[l008com]

Just got booted of off 2.04, so I need to get up to speed. Is one of these faster? I'm reading about people having trouble connecting to the seti server, i never ever had such problems on 2.04, but my first unit on 3.0 seemed to take hours before it would download. I guess 3.04 make connections faster, but I think I heard that 3.04 computes slower!???

[reader50]

3.03/3.04 uses a new server address, so you might well get faster connections. It is still under the Berkeley bandwidth limit, so data transfer speed after connection should not be affected. The idea is that the old server address will presently be dropped, a very final way to lock out the earlier client versions once and for all. I believe 3.0 will fall into the locked out category presently.

3.03+ is definitely slower than 3.0 on all platforms. Berkeley increased analysis to slow down the units. This was required by the UC Berkeley bandwidth limit - although the problem really traces to the release of 3.0 for Windows. It was so much faster that bandwidth demand probably increased 40% or more. If they had just kept the platform releases uniform speed-wise instead of doing some optimization on the Windows code, I don't think we would have run into this problem just yet.

[l008com]

This really sucks, Seti is all about the stats, and not they have totally smashed everyone's stats page.
Whatever the case, lets hope they "optimize" our code next, do all those Dual G4's out there will really haul ass. Even though it will make the bandwidth problem worse

------------------
Make Money With Your Macintosh!

[l008com]

I seperate dedicated internet connection can't possibly cost that much, for a project of this size?!
It's stupid that so many of us at "fighting" to make seti faster, and doing things like running 20 computer using seti at once, and they go and make set go slower, it's so anoying!
------------------
Make Money With Your Macintosh!

[This message has been edited by l008com (edited 01-06-2001).]

[reader50]

UC Berkeley pays $18,000 per month for net access. UC is letting SETI use some of that bandwidth for free, but limits them to about 25% I believe. That is a lot of free net connection to walk away from. And if they do go to their own expensive connection, they would really need to get new servers too to handle the extra load.

Still, you are right. This sucks on several levels.

[l008com]

They could get a dedicated T1 and use that AND thier free bandwidth from the school, that would pretty much solve the problem, wouldn't it, and a T1 is NOT that expesive right?

------------------
Make Money With Your Macintosh!

[reader50]

I don't think a T1's capacity would even come close to what SETI uses.

They are currently limited to 30 Mbit/sec upload. I don't remember right off how that compares to a T1.

[This message has been edited by reader50 (edited 01-06-2001).]

[l008com]

Well then come on, lets talk numbers.....

A T3 can do 45mbits/second. that will much more than double the current bandwidth of seti. How much does a T3 cost? certainly not nearly $18,000/month!?

------------------
Make Money With Your Macintosh!

[Misha]

A T1 is 1.5 Mbps and costs about $800 a month.
A T3 is 45 Mbps and costs about $15,000 a month.

These are business rates. I'm sure schools and non-profits get it less. My guess is Berkeley has an OC3. I'm sure you could figure this out by looking into it (OC3 is 130 or 150 Mbps, I forget which).

A better solution would be the 100 Mbps connection for $1,000 a month from some company whose name escapes me... it's only available in NYC right now. It will be in CA in the Spring. Talk about value...

[l008com]

My cable modem is 1.5 Mbit and it costs $29.99/mo. hmmm..............

------------------
Make Money With Your Macintosh!

[Scotttheking]

In the fine print of the cable modem contract it says, "Up to 1.5Mbit/s". Also, most limit upstream to 384k/sec. Most of Seti's load is upstream.

A T1 is guarenteed at 1.5Mbit/s both ways. That is why it costs so much.

[l008com]

yes my upstream is not 1.5, although it's a little faster than 384K, but my downstream is as much gaurenteed as any other connection. I get 1.5MB or more when i'm connected to somethign thats faster, like a college (napster), and when I connect to someone on a 56K, it crawls, jsut like if i was on T1, except for that upstream.

