"Universities can cobble together a cluster of off-the-shelf PCs in a couple of weeks and create a relatively cheap cluster that takes at least the 20 spot on the Top500 fastest machines list" trumpets El Reg.
Keep in mind that the NICs and switch ports for networks other than Ethernet tend to be quite expensive (with a better price/performance ratio than 10GE at the moment).
Let's review:
(source material)
Rank Interconnect
1. Custom IBM
2. Custom SGI (NUMAlink) and InfiniBand
3. Custom NEC
4. Myrinet
5. Quadrics
6. Quadrics
7. InfiniBand
8. Custom IBM
9. Custom IBM (Federation)
10. Myrinet
11. Custom IBM (Federation)
12. Custom IBM (Federation)
13. Myrinet
14. InfiniBand, Myrinet, Custom NEC
15. Custom IBM
16. Quadrics
17. Myrinet
18. Myrinet
19. Quadrics
20. Custom IBM (SP2)
None of the top 20 (with the exception of the Riken Super Combined Cluster at #14) use Gigabit Ethernet as an interconnect.
So "relatively cheap" is cheap compared to what?
Saturday, November 13, 2004
Irony is everything
ABC and several (up to 65) affiliate stations recently self-censored, by choosing not to show Saving Private Ryan in order to avoid possible FCC indecency rulings.
I was all set to write an article about the Parents Television Council and their statement of "context is everything".
On the surface, this seems reasonable, but consider, for example, their attempt to change the context itself. I'm sure that a definition that can be re-moulded on demand by massive mobilization spurred by the PTC is exactly what they would prefer.
To cut a short article long, Lisa de Moraes has beaten me to the punch.
I was all set to write an article about the Parents Television Council and their statement of "context is everything".
On the surface, this seems reasonable, but consider, for example, their attempt to change the context itself. I'm sure that a definition that can be re-moulded on demand by massive mobilization spurred by the PTC is exactly what they would prefer.
To cut a short article long, Lisa de Moraes has beaten me to the punch.
Monday, November 08, 2004
Selection is censorship
The Register rabbits on (again) about the dangers of news.google.com using PR as a source of news.
The problem is not whether PR has news value. PR is just pre-packaged and ready for aggregation (a.k.a. 'publishing') by a news intermediary1. Information wants to be [re-]published.
The problem is really whose PR do you re-publish?
If you publish PR from company (or party) of size x, why not publish PR from those of size >x?
Lettice writes "you need to interface the automated system and the human judgment" and that "censorship has to be a basic component of the selection process".
Now this is just plain wrong. Why censor?
This is Google. They of the 10,000 server back-end (appropriately enough the story also talks about PR flacks having difficulty censoring company geeks). They of the mysterious (yet published) PageRank algorithm. A few simple (ahem) modifications and you could have your own biased outlook on the web.
This should be an exercise in hermeneutics, rather than hermits.
- Al
1 Why not dispense with the taxonomy of publisher, PR agency, news agency, broadcaster, aggregator, blog, et al? Any secondary source is an intermediary between a primary source and the end consumer (i.e. you! And you communicate with your friends, so you are an intermediary too, from their point of view). Envision endless chains of re-publication, eventually arriving at a consumer, only to be bounced further along to other consumers.
The primary source then becomes a special case of intermediary, with no input from other sources. All other intermediaries accept and re-distribute after applying idiosyncratic changes to the data.
The problem is not whether PR has news value. PR is just pre-packaged and ready for aggregation (a.k.a. 'publishing') by a news intermediary1. Information wants to be [re-]published.
The problem is really whose PR do you re-publish?
If you publish PR from company (or party) of size x, why not publish PR from those of size >x?
Lettice writes "you need to interface the automated system and the human judgment" and that "censorship has to be a basic component of the selection process".
Now this is just plain wrong. Why censor?
This is Google. They of the 10,000 server back-end (appropriately enough the story also talks about PR flacks having difficulty censoring company geeks). They of the mysterious (yet published) PageRank algorithm. A few simple (ahem) modifications and you could have your own biased outlook on the web.
This should be an exercise in hermeneutics, rather than hermits.
- Al
1 Why not dispense with the taxonomy of publisher, PR agency, news agency, broadcaster, aggregator, blog, et al? Any secondary source is an intermediary between a primary source and the end consumer (i.e. you! And you communicate with your friends, so you are an intermediary too, from their point of view). Envision endless chains of re-publication, eventually arriving at a consumer, only to be bounced further along to other consumers.
The primary source then becomes a special case of intermediary, with no input from other sources. All other intermediaries accept and re-distribute after applying idiosyncratic changes to the data.
