I have been reading this thread and post some of my observations and
comments on the matters.
Much, if not all, experimental/hardware science and engineering systems
applications (e.g., special drivers, modified interfaces and bus
control, etc.) are targeted towards EL or the SuSE equivalent. There
will not be a CentOS 9, although there had better be a RHEL9 unless IBM
decides upon yet another path (because of for-profit business
parameters, as well explained by another correspondent to this list who
used very candid and honest language).
For example, at one time we put SuSE SLES (SuSE EL equivalent, just as
Ubuntu LTS is aimed at the same sector) on a HPC compute engine that had
a SAN and other infrastructure as well. SuSE was dismal because we
bought the least expensive support level (not cradle-to-grave full
handholding). We then switched to SL for RHEL, and at least one of the
for-profit HPC vendors we used -- that was a subcontractor to US
Government primary contractors -- also used SL. SL worked, and the
hardware support software we needed (including modified drivers, etc.)
was available, tested, and installable -- from this vendor. No games,
no special support contracts, we did all of the software maintenance and
verification such that things worked on our systems.
Again, as with SuSE, SL, and Ubuntu LTS, this was professionally ported
and, in terms of bug fixes and sample configuration, etc., files,
professionally supported by the vendor -- not amateur volunteers.
Cloud Linux looks -- again, looks -- as though it will be a SL type
support, and presumably there there will be CL9 if RH is required to
release the full source of RHEL9. CL 8 (and upon RHEL9, CL9) should
(should -- not will) be compatible with all of the
scientific/engineering hardware specific ("experiment-use") drivers,
etc., to which correspondents have referred.
Springdale (Princeton EL) is a possibility, but it is unclear how long
Springdale will have real professional support depending upon how
Princeton internally allocates fiscal (and hence personnel) resources,
and how much Springdale, like Fermilab and thus SL, depends upon
discretionary support from the USA Federal government.
I have signed up for the Rocky EL list to keep abreast of the situation,
but it appears Rocky EL will not have the paid professional support of
SL (not cradle to grave hand holding) but rather volunteer developers,
porters, testers, etc.
CentOS Stream, as with Fedora, and Ubuntu non-LTS is completely
unsuitable for my uses. These can be not only "unstable", but have too
many software defects (bugs) to of safe use. Moreover, continual
professional support is necessary to mitigate threats (the various
compromises that are in the wild) -- keep things up to date against the
latest attacks.
There has been a discussion of the shorter life-cycle of Ubuntu LTS
versus EL. In the case of CentOS 8, the life-cycle is much shorter than
what seemed to be promised. However, the issue arises as to against
which system (and application) libraries (typically, .so) a scientific
application is built, and against which Linux kernel (and kernel
"features"). EL 7 (SL, CentOS, etc.) may have a longer supported
lifecycle, but if a vendor does not back-port a vital application
(including a driver) to a "supported" environment that has been
"obsolete" for many years, than the support will be meaningless. I
recall once that we were required by a vendor of scientific equipment to
support versions of the kernel and libraries that were not available for
the "stable supported" Linux release we were using, and we were *FORCED*
to migrate. Thus, the shorter time interval of LTS versus EL supported
environments may not always be of real significance. However, the fact
that some applications (including drivers) may not be available for
Ubuntu (and in .deb from) but only for some variant of EL (in .rpm form)
is of greater concern. Cloud Linux may be the answer here -- if the
vendor really meets their promised time line.
Again, my own needs are such that it is unacceptable to have a volunteer
(and in many cases, amateur) developer/support arrangement for "mission
critical" systems and applications software.
Yasha Karant
On 12/10/20 8:47 AM, Vinícius Ferrão wrote:
I’ve done this mistake in the past.
The major issue with Debian is its lifecycle, even LTS is 5 years only. Same
for Ubuntu. It’s just too little. If you need to install it near the end of the
2yr lifecycle you’ll get effectively something like 3yrs of support.
The other issue is that the vast majority of academic and scientific software
is targeted for Enterprise Linux. As an HPC engineer we always needs to use
RHEL/derivatives or SLES/Leap. OpenHPC is only available to those flavors.
Mellanox OFED? Ok there’s Ubuntu support nowadays, but the default branches are
still for EL/SLE.
That’s how things work in our environment. I think the vast majority of people
here works on Academia or with science/research/etc.
And finally I don’t want to adapt everything to Debian. The FHS is different,
scripts will break, etc.
Best regards,
Vinícius Ferrão
Sent from my iPhone
On 10 Dec 2020, at 13:38, Maarten
<000011ce72e232d2-dmarc-requ...@listserv.fnal.gov> wrote:
I might also consider switching to Debian since it will be hard to tell if any
other still existing rhel clones will continue and Debian has been around for
quite some time.
On 12/10/20 8:34 AM, Maarten wrote:
I will probably be more like to go for Springdale Linux since they've been
around since before CentOS, I find it hard to put trust in a project that's
just getting started unless of course CERN changes their decision about
discontinuing Scientific Linux since they were migrating to CentOS.
On 12/10/20 5:17 AM, ~Stack~ wrote:
On 12/9/20 9:16 PM, Yasha Karant wrote:
One thing does concern me: having left CentOS (it was all "volunteer" effort
at that epoch as I recall) for SL, a primary motivator was that SL had professional
(employed, not volunteer) persons doing the distros, and this SL list amounting to
support.
If Rocky is to be all volunteer, how reliable and professional will it be?
This is not a minor issue, as very few enthusiasts or other non-professionals
provide a truly reliable deliverable.
I would say, give it time. It wouldn't be the first time Kurtzer started an
open source project and turned into a company. :-)
For my use, is EL going to continue to be workstation friendly (e.g., laptop in which one cannot
pick and choose to integrate only Linux traditionally supported controllers with appropriate
drivers, such as sound "cards", but is stuck with whatever the laptop vendor has used --
typically MS Win "supported") or is it primarily a server distro? Ubuntu LTS still seems
to be laptop friendly.
They are aiming for complete RHEL reproducibility. If the goal is to be
as-true-as-possible-RHEL variant then the answer would be in how you use RHEL.
But do give it sometime. It's only been two days and the announcement I just
saw said that there are now 750 people actively participating in the various
forms to communication and they have direction, a plan, and leaders making it
happen. And there's thousands of people who have noticed and are talking about
it on /. , reddit, lwn, ect. That's pretty impressive and it speaks volumes
about the number of people who really do want a true-to-RHEL variant.
~Stack~