384*117=45000
117*29.99=$3,508

So even 117 Mediaone Cable connections will be cheaper than a T1. With all the contributions and stuff, it jsut seems like they should stop focusing on making seti slower and instead splurge on some faster connection

------------------
Make Money With Your Macintosh!

[Cipher13]

A faster connection would be of limited help.
You still have to remember that they just aren't getting enough data for everyone to process.
The worlds larges supercomputer is too powerful.
They need another radio satellite.
More data...

Cipher13

[LouZer]

Seems like everyone here thinks the Seti folks are just rolling in dough and its goal is to stage one mother of a CPU race. First, remember that UC Berkeley is DONATING the bandwidth, the Seti folks aren't paying for it. Second, most likely most of the work done on the project is probably by professors, students, and some researchers, but certainly not a group with a huge budget. Third, this is, to the Seti people at least, about science, not about whether it takes your computer twice as long to run a work-unit.

The whole point of v3.04 and above is (1) it takes more time, thus cutting the bandwidth issue, (2) it does more calculatations over a wider area, which in turn gives them more of an area to search, and (3) everyone has to use it! (Well, everyone is supposed to use it, but there are those who will try to keep away from it as long as possible because it drops them from the 99.1 percentile to the 99.0 percentile. "Oh no! Life won't be worth living anymore if I can't crunch a work unit in less than 10 hours!"

[If you want to be really pissed off, why don't you get upset over the fact that each workunit is actually tested by several users, not just one. So you're precious computer time is spent testing a unit most likely already tested by another.]

[Scotttheking]

3.03 times aren't that bad.
18 hrs per unit on my G3-400 with 384MB RAM (does this make a difference)
44(approx, haven't calculated it out perfectly) hrs per unit on my K6-233 Linux machine running in background (no graphics) 64MB RAM
56 hrs on my PM8500-120 48MB RAM
120 hrs on my PM6100-66 40MB RAM

See, not so bad.
I've got 34 units in 2.5 weeks. So just upgrade already. If it is too slow, donate your slow comps to me

[l008com]

I have a 7500 upgrded to 500MHz G3, and I was doing units in just under 6 hours. Now I'm on 3.0 doing them in over 8 hours. 3.03 will probably take me back up to 10? That's a big difference. And that might not be a huge deal for fast computers, but for the slow ones, that took 100+ horus on 2.04 will be taking 200+ on 3.03. Why bother!?!?!? With a comptuer that slow, there's no way you could ever finish a unit before someone else got it anyway. And how much electricity does your computer use to do just one signle work unit? ALOT.

------------------
Make Money With Your Macintosh!

[ddiokno]

Is the Mac OSX PB version still 3.0, or was it also updated to 3.03 or 3.04?
thanks
dave

[Scotttheking]

6100 is the slowest. I don't have my 60MHz one anymore, so I can't test on that, but it isn't much slower then my 66MHz one, and that one is running 3.03 and can do a unit in 110-120 hours. All powermacs can get a unit done per week. I justify running the 6100 because it bridges my localtalk and ethernet networks. Comps don't use THAT much electricity if they have the monitors off. If people don't want to run SETI on their powermacs, I'll take em. I've got room for 7 more comps with what ethernet ports and localtalk boxes I have.

[EqualTime]

SETI has never optimized any of its clients for Intel processors. Or any other processor. The clients are coded only for compatibility with operating systems, not for performance. The reason you see 600-700MHz PCs with lower average times than the fastest MACs is because those PCs are simply faster at calculating the data. The older 2.0-series SETI client benefitted from the large L2 cache of the MACs, Intel Xeons, and to some degree the katmai core Pentium3(512K). The 3.0 client is not as cache-dependent, so the work unit 'process time' scales downward with increases in clockspeed. The large L2 cache is of (nearly) no benefit in the 3.0-series of clients. The Intel machines will continue to be fast because their clockspeed is increasing almost by the week. Their is no optimization of MMX, SSE, or anything else in the SETI clients. Its performance depends on the performance of the Floating Point Unit in your processor. They do not have an Altivec-optimized SETI client because they don't optimize any of their clients for any particular processor.
Don't blame the client. Upgrade your machine.

