Bug 463330

This request is for Red Hat to do.  Thanks

In order to consider adding this support, we're going to need two things:  First, we don't have the hardware in installer land to develop or test this feature.  Even if we can get someone else to do the development, having hardware with us really helps when we need to do bug fixing.  And even if QA here has hardware, that's usually not very helpful to doing development.

Second, we're going to need the expertise to know what needs to be done to support this.  Ideally, someone here could talk to someone there to see what needs to be done, discuss UI mockups, learn where the problem parts are going to be, etc.  Having some reference documentation might also be useful.

Having both of those things, we can then work on adding this to a Fedora release and then get it into RHEL.  Is there any way you can help with the two things we need in order to get this bug moving?

We need a technical contact to describe what needs to happen here.

There's not much information in this bugzilla, but we are guessing that 
this is something which requires more then just adding a kernel driver and letting udev load it for us. 

IOW this seems like something which needs manual setup from anaconda like iscsi / FCoE.

If this is the case, it is too late for 6.0 - without more information, I have to go on that assumption and nak this.

But maybe we just need to add an additional kernel module somewhere and it's just not clear from this BZ. What we need most of all is more info. What would really help is a technical contact inside IBM who can explain the details of this implementation in an irc dialogue.

adding Bob Sisk, LTC technical contact.  Bob, can you please find a contact to answer Denise's questions?

John, I am looking for someone. Bob

------- Comment From tsenglin.com 2009-12-17 16:25 EDT-------
ib_ipoib.ko is an InfiniBand Upper Layer Protocol (ULP) kernel module. By loading ib_ipoib.ko after loaing an InfiniBanf hardware driver (e.g. ib_mthca.ko, iw_cxgb3.ko), one or two (depending on the number of ports the InfiniBand adapter has) ibX network devices are created. An ibX device is similar to an ehernet ethX device. It can be configured with network configure files /etc/sysconfig/network-scripts/ifcfg-ibX just like ethernet devices.

At this moment, users need to create ifcfg-ibX configuration files manually. What this bug requests is to add IPoIB as a known network device in anaconda to allow users to configure ibX devices with anaconda.

There are two separate requests here:

1) Based on my understanding, we need anaconda to load ib_ipoib.ko if ib_mthca.ko or iw_cxgb3.ko gets loaded.  If that's the case, why isn't that module always loaded?  Is it a module that users may not always want to load?

Assuming the ibX devices work like an Ethernet device, they will show up in the loader network configuration screens and you can assign IP addresses or choose DHCP or whatever.  Is this the case?  Are ibX devices essentially Ethernet and use the same kinds of configuration settings or are there other things that need to be requested?

2) Since we use NetworkManager in anaconda now, nm-system-settings may have to be modified to support ifcfg-ibX files as well as the configuration interface supporting ibX interfaces.

------- Comment From tsenglin.com 2010-01-21 18:56 EDT-------
(In reply to comment #10)
> There are two separate requests here:
> 1) Based on my understanding, we need anaconda to load ib_ipoib.ko if
> ib_mthca.ko or iw_cxgb3.ko gets loaded.  If that's the case, why isn't that
> module always loaded?  Is it a module that users may not always want to load?
> Assuming the ibX devices work like an Ethernet device, they will show up in the
> loader network configuration screens and you can assign IP addresses or choose
> DHCP or whatever.  Is this the case?  Are ibX devices essentially Ethernet and
> use the same kinds of configuration settings or are there other things that
> need to be requested?
> 2) Since we use NetworkManager in anaconda now, nm-system-settings may have to
> be modified to support ifcfg-ibX files as well as the configuration interface
> supporting ibX interfaces.

ib_ipoib.ko creates ibX devices on top of InfiniBand devices. It creates N ibX devices if it sees N InfiniBand ports. Hence InfiniBand device driver, e.g. ib_mthca.ko, need to be loaded before ib_ipoib.ko. It does not create any ibX device if no InfiniBand device driver is loaded. However, it won't break anything. Hence, the answer is "yes, it can be loaded always".

The ibX devices are network devices on top of InfiniBand fabrics. They are not exactly the same as ethernet due to different media and hardware type. However, they work just like ethX devices for every thing above device driver level. For static IP, you can just create ifcfg-ibX files and use ifup/ifdown to configure ibX dveices. For DHCP, since DHCP needs to obtain the hardware address (MAC), you need to make sure the DHCP code supports InfiniBand.

We are out of capacity to complete this for RHEL 6.0, adding to RHEL 6.1 tracker for consideration for that release.

Who is willing to provide hardware access for development and testing?  Before we can seriously consider this feature for any release, we need to have hardware.

If someone (Penguin Computing?) were to provide hardware, who would we contact to make the equipment available?

(In reply to comment #12)
> If someone (Penguin Computing?) were to provide hardware, who would we contact
> to make the equipment available?    

Please email myself and John Jarvis directly and we'll help you out:

dcantrell
jjarvis

How's this progressing, and do you have an idea of what hardware you need?

We have spare Infiniband network cards and cables here that could be shipped if there's still a need.

However, it's probably easier to get Vendors involved and they'll likely ensure new hardware gets given into the future too.

people interested in this bug may also want to look at bug 590578, which is about adding support to anaconda for storage connected through infiniband.

------- Comment From rsisk.com 2010-10-04 10:42 EDT-------
Code Upstream Status: No Code Required

*** Bug 641031 has been marked as a duplicate of this bug. ***

IBM is signed up to test and provide feedback, setting OtherQA.

This request has been deferred from RHEL 6.1 but added to the list for consideration of inclusion in RHEL 6.2.  Resetting flags, adding to the ibm6.2 tracker.

I need access to hardware to do development on this.  Please contact hamzy.com