It seems that there is a naming problem with
INIT_SET_IB_IN_SIZE. I guess it should be SET_IB_IN_SIZE.
Shahar
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, pl
Eitan,
Your analysis is not completely accurate. The SM configure the
subnet using direct mads only, and it builds a spanning tree of direct
routes. What I want to say, is that that it doesn't matter why exactly a
port is unreachable. Once a port can not be reached, you can either
retry t
port the Simulator to gen2.
I am not aware to any
other opensource simulators.
Shahar
From:
abhijitngpune [mailto:[EMAIL PROTECTED]
Sent: Wednesday, March 30, 2005
6:52 AM
To: shaharf
Cc: openib-general@openib.org
Subject: Re: RE: [openib-general]
OpenSM
hi,
Is
Hi abhijitngpune,
OpenSM do not know care
about the topology of the network. Every connected graph is valid for it. BTW,
fat tree can have cycles too. If I don’t err, the algorithm used by the
OpenSM is a variation of some well known graph algorithm invented by Dijkstra
or based on one
I am not sure I follow. Do
you mean that for example, a separate Qos tag is used for each stream? If this
is the case I am not sure how to handle Qos tags from the VM.
I thought about this so that at a higher level you can
use policy routing or QoS or traffic shapping or maybe even C
> shaharf> If not, how complex it should be to implement it?
>
> I think you could do it quite easily. ipoib_add_port() encapsulates
> everything about creating a device, so you just have to call it more
> than once. The hardest part is probably figuring out the rig
Hi,
Is
it possible somehow to create multiple IPoIB devices over a port with the same
pkey but with different QP?
If
not, how complex it should be to implement it?
Thanks,
Shahar
___
openib-general mailing l
>
> > It doesn't seems like a good idea to reinvent cluster management in
an
> > IB-specific way. I would rather see this sort of thing built on top
of
> > an existing tool like Ganglia (http://ganglia.sf.net).
>
> While I don't agree taht ganglia is the right thing, I agree
completely
> with th
Hi all,
I saw that a lot of ping-pong flew around during (my) weekend
;-). I guess that I have few things to explain regarding the ibping.
First, the origin of the utility is in pathforward SOW. As I understand
it, the ibping (vping in the SOW) should imitate the ICMP ping and
should be u
Hi, Does openSM supports non-fat tree (irregular such
as graph) topologies? Abhijeet
[shaharf]
Yes. OpenSM supports any type of mesh (connected graph).
___
openib-general mailing list
openib-general@openib.org
http
>
> The advantage of ATS is that it "just works" whether wired point to
> point, or via a switch, or whatever. It requires no central
> administration,
> works as transparently as ARP and ND, and supports IP addressing so
> applications don't have any ambiguity in how they resolve names.
>
> If
>
> Roland> First, let's understand the problem we're trying to solve.
> Roland> Who are the consumers of this address translation service?
>
> shaharf> Any ULPs at user & kernel, and also some
> shaharf> applications.
>
> I
> This API seems overly complex and at the same time too inflexible to
> me. However, rather than getting bogged down nitpicking about APIs, I
> think we have to take a few steps back.
>
I am very open to suggestions.
> First, let's understand the problem we're trying to solve. Who are
> the c
The
following email describes few new function calls for OpenIB access layer that
enable better/simpler use of the InfiniBand fabric resources, called ib_at (IB
address translation) See the attached preliminary ib_at h file.
There
are 3 main functions and few additional helper function
Title: RE: [openib-general] [Utils] IB Management Simulator
Eventually is should be
part of the management tree (gen2/trunk/src/usermode/management), or even under
the osm tree (gen2/trunk/src/usermode/management/osm) due that fact that it requires
the opensm vendor layer to be able to func
Hi,
> > There is a sys fail red light on the CPU on the 96-port switch that
the
> > opensm host attaches to.
> >
> > What's weird is none of the ib admin tools found anything.
ibnetdiscover
> > happily walked the whole subnet. The only problem was that opensm
would
> > not run, but the errors were
Hi
all,
I
added two new utilities to the management tree (gen2/userspace/management):
ibaddr:
shows the lid range and the default GID of the target, the default is the local
port.
perfquery:
dump (and optionally clears) the performance counter
Hi
all,
A
new utility is added to the management tree (gen2/userspace/management/diags/net/ibtracert).