Sunday, November 07, 2004
Recognizing blogging bias
(from http://osviews.com/modules.php?op=modload&name=News&file=article&sid=2722)
Kelly McNeill, operator of osviews.com, writes about bias in blogging:
"publishers of tech news sites that cherry pick links, and claim that this strategy keeps the site from being associated with bias", "a trend in continually linking to select number of articles that cary [sic] the same controversial subject matter", etc.
Blogs, as opposed to more traditional news aggregators (e.g. Slashdot) and news sites (e.g. The Register), are surely more personal in nature. The 'party line' that is followed is the often varying (based on differing input), often inconsistent (based on similar input) point of view of one individual. This point of view has a low "ideological inertia" and so can be tossed on the tempest of perception of an individual subjecting his- or herself to such a wide or narrow variety of inputs.
'Man is not a rational animal; he is a rationalizing animal.'
- Gulf by Robert Heinlein
McNeill then writes two closely entwined but seemingly contradictory statements:
McNeill's disillusionment seems to stem from tiring of sorting the sheep from the goats, the wheat from the chaff, looking for the diamond in the rough and the Emperor's New Clothes. Is it bad that the "publishers are sifting it for you"?
This is no more than a change in roles. The publisher is currently the teller of the story. The 'facts' are presented. Do you want the AP, Reuters, AFP, Xinhua, ABCD[isney], CBS, Fox, SABC, BBC, DW, Ananova, Daily *, Sunday *, *Times or even usinfo.state.gov version?
News.google.com, Slashdot, Fark, et al aggregate these for you. Browsing from a central repository might show inconsistency in interpretation, but probably not on basic facts (at least, not after the first few versions have been proofread and fact-checked by millions).
The choice is NOT: Who is your source?
The choice is NOW: Who is your source-choosing proxy?
If your proxy is news.google.com, how do you evaluate news.google.com? Is it systematically leading you to the Dark Side (Bush / Microsoft / Tory / Nuclear - delete as appropriate)? The Online Journalism Review's article Balancing Act shows that even simple differences in search terms can show difference in bias.
So, who do you trust?
Who do you trust?
You cannot personally verify details of friends of friends of friends ad nauseam. Google's Orkut helps build social networks, and all that remains is for trust (á là PGP) and recommendations (á là Amazon, Firefly, et al) to be integrated and presented individually.
Google does it already, and makes the results available free of charge to the public. It's called PageRank :) But it's not individualised.
Not yet.
Kelly McNeill, operator of osviews.com, writes about bias in blogging:
"publishers of tech news sites that cherry pick links, and claim that this strategy keeps the site from being associated with bias", "a trend in continually linking to select number of articles that cary [sic] the same controversial subject matter", etc.
Blogs, as opposed to more traditional news aggregators (e.g. Slashdot) and news sites (e.g. The Register), are surely more personal in nature. The 'party line' that is followed is the often varying (based on differing input), often inconsistent (based on similar input) point of view of one individual. This point of view has a low "ideological inertia" and so can be tossed on the tempest of perception of an individual subjecting his- or herself to such a wide or narrow variety of inputs.
'Man is not a rational animal; he is a rationalizing animal.'
- Gulf by Robert Heinlein
McNeill then writes two closely entwined but seemingly contradictory statements:
- 'The only logical alternative is to let the community dictate coverage'
- 'News editors that feel inclined to incorporate personal bias into the content only stand to eventually cause their readers to be disenchanted with the content they publish'
McNeill's disillusionment seems to stem from tiring of sorting the sheep from the goats, the wheat from the chaff, looking for the diamond in the rough and the Emperor's New Clothes. Is it bad that the "publishers are sifting it for you"?
This is no more than a change in roles. The publisher is currently the teller of the story. The 'facts' are presented. Do you want the AP, Reuters, AFP, Xinhua, ABCD[isney], CBS, Fox, SABC, BBC, DW, Ananova, Daily *, Sunday *, *Times or even usinfo.state.gov version?
News.google.com, Slashdot, Fark, et al aggregate these for you. Browsing from a central repository might show inconsistency in interpretation, but probably not on basic facts (at least, not after the first few versions have been proofread and fact-checked by millions).
The choice is NOT: Who is your source?
The choice is NOW: Who is your source-choosing proxy?
If your proxy is news.google.com, how do you evaluate news.google.com? Is it systematically leading you to the Dark Side (Bush / Microsoft / Tory / Nuclear - delete as appropriate)? The Online Journalism Review's article Balancing Act shows that even simple differences in search terms can show difference in bias.
So, who do you trust?
Who do you trust?
You cannot personally verify details of friends of friends of friends ad nauseam. Google's Orkut helps build social networks, and all that remains is for trust (á là PGP) and recommendations (á là Amazon, Firefly, et al) to be integrated and presented individually.
Google does it already, and makes the results available free of charge to the public. It's called PageRank :) But it's not individualised.
Not yet.
Subscribe to:
Posts (Atom)