[GE users] LSF vs SGE
andy.schwierskott at sun.com
Tue Apr 12 15:16:24 BST 2005
> You can always download SGE binaries and try to play with it. You don't
> need to be root to install SGE (use TCP ports over 1024), and you don't
> need a dedicated machine.
> (For less common OS/platforms, you do need to compile from source,
> SGE and LSF are very similar, except:
> - LSF 5.0 or above has a scheduler API, and SGE is in the process of
> it (AFAIK, the code is ready, and will be checked into cvs after
> is released)
indeed Shannon wrote code which provides an interface to influence the
scheduler. After the 6.0u4 release this will be checked in but we don't yet
make it part of the standard compile (and thus is not immediately part of
Sun's N1GE or the open source courtesy binaries).
I assume we'll get more background information from Shannon and some kind of
documentation or use scenarios about this API once it is available.
We (i.e. the community) need to talk and decide later if and how to make
this a standard part of the standard build in the future. I think first the
most important is that we will have a good opportunity to test and disuss
the functionality of Shannon's extensions to the scheduler interfaces.
> - only the commercial version of SGE offers the accounting module. For
> if you are willing to pay, there are similar products.
> - only the commercial version of SGE offers the Windows port
> - SGE has more DIY stuff that you can (or you needed to) play with, like
> AFS/Krb. You can even take the SGE source and compile under Cygwin to get
> free version of SGE under Windows.
As I understood the discussions on the mailing list the Cygwin compile just
ports the command line utilities but not the execution host functionality.
Anyone who is interested on doing more on the Windows port (be it using
Cygwin or Interix) is welcome to study the code for the #ifdef INTERIX
occurences and reengineer the code and contribute it to the open source:-)
We as Sun are fully committed (and I think we've quite well proven it over
the past four years) to develop, contribute and sustain the "core" system
functionality (qmaster/scheduler, Unix exec hosts, qmon) and share it woth
tthe open source community users. "Really" new stuff (like ARCo or Windows
support) is aprt of the Sun product.
Full interoperability between the OS binaries and Sun's N1GE binaries is
guaranteed. Since as Sun we have to follow some rules when adding major
features to the software and we on the other side want to keep the full
compatibility between SGE and N1GE we may have some hard findings about the
prper integration of Shannon's exensions - but let's have this discussion
> In terms of scaling, both SGE and LSF can handle clusters with over
> thousands of nodes (Linux has a system bug, so you need to use SGE 6.0u4 or
> above to get a workaround). In terms of scheduling, there is 1 thing that I
> think can be improved in SGE (issue 35, 1245, and friends) - there are
> workarounds, but if we can get it fixed in SGE 6.1, it would be nice. In
> terms of security, SGE offers more features, like SSL/SSH,
> submit/admin/exec only node. And also, SGE supports the BSDs. On the other
> hand, LSF offers SLA in 6.x - I don't think I need it, but I don't know
> your setup/requirements so I can't tell.
> Also, in terms of SGE support, this list is good enough for many people.
> For more advanced support, I believe you can find from:
> But if the cluster is mission critical, then I would choose to get it from
> Sun or from Raytheon (ie. Shannon), as they understand the code very well.
> For LSF, commercial support is not an issue (as long as you are willing to
> pay, but SGE commercial support is non-free as well!).
>> thanx all for your helpful responses. Looks like I have a lot of
>> research to do.
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net
More information about the gridengine-users