?

Log in

No account? Create an account
Pete Zaitcev's Journal -- Day [entries|friends|calendar]
Pete Zaitcev

[ userinfo | livejournal userinfo ]
[ calendar | livejournal calendar ]

[19 Aug 2004|10:52am]
I am a corrupter of minds.

Red Hat Marketing posted a Summer Reading List which included a recommendation of mine, Schimmel's "UNIX Systems for Modern Architectures". It is a wonderful book, but it has to be kept in mind that it is great primarily because it clearly delineates its area and covers it comprehensively, but without "water". If you hack on kernel and you see a data corruption after DMA, it is the book for you. If you hack on GNOME, it is unlikely to be useful.

Tangentially, our website search engine is worse than useless. I tried to search for "summer list" and "reading list" without success. Sure, it probably finds the needed document, down in 10,000th hit. This begs a question, if a decent implementation of page rankings exist in a small Free Software package, such as ht://Dig. Advogato's mod_virgille is not that big. You do not need a million computers to search one small website, and you do not have to be terribly attack resistant if the content is under central control. Google probably patents its Page Rank, but you cannot patent the eigenvector or such math (thanks heavens).

Interestingly enough, I saw firms buying mini-Google to scan intranets, with strangely mixed results (results missing, poor rankings). One explanation might be that Google does actually require a lot of concentrated horsepower to function effectively. If so, it bodes ill for "a small mod_virguille derivative".
post comment

[19 Aug 2004|11:48am]
Val, blogs, SHA-1, Slashdot

http://blogs.sun.com/roller/page/val/20040817

``Thanks to Fred Douglis for adding a comment to my weblog pointing me to these results. For the record, no, I don't read Slashdot, but I'm beginning to think I should get back into the habit...''

I fail to see the logics here. The fact that Val did receive the link despite ignoring Slashdot proves that she obtains all the needed information successfully without reading Slashdot. So, why go back? Slashdot is obsolete. It lags LWN and Groklaw, it has no filtering, it's slow (especially in archives), it is unfocused. The network of blogs feeding on competing publications and a range of better content aggregators replaced Slashdot as the aggregator.
post comment

[19 Aug 2004|11:09pm]
New day, new setback for ub (or ub's schedule). This time it turned out that urb->timeout field is just for show and setting it may have no effect in 2.6. In 2.4, both usb-uhci and usb-ohci processed timeouts and everything worked fine (except USB 2.0 with its EHCI, which was "experimental" anyway).

In 2.6, several things happened. On UHCI side, JE's uhci-hcd does timeouts by comparing timestamps every frame. On OHCI front, the timeout implementation was hockey, so it was declared broken and removed. It was easier to do than to figure how to make it not to conflict with asynchronous unlink feature. And for EHCI, nobody in general and not David-B in particular bothered to implement timeouts. The common mood (by Stern et. al.) is to look at all this brothel and to declare it a feature: URB timeout are officially non-existent in 2.6. David-B even claimed that "Only UHCI has ever used urb->timeout."

It was entirely my failure to verify that timeouts worked before relying on them. Now I have to go in and implement explicit timeouts and unlink in ub. It is a relatively simple task, but it takes time.
post comment

navigation
[ viewing | August 19th, 2004 ]
[ go | previous day|next day ]