ajdecon's questions

I have an HP Proliant SL250s Gen8 (right-hand) server which is apparently not able to boot from the network. If I configure the system to PXE boot in the BIOS, or if I hit F12, it does start up the Intel boot agent (for netboot) and claims to be attempting to DHCP. However, the system does not boot. In addition, running tcpdump -i eth0 -s 1500 port bootps or port bootpc on the PXE server during boot does not show any DHCP requests from the server MAC address.

I have confirmed that a second server, using the same switch port and cable, is capable of booting using my PXE server. When booting the second server, the above tcpdump command successfully captures the DHCP request.

I haven't worked with HP servers much before. Is there a "make it boot" button in the BIOS somewhere that I've missed? ;-)

Is there any way to create a link or "saved profile" for launching an EC2 instance with a given configuration?

I'm aware of the mechanism where you can create a link to launch a given AMI, i.e. http://console.aws.amazon.com/ec2/home?region=us-east-1#launchAmi=ami-XXXXXX. However, there are still a lot of configuration steps left (instance type, security group, etc).

Is there anyway to automate this process, in such a way that it can be launched by an arbitrary Amazon account?

I recently inherited an Infiniband network containing multiple switches, and I know that one of these switches is running the subnet manager. The rest supposedly have that feature turned off, or were never enabled. The trouble is, I have no idea which one it is...

I'd like to replace the switch subnet manager with OpenSM running on a couple of my infrastructure servers. Is there any way, short of logging into each switch individually, to determine which switch is running the SM?

This might be a silly question, but... Is there any hypervisor or other virtualization solution which supports a "virtual IPMI" for virtual machines? I.e. would allow access to something that looks like power control, a serial-over-LAN console, etc. to a virtual machine.

This is really not the right way for managing virtual machines, I know. I'm looking for this because we make heavy use of IPMI for systems management and it would be interesting/useful to integrate VMs into the same framework, and because it'd be useful to have such an interface when I use VMs as a dev environment for systems management tools.

As I understand it, the subnet manager of an Infiniband network calculates the best routes between each pair of nodes on the network and provides these routes to the nodes when they want to communicate. Is there any way to get the subnet manager (and in particular, OpenSM) to print out these routing tables in a human-readable fashion?

At the moment this is purely for my own curiosity, but I can imagine cases where this would be useful for debugging.

iWARP is a protocol for doing remote direct memory access (RDMA) on top of TCP/IP, so that it can work with Ethernet and other network types as opposed to Infiniband. It works with many of the standard IB interfaces - the IB verbs, for example - so it's all pretty transparent.

I'm doing some IB-verbs programming (mostly for the sake of learning about how they work better), and it'd be wonderfully convenient for me if I could use iWARP to do RDMA over my loopback interface, so that I could test some of my code without getting on our IB-connected cluster. :-)

But I cannot figure out how to get a "local development environment" set up: there are no tutorials I'm aware of for even setting up iWARP from scratch on a server or a network interface. Can anyone give me a tutorial or point me in the right direction?

Environment is Fedora 16 running in VirtualBox.

We're building a tool right now which includes booting a tiny Linux environment via PXE, and having it perform a number of system configuration operations without loading a larger system. (Formatting disks, creating filesystems, etc.) We've been planning to use BusyBox in this system, as having most of the system tools in one tiny static binary is obviously great in this context.

Problem, though: we've had a requirements change, and need support for creating ext3 (and maybe ext4) filesystems, which BusyBox does not support. Is anyone aware of a similarly self-contained binary tool which can be used for creating these filesystems? It doesn't need to also be BusyBox (I can include a few extra tools), but it can't require a bunch of dynamic libraries.

I'm trying to figure out how to set up a simple storage system which exports block storage over Infiniband, using either SRP or iSER.

I'm very early in the process, and at the moment I'm basically just looking for a tutorial on the level of, "You have two servers on your fabric: here's how to export a block storage device from one to the other." Preferably on RHEL or variants thereof. Any pointers?

