Re: [Wg-test-framework] Machine suggestions from AMD

[Aravind Gopalakrishnan <aravind.gopalakrishnan@xxxxxxx>]

On 9/22/2014 7:57 AM, Ian Jackson wrote:
> Here is a detailed proposal for selecting particular machines from
> Paul's suggestions, in the form of a spreadsheet.
>
> (You can find everything here too:
>    http://xenbits.xen.org/gitweb/?p=people/iwj/colo-for-testing.git;a=summary
>    git://xenbits.xen.org/people/iwj/colo-for-testing.git)
>
> Summary:
>
> Total budget $57,076 for 26 machines, 2 to become the main servers and
> 24 to become test boxes.  (Switches, etc., not included.)  Of the test
> boxes: 42% are AMD (54% by cost).
>
>
> I have added a pair of higher-spec versions of the Dell R415.  The PDF
> from the dell.com website is attached.
>
> There are two very expensive AMD test boxes at the top of the list.
> This is to try to balance the quantity and spend between Intel and
> AMD.  Conversely there are two very cheap AMD machines at the bottom
> of the list.  (It appears that there is a greater variety of Intel
> machines available.)
I could see a mix of different micro-architectures and socket config for 
the AMD systems on the spreadsheet.
(In fact, I think the second R515 (the cheaper one) is a K10 micro 
architecture which is a little old. But can be
used for regression tests, etc. I guess)

So, as far as variety, we should be fine.

> This amount is bigger than was approved by the Xen Project Advisory
> Board but I think we can just go back to them to approve it and this
> won't be a problem (and can be done quickly).  We should seek
> re-approval for the bigger amount plus (as before) some headroom.
>
> Alternatively we can go ahead without the two big AMD machines; that
> would be $42,342, well within the budget (and would have 36% AMD test
> boxes by number or 34% by value).  If it is convenient we could start
> work now and get approval for the extra budget (which I am confident
> of getting) in parallel.
Instead of the IBM machine, why not swap that with a HP 385p server? 
(attaching a model I found on the website).
Looking at IBM x3755's spec, I see it has a 6300 series opteron which is 
a Piledriver micro architecture.
The HP model I have attached is the same; and is also a 2U (a space 
requirement that we have already factored in).
It's different from the other AMD ones in that it's a 2P box. (i.e) a 
MCM part. Afaict, others are 1P boxes.
Aside from that, it's also less costly compared to the IBM one.
But, it still keeps the ratio of AMD boxes (by  number) intact and the 
cost balancing identical still-
Costs after swapping the HP 385p for the IBM x3755:
Total cost: $50140
Cost of AMD systems: $18956
Ratio of AMD test boxes (by value) : ~47%


Thoughts?

Thanks,
-Aravind.

> I'm assuming that one pair of 2U test boxes will fit in the rack.
> That will give us 26 machines of which 4 in total (the two main
> servers and the two big AMD test boxes) are 2U, for a total of 30U.
> (Of course we need to fit some cable management, switches,
> etc. etc. in as well.)

Attachment: HP_server_DL385p.pdf
Description: Adobe PDF document

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