 
> > > Hi.
> > > 
> > > > What version of Rocks are you running?
> > > 
> > > Versions since 2.x, but I don't think you've been around since then,
> > > otherwise you wouldn't be asking me this question.
> > > 
> > > > And how would it be impossible for a universal script / binary to be
> > > > created to make these changes? Obviously "its trivial, one only
> > > > needs to know how to do simple find/replaces, a little sed/awking,
> > > > vi and
> > > > change a few sports in the database and you are like the initial
> > > > installation"
> > > > 
> > > > If its that simple a script can be written to make it simpler, ...
> > > > isn’t that why we write them in the first place?
> > > 
> > > Please, go ahead and script it up, show your support as a commercial
> > > vendor who gives back to the community. I already develop a high
> > > number of rolls for the community and have for years.
> > > 
> > > > Besides adding different rolls, Rocks environments are pretty
> > > > similar, unless you go *way* custom, and even then its usually just
> > > > application packages, partitioning / appliances.
> > > > 
> > > > The thing is there is no documentation on WHERE all this information
> > > > is in the database for one, ... so unless you want to spend all day
> > > > digging through the thing, finding all the tables and fields etc etc
> > > > etc .. unless you have over 50 nodes, id rather just reinstall the
> > > > damn thing.
> > > 
> > > I'd have to guess you're not managing systems for years, then have to
> > > move it or change networks. That's an easy route to take as a vendor
> > > building systems, but not practical in the real world.
> > > 
> > > > Minus install time, I can roll out a cluster, have it up and running
> > > > and processing jobs in a few hours.
> > > > 
> > > > I work with TONS of people that swear by Rocks, .. and it really
> > > > sucks when this issue comes up and I tell them "well, our best bet
> > > > is to
> > > > reinstall it" ... I actually have a complete write up on how to
> > > > rebuild a cluster step by step just for this case, ... whether its
> > > > Torque / SGE / GigE / OFED / Qlogic OFED / etc /
> > > > 
> > > > All we are doing is changing the public interface, .. a core
> > > > component of the Rocks system, .. well of any cluster really, ...
> > > > and anyone who
> > > > has been an admin for any group of systems knows there are times
> > > > when you have to move machine(s) to new ip ranges, ... finding a way
> > > > to simplify this in Rocks instead of playing surgeon with the rocks
> > > > core database and config files would be very beneficial to everyone,
> > > > especially hardware vendors like myself who preload Rocks on
> > > > clusters as a value add for customers, .. I deal with a lot of
> > > > research / edu /
> > > > etc who have no idea how to really build / maintain a clustered
> > > > environment, so, using Rocks, which is an amazing package makes
> > > > their lives A LOT easier.
> > > > 
> > > > Because if you slip up while playing surgeon on your cluster, ...
> > > > you're going to have to reinstall anyways :P ... and its not that
> > > > easy to remote into a cluster, ... to change its public interface :)
> > > > and walking through someone who knows how to run jobs, .. but not a
> > > > cluster, on how to rebuild it can be somewhat frustrating :)
> > > > 
> > > > Maybe even something as simple as creating a script that builds a
> > > > NEW database (like the one that’s created at install) ... but with
> > > > the new
> > > > IP information, ... throws out some changes in some config
> > > > scripts... dumps the running database, ups the new one and then
> > > > pushes a
> > > > cluster-kickstart out to all the nodes so they are reimaged with
> > > > this information.
> > > > 
> > > > It could even query the running database so that it pulls all the
> > > > node information (hostnames, ip's, mac's, etc) ... and even if you
> > > > had a
> > > > super custom extend-computes file with a crazy node config, ..
> > > > nothing would really change.
> > > > 
> > > > But the public ip.
> > > 
> > > blah blah blah. Write the script, give back to the community.
> > > 
> > > Steve
> 
> 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20100626/e3e5aacb/attachment.html \



[prev in list] [next in list] [prev in thread] [next in thread]