This loads a font easier to read for people with dyslexia.
This renders the document in high contrast mode.
This renders the document as white on black
This can help those with trouble processing rapid screen movements.

RE: [pafska] RE: PAFSKA teleconf 27 April 21 UT ; additional item for discussion

From: <John.Bunton_at_email.protected>
Date: Thu, 28 Apr 2011 17:52:10 +1000

Hi Tim,

I see my top-down and your bottom-up analysis agree: computing is not a risk.
The point i am trying to address is some people, and Jaap is one, are say computing is a risk.
So we need to show that it isn't. I think we are in good shape for SKA1 or SKA2 now.

But it does depend on the science people giving us specifications with reasonable baselines. For example interpreting memo 100 can give a max baseline of 200km because there are PAFs an antennas out to 200km, but table 1 seems to specify 10km for SKA2 . So it seems it is contradictory in the one document. Does this mean for SKA1 we only have a PAF core and inner region (70% of dishes radius 2.5km, memo 130). This will reduce our cost impact somewhat. A PAF costing 30% of a dish (~100keuro) would reduce the number dishes by 18% (207 dishes built) not the 23% (192 dishes built) it would be if PAF were installed on all dishes.

cheers
john
________________________________
From: Cornwell, Tim (CASS, Marsfield)
Sent: Thursday, 28 April 2011 2:04 PM
To: Bunton, John (ICT Centre, Marsfield)
Cc: Jackson, Carole (CASS, Marsfield); pafska_at_atnf.csiro.au
Subject: Re: [pafska] RE: PAFSKA teleconf 27 April 21 UT ; additional item for discussion

Hi John,

Your analysis assumes that the computing is a significant component of the cost for SKA1. I doubt that it will be if we go by memo 100. For ASKAP, we expect to be able to do a 2km survey with 36 antennas/36 feeds on a roughly $1M (2010) machine. Scaling by number of antennas (36 to ~ 175) and baseline length (2 to 5) we get 390 times CPU = 39PF. Leaving storage aside, we would have to spend $390M in 2010. Suppose that we actually buy the machine in 2018, say, the cost is then about $13M. Scaling from the $200M expected for a 1EF machine in 2018, we get $8M.

All of this is very handwavy but it does indicate that we shouldn't be too frightened of the cost of the computing for PAFSKA1. I'm working on a memo analysing the costs working from simulations which will firm up some of these numbers. The simulations for the dishes are already feasible on the Pawsey Stage 1A system that we have access to.

One important caveat is that if memo 100 is incorrect then the costs can increase quickly as baseline increases.

Note also that although we desire accuracy in estimates of the total processing cost, the unpalatable truth is that the required processing performance for a given science observation is hard to determine since it depends upon some poorly constrained choices. For example, the processing time increases with the data volume, and the data volume goes as the inverse of the correlator integration time and the channel bandwidth. In turn these are constrained by the requirement to limit decorrelation losses. Changing both by a factor of two will change the processing by a factor of four. Similarly changing the maximum baseline processed by a factor of two will change the processing by a factor of eight. Thus without changing the science case much, one can vary the processing in this example by a factor of 24 in each direction.

Cheers,

Tim

On 28/04/2011, at 1:37 PM, <John.Bunton_at_csiro.au<mailto:John.Bunton_at_csiro.au>> wrote:

Hi,
Apologies for not being heard at the teleconference, I hit *6 once too often.
Some comments:

For monitoring the big difference for PAFs is the monitoring of calibration data (array covariance matrix and correlations with noise sources). These are calculated in the beamformer but processing of the data is in the command and control system and used to determine beam weights.

For the requirements we have science wanting more and costing pushing us to less. But we should emphasis that we have an adaptable system. We a free to scale the PAF from a ~10 beams at the high end to ~100 by choosing the size of the PAF. The science at the high and low end of our 0.6-1.4 GHz bans is different. At the high end it might be better to restrict BW to that that will give us galaxy detections and use the beamforming resources to implement more beams. At the low end fewer beams and wider bandwidth might be better.
In any case costs might cause us to go for fewer beams or less bandwidth but even so we are aiming at a survey speed that is an order of magnitude more than that of a single pixel feed.
In a sense our requirements are quite plastic at the moment.

Maximum baseline is also an important requirement. Imaging cost go as the cube of maximum baseline. If we had 27 beams then for the same image processing cost as a WBSPF we are limited to a maximum baseline that is 3 times less. Is this OK? Memo 100 specifies max baseline for WFoV Surveys as 5km for phase 1 and 10km for Full SKA (Table 1). But baselines for Phase 1 go to 50-100km. If there is imaging of the full beam of SPF at 50km then we have plenty of margin for doing PAF survey imaging. The science component of the presentation should describe the baseline/resolution requirement and possibly elsewhere we make the point that with this baseline limitation the image process is less than that for a longer baseline SPF. This retires one of our risks. Jaap in memo 131 points out that the imaging of our field of view is one the PAFs major risks.

Also here is how i see the system. We have a choice of PAF, a choice of how to get the data of the PAF and after that it is "just" engineering. All links from the focus to the "remote" site are optical.

Regards,

John
<att8381d.png>
-----Original Message-----
From: owner-pafska_at_atnf.csiro.au<mailto:owner-pafska_at_atnf.csiro.au> [mailto:owner-pafska_at_atnf.csiro.au] On Behalf Of Carole.Jackson_at_csiro.au<mailto:Carole.Jackson_at_csiro.au>
Sent: Wednesday, 27 April 2011 6:30 PM
To: pafska_at_atnf.csiro.au<mailto:pafska_at_atnf.csiro.au>
Subject: [pafska] PAFSKA teleconf 27 April 21 UT ; additional item for discussion

Hi

Attached is a draft agenda for the PAF contributions to the WP2.2 Dish Array CoDR scheduled for mid-July.

For discussion at the telecom; please review and consider where your institute will be contributing.

Best regards
Carole

.................

Australia (0)282239584
Canada 18004571636
China 4008811051
France (0)171230600
Germany (0)6995096471
Netherlands (0)202015232
Portugal 214159140
South Africa 0214273061
UK (0)2079405700 or 08448002845
USA East 17183541291
USA West 14089160665

The participant pass code is 21299063#

Mute/un-mute your line with *6

Dr Carole Jackson
Business Development Manager & WP2 leader
CSIRO Astronomy and Space Science

Phone: +61 2 9372 4407 | Mobile: 0432 925 932
carole.jackson_at_csiro.au<mailto:carole.jackson_at_csiro.au> | www.csiro.au<http://www.csiro.au> | www.atnf.csiro.au<http://www.atnf.csiro.au>
Address: CSIRO RadioPhysics Laboratory, cnr Vimiera & Pembroke Roads, Marsfield, NSW 2212

PLEASE NOTE
The information contained in this email may be confidential or privileged. Any unauthorised use or disclosure is prohibited. If you have received this email in error, please delete it immediately and notify the sender by return email. Thank you. To the extent permitted by law, CSIRO does not represent, warrant and/or guarantee that the integrity of this communication has been maintained or that the communication is free of errors, virus, interception or interference.
Please consider the environment before printing this email.
Received on 2011-04-28 17:52:38