Tracert shows the path between two lids. It can show also multicast path, once
the mild is given (see the usage text).
As
a general note, please report any
New
utility named ibroute is added. This utility shows the linear forwarding
configuration and multicast forwarding configuration of a switch.
Any
comments and suggestions are welcomed.
Shahar
___
openib-general mailing list
openib-g
-Original Message-
From: shaharf
Sent: Tuesday, January 25, 2005 8:04 PM
To: 'Tom Duffy'
Cc: 'openib-general@openib.org'
Subject: RE: [openib-commits] r1642 -
ingen2/trunk/src/userspace/management: . diags/host/ibstat
diags/netdiags/net/ibnetdiscover diags/
>
work/gen2/trunk/src/userspace/management/diags/net/sminfo/sminfo.c:146:
> undefined reference to `ib_resolve_portid_str'
> sminfo.o(.text+0x3b2):/build1/tduffy/openib-
>
work/gen2/trunk/src/userspace/management/diags/net/sminfo/sminfo.c:149:
> undefined reference to `ib_resolve_smlid'
> collect2:
HI
all,
In
order to be able to support several umad layers (i.e. non gen2) and to cleanup
things, I separated any vendor specific stuff from the opensm directory and
moved it to a new library – lib vendor (usermode/management/osm/libvendor). The
includes files in the opensm dir
Roland,
Can you please run ibnetdiscover on one of the openib gen2 hosts
and send me the output file?
Please verify that there are no errors during ibnetdiscover run.
Without errors I believe that ibnetdiscover will scan the net withing
less then 30 sec (maybe even 10 sec). If there
Hi Sean,
> Hal Rosenstock wrote:
>
> >>>[SHAHAR] Attributes filtering can be done with O(1). First you
don't
> >>>have to support the entire theoretical range. Currently less then
> >>>256 attributes are used per class.
> >>
> >>You can't make this assumption for vendor defined classes.
> >
> > A
Hal and Sean,
sorry for now responding earlier, but I was sick for few days.
On Tue, 2005-01-11 at 12:15, Sean Hefty wrote:
> An issue with attribute filtering is that is results in an
O(lg n)
> algorithm being used to route MADs to the correct cl
From: Hal Rosenstock [mailto:[EMAIL PROTECTED]
Sent: Mon 1/10/2005 11:22 PM
To: shaharf
Cc: Michael S. Tsirkin; Roland Dreier; openib-general@openib.org
Subject: RE: [openib-general] Some Missing Features from mthca/user MADaccess
On Mon, 2005-01-10 at 11:30, shaharf wrote:
> Thinking about it, I t
Original Message-
> From: Michael S. Tsirkin [mailto:[EMAIL PROTECTED]
> Sent: Monday, January 10, 2005 7:05 PM
> To: shaharf
> Cc: Eitan Zahavi; Roland Dreier; openib-general@openib.org
> Subject: Re: [openib-general] Some Missing Features from mthca/user
MAD
> access
>
> Hel
: Monday, January 10, 2005
6:43 PM
To: shaharf; Eitan Zahavi; Michael
S. Tsirkin; Roland Dreier
Cc: openib-general@openib.org
Subject: RE: [openib-general] Some
Missing Features from mthca/user MAD access
Shahar> Yes, but when you want to respond to
attributes you have to specify a mask. Us
Title: RE: [openib-general] Some Missing Features from mthca/user MAD access
Shahar,
>
Thinking about it, I think there is another alternative which maybe
> cleaner: automatically raise
the issm bit if someone registers to answer
> sminfo attr. The cleanup will
be of course when it wi
> > Personally I think that non binary format is better when applicable
-
> > this mean issm file (you can say it is Plan9 style...). It would not
> > necessarily use the open method. You may want to use the write
method
> > and to enable everyone read the file. This read may return the pid
of
> >
> > After consideration, I think the proper way is add a reference count
> > and clean is_sm when it falls to 0.
> > This way runnning two opensms on the same HCA and two different
> > HCAs in the same subnet behaves the same.
