nfsnet.org updates

[Wacky]

Originally Posted by bdodson

Richard does respond to posts in this thread/subforum. Did so yesterday, in fact. NFSNET is his project. I know for a fact that he'd be a lot happier to have more contributors from among the non-diehard-fanatical.
He seems a bit gloomy about the prospects. I don't know why he hasn't
pursued fivemack's offer.

Actually, I sent "fivemack" a private message and he responded with an e-mail which I answered.

Until I know more about the nature of the services which we could utilize on this 'well-connected system', I cannot assess whether or not it is likely to be more helpful than trying to get Tom to set up something on one of my servers (which have their own problems).

However, please do not think that I am ignoring Tom's offer.

Most of you are not aware of just how "fragile" our previous "stats" system was. It was held together by quite a bit of "behind the scenes" manual intervention on the part of multiple people. If we are to have stats again, someone will need to either be willing to supply that effort or expend the effort to rewrite parts of the system to improve the reliability/error recovery.

In the past few years, I have had less time to devote to NFSNet and have been quite disappointed because of the lack of useful results from the time that I have spent educating potential contributors about the inner workings of the system only to have them lose interest before they have contributed any visible improvements.

[frmky]

I apologize for the tone of my reply above. The reply came more from genuine confusion than annoyance, but despite the emoticon that wasn't at all clear. Although I've only contributed a little to the project over the past year (I've instead been using GGNFS to do a few hundred smaller SNFS factorizations), I do have a genuine desire to see it succeed. After all, my first factoring contributions were to NFSNet when it was run by Conrad Curry.

Greg

Originally Posted by bdodson

Uhm, you're suggesting that I ought to be scanning for the owner of
/~twomack/homcun.pl rather than for fivemack? To be honest, the
"colocated dual-P3" as server threw me off a bit. Bad judgement on my
part for a misdirected reply to the 10am post; then an apparently
unsuccessful attempt to back-off. I did my best to provide some
additional info; so far as I can see, the reply borders on non-responsive.
Perhaps I got off lucky, as I can see that I might not have fared well with
something more directly responsive. Thanks.

Richard does respond to posts in this thread/subforum. Did so yesterday,
in fact. NFSNET is his project. I know for a fact that he'd be a lot
happier to have more contributors from among the non-diehard-fanatical.
He seems a bit gloomy about the prospects. I don't know why he hasn't
pursued fivemack's offer. Of the reservations you refer to,

3^451-2^451, c135.6, difficulty 215.2 is the c135 gnfs? And
6^256+5^256, c199.3, diff 199.2 would surely be the snfs.

I'm not completely sure why you're so annoyed; these seem to me
likely to provide useful experience for someone interested in promoting
the nfsnet project. Some years ago I did a c137 gnfs myself (using
the CWI suite), and the sieving/post_processing of a c142 gnfs, before
sending the matrix off to Paul. And M727 by snfs, before sending the
matrix off to CWI (where it was the first hard test case for their parallel
Lanczos, using 25 cpus). Each with some sieving contributions from other
people. I also managed contributions from 6-7 people on the factorization
of 2^773+1, the first 768-bit snfs, with hand task-distribution and data
collection by ftp. Even earlier, our paper on RSA130 is titled "a world-
wide factoring record ...", a reference to the attempt to get contributions
from non-diehard-fanatics, using the web. You most likely don't want to
hear about factor-by-email, from '87-'91 or so, before the web; of which
I caught the last couple projects.

The skill-sets needed for these collaborative projects certainly include
a non-trivial amount of fanaticism. Paul's probably been the most
successful, with RSA129 (and the famous 5000 computers); but the
current nfsnet setup, together with the lamented stats pages was/is
a considerable improvement beyond anything seen before. I'd like to
suggest that a little more tolerance may be in order. Richard has been
doing some fundamental re-thinking. I haven't been deligated as
recruiter for the stats-page position, if there is one. I don't want to
discourage fivemack; on the contrary, he seems to be doing all of the
right things. Managing a stats-server without sys-admin privileges does
seem a bit unlikely, but that's just me --- actually, my initial post from
March does report that server access/experience, say at the univ
sys-admin level, was part of what the previous stats person did. But,
again, that list-serve/email_post was from the previous stats person,
not Richard, who has both the first and last word on what's wanted/needed
for the position (if there even is one ...). -Bruce

[fivemack]

Originally Posted by Wacky

Actually, I sent "fivemack" a private message and he responded with an e-mail which I answered.

Until I know more about the nature of the services which we could utilize on this 'well-connected system', I cannot assess whether or not it is likely to be more helpful than trying to get Tom to set up something on one of my servers (which have their own problems).

Unfortunately, the sysadmin reacts to the idea of running a CGI program written in C++ on his machine as if I had insulted his sister's honour, so I fear it may be necessary to move to a plan B of some kind ... a virtual dedicated server in England seems to cost about $40 per month, which would buy me a decent compute-farm node every nine months, and I think I'd rather have the nodes. Does anyone know of cheaper hosting options?

I'm going off now to buy the sysadmin a drink and try to convince him that statsreporter.cpp is not in fact personally forged by demons out of Satan's infernal mucus ...

[Wacky]

Originally Posted by fivemack

Unfortunately, the sysadmin reacts to the idea of running a CGI program written in C++ on his machine as if I had insulted his sister's honour

Tom,

Is he more agreeable to something in php, perl, etc. ?
I cannot imagine that he can reasonably object to some kind of program to "massage" incoming data and format it as a report.

[fivemack]

On further discussion over a glass of Frambozenbeer, it seems that what I read as a veto was in fact a strong claim that C++ wasn't the right language; he'd be happier with perl, and statsreporter isn't so big or so complicated a program that recoding it in perl would be troublesome. So, less worry there than I'd feared.

[Jushi]

Originally Posted by fivemack

On further discussion over a glass of Frambozenbeer, it seems that what I read as a veto was in fact a strong claim that C++ wasn't the right language; he'd be happier with perl, and statsreporter isn't so big or so complicated a program that recoding it in perl would be troublesome. So, less worry there than I'd feared.

Why should it matter in which language the problem is written? Why does he care? If he's afraid of security (which he should be in any case), then let the script run on its own account. I don't see what the problem could be then.

In case: I have a server at the University of Gent on which I can give you an account for NFSNET, it's a Pentium D 2.8GHz (dual core) with 4 GB of RAM. The only problem is that I might be leaving UGent in some months (I did my PhD, and I don't know yet where I will go to).