We are currently running an interactive HPC application which presents a graphical interface to the user, attaches to an HPC cluster and allows them to run and observe some computation. The user logs in to a front-end node via NoMachine NX Server (this machine does not participate in the computation). He typically sets up his problem, does a few tiny trial runs, then starts a large job. After that he disconnects from the NX session, expecting the computation to continue.

Except it doesn't. All execution within the NX session, and across the cluster, seems to pause when the user disconnects. If he resumes a session he can resume the computation, but this is a job which he expects to run for several days, so it may not be feasible to expect a connected NX session throughout.

We're aware that in many ways the correct use-case would be for the user to figure out his parameters and then submit a batch (non-interactive) job via ssh, but he strongly prefers the workflow I outlined above so we're trying to make it work.

The cluster resource manager Torque typically allocates compute nodes on an exclusive basis. However, when you have a lot of small jobs (like we do) running against multi-core compute nodes, this can result in a lot of wasted resources. Is there any way to configure Torque to allow non-exclusive allocation of the cores on a compute node?

(These jobs are all embarrassingly parallel, so we aren't concerned about contention for the shared network resource. We can't switch schedulers as our customer's job-scripts are all in PBS/Torque.)

I currently have a domain, call it example.com, hosted on Dreamhost, and also a Linode VPS. The Linode lives in a subdomain of the Dreamhost domain, foo.example.com. What I want is to configure reverse DNS so that it will report itself as being at foo.example.com.

What I don't know is how to accomplish this. I don't understand DNS very well, and while Linode seems to have facilities for managing a DNS slave zone, I'm worried about screwing up the DNS on Dreamhost. I still want the domain registered there for unrelated reasons.

We are currently in the process of setting up several Windows HPC clusters using Windows 2008 HPC R2. We would like to be able to perform Windows licensing commands across the clusters via the command line, i.e. display the current license status, install a product key, re-arm a trial key, etc.

Unfortunately, the slmgr command always opens a dialog box to display its output, rather than writing to standard output. This is extremely unpleasant, as we'd like to be able to run parallel operations on the license status using the HPC tools.

Can anyone suggest a command for managing Windows licenses that works completely on the command line?

We have a couple of servers (part of an HPC cluster) in which we're currently seeing some NFS behavior which is not making sense to me. node1 exports its /lscratch directory via NFS to node2, mounted at /scratch/node1. node2 also exports its own lscratch, which is correspondingly mounted at /scratch/node2 on node1.

Unfortunately, whenever I attempt to mount either NFS export on the opposite node, I get the following error:

mount: node1:/lscratch failed, reason given by server: Permission denied

This despite the fact that I have included first the IP range (10.6.0.0) and then the specific IPs (10.6.7.1, 10.6.7.2) in /etc/exports.

Any suggestions?

Edit to remove ambiguity: I've made sure that exports only contains either the range, or the specific IPs, not both at the same time.

We have a cluster being used to run MPI jobs for a customer. Previously this cluster used Torque as the scheduler, but we are transitioning to Grid Engine 6.2u5 (for some other features). Unfortunately, we are having trouble duplicating some of our maintenance scripts in the Grid Engine environment.

In Torque, we have a prologue.parallel script which is used to carry out an automated health-check on the node. If this script returns a fail condition, Torque will helpfully offline the node and re-queue the job to use a different group of nodes.

In Grid Engine, however, the queue "prolog" only runs on the head node of the job. We can manually run our prologue script from the startmpi.sh initialization script, for the mpi parallel environment; but I can't figure out how to detect a fail condition and carry out the same "mark offline and requeue" procedure.

Any suggestions?

OK, I know the basics of how to set up passwordless ssh: create a keypair using ssh-keygen then add the public key from id_dsa.pub into the host's .ssh/authorized_keys. So for two different machines client and host, I can ssh from client into host without entering a password. Good so far.

But I have some software which requires that client be able to ssh into itself without a password. Naively I do the same thing: simply copy .ssh/id_dsa.pub into .ssh/authorized_keys, same directory. But it doesn't work. If I type

user@client:~$ ssh client

then it still asks for my password, even though the public key of client is in its own authorized_keys file.

Am I missing something?