>
> I don't understand this. IsSM is per-port, so what does it mean fo
Title: RE: [openib-general] OpenSM died a horrible death
Hi Tom,
Are
you able to reproduce this problem? If you are I would like you to reproduce it
will full verbosity (-V). If you can’t or the scenario is not consistent,
please tell me too. It might direct us to some other direction
> Yes, I guess it's not so bad to add port_phys_state to struct
> ib_port_attr.
> But I'm still wondering whether it's useful at all.
>
> - R.
You are correct that this is not critical. The Logical state is enough
for most cases. It is just cleaner for the umad library to check the
Phys state
> Hello, Roland!
> Quoting r. Roland Dreier ([EMAIL PROTECTED]) "Re: [openib-general]
> Some Missing Features from mthca/user MAD access":
> > > 1. The SM needs the ability to write (set/clear) the 'is_sm'
PortInfo
> > > capability mask bit (from user space).
> >
> > Yes, I'll implement this.
>
>
Title: RE: [openib-general] OpenSM died a horrible death
Hi Tom,
Can you send me the
original mail concerning the SM horrible death? It was corrupted in our
Exchange or it was very large (over 5 MB). If you want to send very large log
files, please send it tared and zipped. If it is only
ibnetdiscover is the
most recent topology (right to the few last milliseconds). It doesn’t
depend on nothing.
Shahar
From: Eitan Zahavi
[mailto:[EMAIL PROTECTED]
Sent: Tuesday, January 04, 2005
7:44 PM
To: shaharf;
openib-general@openib.org
Subject: RE: [openib-general] New
utilities
Hi
guys,
I
started to add new mad utilities:
Sminfo
: issue a sminfo and dump the reply. Of course you
must direct it to the sm lid/path in order to work, but it can query also
standby SMs.
Ibnetdiscover:
do IB net discover and print out the
> shaharf wrote:
> > I added very simple retries mechanism to the SM. It is
not
> > really checked, because I don't have a controlled way to loose
packets.
>
> You can simulate this by modifying validate_mad in mad.c to return
> occasionally. E.g. declare
>
> I assume libmthca was moved to userspace/management by mistake. I
> moved it back to the top level userspace directory.
>
> Thanks,
> Roland
Oops, sorry.
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/listi
> > > > Oh. Is that a udev problem then?
> > >
> > > Yes. I Forgot to modify umad.c for the new udev scheme. I will do
it
> > > soon.
> > >
> > > Shahar
> >
> > Why cant I give the udev device path on a command line to opensm?
> > that would solve this once and for all, and let the user play with
>
>
> Hello!
> Quoting r. Michael S. Tsirkin ([EMAIL PROTECTED]) "[openib-general]
Re:
> opensm failure":
> > warn: [32666] umad_open_port: open
/dev/infiniband/mthca0/ports/1/mad
> failed
>
> Oh. Is that a udev problem then?
Yes. I Forgot to modify umad.c for the new udev scheme. I will do it
soo
I
moved all osm and umad related directories to usermode/management.
I
also fixed some minor bugs in the umad libraries:
-
Minor make files error fix (missing .depend warning)
-
force terminating null at sys_read_string
-
fix sys_read_uint retur
> Hi!
>
>
> # ./osm/opensm/opensm -V
-
>
> OpenSM Rev:1.6.0-rc2
> Command Line Arguments:
> Big V selected
> Log File: /tmp/osm.log
> -
> using default guid 0xc900012a3981
>
> Error from osm_opensm
> Michael> Could you please remove openib/src/userspace/osm.old as a
> Michael> first step? Its really painful to get two copies of
> Michael> opensm each time. If you thjink you need it for
> Michael> reference, you could always just move it somewhere ...
>
> I agree osm.old sho
Hi
all,
I
added very simple retries mechanism to the SM. It is not really checked,
because I don’t have a controlled way to loose packets.
