Kevan Miller wrote:
On May 23, 2008, at 7:54 PM, Justin Erenkrantz wrote:
On Fri, May 23, 2008 at 11:08 AM, Joe Bohn <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
The Geronimo PMC would like to make the following request of the ASF
Infrastructure team. The request is specifically for machines to be used
for collaboration and support of our certification efforts. This is an
extremely extensive set of tests that require significant computing
resources, often not available to most community members.
Thank you for your consideration of this request. We will gladly
answer any
questions or provide additional information.
Rationale:
The Apache Geronimo community has a need to support the execution and
sharing of results from Sun certification tests (cts) which are necessary
gain JavaEE5 certification compliance. This information is only
available
to those Geronimo committers that have signed the Sun NDA and other
Apache
committers that have signed an NDA and gained approval of the Apache
Geronimo PMC. This has allowed other Apache projects to test new
products/releases by running JavaEE 5 TCK tests using the Apache Geronimo
test infrastructure.
In the past these resource intensive tests have been run on private
machines
by individuals. As more people become involved with Apache Geronimo and
related projects, it is becoming obvious that we need a central system to
run and share the results of these tests. A centralized testing
environment
allows the Geronimo community to more fully participate in the TCK
process.
Some committers don't have access to the hardware resources needed to run
the Java EE TCK tests in a timely manner. Although some ad-hoc sharing of
private machines has occurred, this is not ideal from a community
perspective. Community controlled systems allow us to equitably share
these
resources.
Request:
To fulfill this requirement, the Apache Geronimo PMC is requesting
the ASF
infrastructure team to provide and host machines that can be used for
this
purpose. Initially, we would like to request two (2) machines that
meet (as
closely as possible) the specification below. However, we can see
the need
for 2 additional machines in the not too distant future.
To fill in others, Joe B. and I have talked about this before - so
this doesn't come as a surprise to me. I've informed Joe that our
current data center does not have space for any new machines, so this
will require a build-out of a new data center. (Joe S. has looked
into this and it's about $500/mo for the data center that he houses
his personal setup at; we also know that we'd have to build out the
new data center - which we'd have to gather the costs for as well.)
I assume building out a new data center is something we will need to do,
regardless? These machines would put us over the tipping point?
Have we determined whether the machines would be donated, or do we
need to purchase them?
To give you an idea, to meet the specs you outlined for a Dell
PowerEdge 1950 III is about $4500/machine (and that's at 2.5GHz not
3.0GHz - see below). We may be able to negotiate a lower price with
our Dell sales rep, but that's the ballpark figure per machine.
My assumption was that the machines would be purchased by the ASF.
Community members have offered to donate some existing hardware. There
was also discussion about building out the hardware from scratch.
Finally, there has also been discussion of $ donations, but I think
that's a separate matter...
IMO, the more inline we are with current Infra practices, the better off
we'll be. I don't think that the existing hardware would meet the
management requirements needed to properly host them. Building up the
machines might save money in the short-term, but we probably lose in the
long-term. We've suffered through some efforts (GBuild) with
non-standard hardware, cheap (i.e. free) hosting, etc. Too frequently
we've ended up with the equivalent of boat anchors as machines waited to
be rebooted/repaired, spotty network reliability, etc.
I agree and based on some earlier discussions I was under the impression
that Infra preferred hardware purchased directly that was more standard
and supported. As Kevan pointed out we did have several discussions on
alternatives and could resume those if it would be more feasible. Just
let us know.
Given that amount, I'd probably want to run authorization of these
machines by the Board once we settle upon a final proposal and cost.
So, it may take until the next Board meeting before I can get final
approval.
Machine specs:
- 8 core (two 3.0 GHz quad-core)
How set is the 3.0GHz specification? The cost difference between,
say, the 2.5GHz (E5420) and 3.0GHz (E5450) is at least $1000/machine
extra. IMO, I doubt it is worth the differential.
IMO, 2.5 GHz is fine.
I agree, 2.5 GHz is fine.
- 16 GB memory
- two 750GB 7200 - rpm SATA 3GB/s disks
In a RAID1 (mirrored config) or striped?
I'd assumed striped, but others may have their own opinions.
striped
- DVD R/W (20x?)
Does it really need to have DVD R/W? (Dell doesn't even sell a DVD RW
in their PowerEdge series.)
Heh. No, not necessary.
This isn't necessary. I just thought that was standard.
- rack mountable specification to work with ASF infra requirements
- LOM or other features as necessary for ASF infra support
- to be developer managed and maintained by the Apache Geronimo Team
- Apache Geronimo would assume all responsibility for:
- configuration
- backup/recovery
- secure access
- Strictly limited to those Apache Geronimo committers with NDAs on
file or additional Apache committers with NDAs and approved by
Geronimo PMC.
- full, admin access would be granted to ASF infra with reboot directions
- At least 2 active Apache Geronimo committers (with NDA authorization)
would identified to manage the machines.
- Running Linux with something like Xen for 4 VM images per machine.
We may
increase the number of VM images if it is feasible.
Is Ubuntu 8.04 sufficient, or does something special need to happen
for Xen? (I haven't looked into Xen lately.)
I assume Ubuntu is the preferred Linux distribution for Infra? I don't
have any experience with it. I see that they choose KVM as their
virtualization technology. From a quick look, seems that Xen support is
not too great. My guess is that we can make Ubuntu/KVM work, but we
should do some investigating. I think G community members are more
familiar with Suse, Red Hat, and Xen.
- We would require both ssh and VNC access to the VMs.
I don't know if or how Xen supports VNC, but I guess so... =)
- It is not yet decided if we would use NAT to access the VMs or
public IP
addresses. Is there are recommendation from ASF Infra?
It depends how secure the Xen instances are.
If we need to have a partitioned network, we may need to get an extra
switch and stuff.
- Automation will most likely be added to run builds, execute tests, and
produce reports.
- Capability to manually run tests on demand would also be supported.
Thanks. -- justin
