Re: [Wg-test-framework] Test-as-a-service deployment costing

[Lars Kurth <lars.kurth@xxxxxxxxxx>]

Konrad,

I agree this can be done more cheaply: Chris did this costing on *very* short 
notice and used our usual suppliers (list prices) to turn this around in less 
than a day. So the costs are *very* conservative, which IMHO is reasonable for 
a budget. My thinking was that if we can get the Hardware cheaper, we can just 
get more boxes or we have more cash in the bank.

On A+B
> I am not sure what that is. Is this an upfront cost to setup the network, 
> switches, terminal servers, remote power on/off setups, DHCP, etc? I presume 
> so.
The exact cost breakdown is in the budget spreadsheet. Here is the detail for 
those two items:

A) Core infrastructure : A pool of servers and shared storage to run the 
various VMs you need to run the network, monitor infrastructure, pxe boot, 
dhcp, scheduling etc. Osstest would presumably need something similar.

B) XenRT âsiteâ infrastructure : XenRT can be built up into multiple âsitesâ, 
each consisting of a number of servers (we have sites of 16, 32 and 64 
servers). Each site needs its own small pool of servers with shared storage and 
switches to run the site. For a small XenRT deployment (one site) we could 
probably share the core infrastructure for the site infrastructure, thereby 
removing (or at least postponing) this cost.

On C
>> An admin / developer to look after and extend the deployment, possibly 
>> part-time
> Is this also considered the person who would triage issues with XenRT or 
> osstest?
No. We only covered HW maintenance (1-2 days a month) and remote management 
(for a much larger deployment than what we are talking about this takes up 20% 
of a person).

The key challenges are:
* We can't employ anybody via the Linux Foundation (they already made clear 
that this is too complex and unusual for what they do today)
* This leaves us with : vendors donating time (which in practice never works, 
unless a single vendor takes full responsibility). 
* Employing somebody on behalf of the project within a vendor : this may be 
politically, legally and logistically difficult and will need to be discussed 
by the board
* And we don't need a full-time person, unless we include triaging, etc.  

So this, needs to be discussed at the Advisory Board meeting.
As an aside: I believe we would end up paying a lot more if we used the Linux 
Foundation (I was told how much OpenDayLight pays, which is very high)

> Meaning they would look at the test results and figure out where the bugs are 
> happening?
The question there is whether we can find somebody with the right skill set AND 
whether we have enough funds. This should be discussed in the Advisory Board.

> These would be one-time costs right? And I presume they could also be 
> eliminated if vendors (AMD, Intel, Oracle, ARM) provided the hardware?
Agreed. 
Although if a vendor donated HW, there will logistical issues, etc. The other 
option is to get discounts.

Regards
Lars

-----Original Message-----
From: Konrad Rzeszutek Wilk [mailto:konrad.wilk@xxxxxxxxxx] 
Sent: 17 December 2013 14:20
To: Lars Kurth
Cc: wg-test-framework@xxxxxxxxxxxxxxxxxxxx
Subject: Re: [Wg-test-framework] Test-as-a-service deployment costing

On Thu, Dec 12, 2013 at 02:15:43PM +0000, Lars Kurth wrote:
> 
> 
> From: Chris Shepherd
> Sent: 12 December 2013 13:09
> To: Lars Kurth
> Subject: External test-as-a-service deployment costing
> 
> Hi Lars
> 
> The costs are in two parts: colo and test-as-a-service deployment.
> 
> TOTALS:
> $100k-$150k capex for the test machines and infrastructure $2.5k-$3.5k 
> per month costs for colo service Additional to pay for an admin
> 
> Note: if colo costs are unacceptable then could see if a Xen Project member 
> would âdonateâ space, power and cooling etc.
> 
> See below for breakdownâ
> 
> Colo costs:
> 
> Â         $2.5k per rack per month (inclusive of space, power and cooling)

> 
> Â         $1k per month for 20Mbps internet with high SLA (could reduce this 
> cost very significantly in return for lower SLA)

I don't think we need the top tier SLA.

> 
> Â         An admin / developer to look after and extend the deployment, 
> possibly part-time

Is this also considered the person who would triage issues with XenRT or 
osstest?
Meaning they would look at the test results and figure out where the bugs are 
happening?

> 
> Test-as-a-service deployment costs:
> A.  Core infrastructure = $24k
> B.   XenRT âsiteâ infrastructure = $12k

I am not sure what that is. Is this an upfront cost to setup the network, 
switches, terminal servers, remote power on/off setups, DHCP, etc? I presume so.

> C.   Mid-range test servers = $4k each (could be lower if happy to go for 
> lower spec machines)
> Two options would be:

These would be one-time costs right? And I presume they could also be 
eliminated if vendors (AMD, Intel, Oracle, ARM) provided the hardware?

> Â      Single 16-host site = A + B + 16xC = $100k
> Â      Single 32-host site = A + B + 32xC = $164k


With these numbers I am tempted to advertise my basement as a perfect place for 
it :-) Much cheaper costs - though there would be an upfront cost to get a new 
100AMP circuit installed.

Free cooling thanks to New England weather.

(and yes, my basement already has two racks of machines that do testing every 
night).


> 
> 

> _______________________________________________
> Wg-test-framework mailing list
> Wg-test-framework@xxxxxxxxxxxxxxxxxxxx
> http://lists.xenproject.org/cgi-bin/mailman/listinfo/wg-test-framework

_______________________________________________
Wg-test-framework mailing list
Wg-test-framework@xxxxxxxxxxxxxxxxxxxx
http://lists.xenproject.org/cgi-bin/mailman/listinfo/wg-test-framework