It
is committed. Please let me know if it you encounter related problems (or
non-related problems ;-)
Shahar
> -Original Message-
> From: Roland Dreier [mailto:[EMAIL PROTECTED]
> Sent: Monday, December 27, 2004 6:37 PM
> To: shaharf
> Cc: openib-general@openib.org
> Subject: Re: [openib-general] Problem with userspace tree structure
>
> We already have osm, diags, a
>
> I think gnome is a pretty good parallel, because the contents of
> userspace/ are really separate projects, with different sets of
> developers and different maintainers. Having one main makefile is
> confusing for users and packagers, and the current scheme is also
> error prone, since it's
Tom,
I debugged the opensm a little, and it seems like a simple
opensm bug. Somehow the destination lid of the report target was not
found, and as there were no checks for that, the opensm faulted. I added
the missing checks. I don't know how this happened - i.e. why was the
dest port inval
Hi Tom,
It seems that one of nodes in your subnet registered itself to be informed on
traps. This is one of the rare cases that require GRH headers which are not
really supported yet.
I didn't think GRH has a high priority, but as I see that it is really used, I
will have to deal with that s
Thanks. You are probably right about the trap. I wonder why it has different
scheme in the spec.
Did this work for anybody else then me?
Shahar
From: Roland Dreier [mailto:[EMAIL PROTECTED]
Sent: Fri 12/24/2004 12:22 AM
To: shaharf
Cc: Michael S. Tsirkin
From: Tom Duffy [mailto:[EMAIL PROTECTED]
Sent: Thu 12/23/2004 11:10 PM
To: shaharf
Cc: openib-general@openib.org
Subject: Re: [openib-general] osm
On Sun, 2004-12-19 at 20:26 +0200, shaharf wrote:
> Hopefully everything will compile and install
Ok, I will send the relevant function from user_mad file. Please forgive me not
to send you a proper patch. I am already at home and I don't want to mess up my
Wife's computer ;-)
Aha - BTW, what I really meant is that OpenSM can bring up IPoIB (small mistake
for me, big progress for the openi
>Hello!
>Quoting r. shaharf ([EMAIL PROTECTED]) "RE: [openib-general] Problem with
>userspace tree structure":
>> BTW, the reason I didn't include tvflash in my make file, because it doesn't
>> build at all on my SUSE9.1 (option). If it is only a loc
OK, I have good news and bad news:
The good news are that the committed OpenSM can bring up OpenIB (at least I
could do it once or twice... ;-).
It is also merged with Mellanox 1.6.1 OpenSM.
The bad news are that I am pretty sure that the committed openib stack will not
work - there is a bug i
>By the way, I don't think we have to mandate autotools for every
>package under the userspace/ directory (although using autotools is
>the friendliest for users, packagers and developers).
>
>My real objection was just to the recent checkin of Makefiles
>etc. directly under the userspace/ direc
>
> One more suggestion: consider using libsysfs (which every modern
> distro will ship) instead reimplementing it in libcommon/sysfs.c.
>
> - R.
Thanks - I will do that. I was not aware to this library.
Shahar
___
openib-general mailing list
[EMAIL
> shaharf> Hi all, I took the liberty to commit the user-mode tree
> shaharf> even though it is far to be complete or stable. However,
> shaharf> the tree contains several applications and utilities that
> shaharf> may help all of us, even at this prelim
Hi
all,
I
took the liberty to commit the user-mode tree even though it is far to be
complete or stable. However, the tree contains several applications and
utilities that may help all of us, even at this preliminary stage.
I
renamed the 1362 osm tree to be os
> > Are you saying to have other forms of MAD like cancel by mgmt_class,
> cancel
> > by SGID and/or SLID in addition to cancel by TID ? We could do this
if
> there
> > is a need. I don't think that need has been demonstrated (at least
yet).
>
> No - I'm not suggesting that we have more ways to ca
>
> Roland> I would create another device special file like the umad
> Roland> file and use reads to get the events.
>
> I thought about this a little more, and I'm not sure this is the right
> approach any more. I would look at using the kobject_uevent mechanism
> to pass these events
> I guess it would be another ioctl. However I'm not sure how useful
> this is for userspace... in the kernel modules like IPoIB want to
> cancel pending sends to avoid a callback when the module is unloaded.
> However in userspace, an application can just close the file on exit.
>
> - R.
Open
> -Original Message-
> From: [EMAIL PROTECTED] [mailto:openib-general-
> [EMAIL PROTECTED] On Behalf Of Michael S. Tsirkin
> Sent: Thursday, December 09, 2004 8:02 PM
> Cc: [EMAIL PROTECTED]
> Subject: Re: [openib-general] Should I use umad -or- osm
>
> > > > I guess that in this stage onl
> -Original Message-
> From: [EMAIL PROTECTED] [mailto:openib-general-
> [EMAIL PROTECTED] On Behalf Of Michael S. Tsirkin
> Sent: Thursday, December 09, 2004 7:07 PM
> To: [EMAIL PROTECTED]
> Subject: Re: [openib-general] Should I use umad -or- osm
> >
> > I guess that in this stage only
> -Original Message-
> From: [EMAIL PROTECTED] [mailto:openib-general-
> [EMAIL PROTECTED] On Behalf Of Michael S. Tsirkin
> Sent: Thursday, December 09, 2004 7:10 PM
> Cc: [EMAIL PROTECTED]
> Subject: Re: [openib-general] Missining Infiniband class fields
> > Please note that you want to
> -Original Message-
> From: Michael S. Tsirkin [mailto:[EMAIL PROTECTED]
> Sent: Thursday, December 09, 2004 7:04 PM
> To: shaharf
> Cc: [EMAIL PROTECTED]
> Subject: Re: [openib-general] Missining Infiniband class fields
>
> Hello!
> Quoting r. shaharf (
> -Original Message-
> From: Michael S. Tsirkin [mailto:[EMAIL PROTECTED]
> Sent: Thursday, December 09, 2004 6:53 PM
> To: Philip Mucci
> Cc: shaharf; [EMAIL PROTECTED]
> Subject: Re: [openib-general] Should I use umad -or- osm
>
> Its clearly not a great ide
From: Philip Mucci [mailto:[EMAIL PROTECTED]
> Sent: Thursday, December 09, 2004 6:44 PM
> To: shaharf
> Cc: [EMAIL PROTECTED]
> Subject: RE: [openib-general] Should I use umad -or- osm
>
> Hi Shahar,
>
> Thanks for the info.
>
> My needs are very 'simple'
Hi Philip,
I am currently implementing umad access library. It much simpler
then the osm_vendor api. I would recommend using umad library and not
osm. As a matter of fact the current osm vendor layer does not support
openib gen2. I am working on that either. Both the umad access library
an
Hi
all,
Another
issue I want to raise concerning the umad access library is some change
indicators for the infiniband class objects – ca & ports. It would be
very nice if we could have a modification version file for each just object. This
will help the safe atomic reading a
Hi
guys,
I
am implementing a user mode access library for the umad. I have encountered few
issues:
1.
I am using the sys Infiniband class to get the HCA/port attributes, and several
fields are missing:
For
the ports:
Physical
port state
>
> The intention is that userspace can read the kernel's ABI version from
> /sys/class/infiniband_mad/abi_version and compare it to the value of
> IB_USER_MAD_ABI_VERSION that userspace was compiled with. We just
> need to be careful to increment the ABI version if we make any
> incompatible ch
>
> It would make sense for OpenIB to ship a package like "libibmad" that
> has all the headers required for using the ib_umad module. Userspace
> and the kernel need to agree on the ABI, obviously, but physicaly
> sharing the same .h file ends up creating more problems than it
solves.
>
> - Ro
>
> I believe the current feeling in the kernel community is that kernel
> headers should be kernel only and if userspace needs a header file,
> there should be a separate userspace version of the file.
>
> - Roland
OK, I accept that for most IB types and structures, but what about
ib_user_mad
>
> Sam> After giving it a second thought my vote goes for:
> Sam> include/linux/infiniband
>
> Could you share the reasoning that led to that preference?
>
> Unfortunately we don't seem to be converging on one choice of
location.
>
> On one side there is the fact that the .h files are
Hi
all,
I
know I am new to this project and I must be naïve but I want to understand few
things concerning the openib architecture. In the course of learning the openib
gen 2 stack and preparing to port the opensm to it (which is my current task),
I have encountered few area
Just a small correction: Voltaire's IPoIB should work with minism as
long it is configured to do it (using the ib-setup option 3,2). As a
matter of fact it should be treated as a hack, but it should be enough
for back to back configuration.
OpenSM should work with a Voltaire Host, but I am not sur
76 matches
Mail list logo