[Todd Madson]

Actually, not only does it depend on the processor, the operating system, and the client but also what you're doing on the machine.

My collection of WinNT 4 machines has had varying performance, the Mac
stats will be shown afterwards.

I've run the both the deschutes (PII cpu) @ 350 mhz and got around 10-11
hrs once I made the move to 3.x client. This was a desktop machine
with 128 megs ram.

The coppermine (PIII) performance I experienced @ 500 mhz has been
roughly similar, but usually from 7-9 hours on 2.04, on 3.x I've been
getting somewhere over ten hours. This was a IBM Thinkpad 600X
believe it or not.

The dixon processor (mobile pentium II) @ 365 mhz was about 9 hours
on 2.04, on 3.x about the high tens to mid elevens average. Another
Thinkpad (600E).

All of these processors have 512k of L2 cache.

The G4/400 I've had with overclocked L2 cache (267.5 mhz) has always
(at 2.04 graphical) given me 5-6 hours block times under MacOS9, under
Mac OS X DP 4 with the 2.04 CLI client I'd get 6-7 hour times (which was nice because I could run three instances of the app and stagger their block times every few hours and not experience slowdown). I never did figure out what the GUI version of OS X's client would give because I gave up well into 15 hours. Way too slow. With the 3.x client on OS9.04 the
projections show about 10 hour block times even with a 1 meg L2 cache.
I've currently harnessed the G4 to try out RC5 since that has a small,
fast efficient altivec client.

My wifes iMac DV SE with 128 megs was pretty much a 10 hour machine on
2.04. Now with 3.x, it's usually an hour or two beyond that. She's got
512k L2 cache.

Prior to 3.x, machines with >1 meg of L2 cache had an unfair advantage in
some ways. Now, machines with 512k of L2 cache seem to have this
advantage - or rather, they've levelled the playing field.

(1) Seti can only get so many blocks from the telescope.
(2) Seti isn't made of money - it's totally non-profit.
(3) They don't want super fast processors ripping thru work units.
(4) They only have so much internet bandwidth.
(5) They are only running this for a finite amount of time.

It's not a CPU race (although that aspect has been fun, it's not
a way to show how big your stick is - rather, it's a science thing.

You figure people would want to run RC5 because there you can
win cash as well as rip through work units like mad.

Of course, the Seti side of things has a tantalizing proof of
alien life thing.

[EqualTime]

The L2 cache size has almost no impact on the 3.0 client's performance - neither does the amount of RAM (within reason).

My machine has the same average time when I disable the L2 cache altogether.

By the way, the folks in charge of SETI Berkeley aren't the best around.

Don't assume there is a good reason for the things they do, mostly it's just mis-managed.

[l008com]

What is this "RC5" thing?

------------------
Make Money With Your Macintosh!

[druber]

RC5 is a security contest, to say it simply. Secure servers on the net are encrypted to keep things like credit card #s safe. RC5 is a contest to crack a specific key; the prize is $10k. Not really sure what the point of it is, but you can find the client and lots of info at distributed.net.
Personally, I really wouldn't mind joining RC5 if they kept stats like Seti does. Maybe if MacNN also hosted an RC5 forum as well... I know they wouldn't want to support every distributed computing program out there, but maybe people would enjoy a contest that's purely a contest as well..

[Todd Madson]

It's essentially a contest to prove to the convernment that 64-bit
encryption isn't enough, that, with concerted effort even an idiot
could break 64-bit encryption given enough computers.

(1) RC5 clients, are small (not 20 megs like seti).
(2) Don't hog the cpu time displaying pretty graphics.
(3) Won't make the machine unusable by running it.
(4) Are significantly faster than Seti.

I think I've got somewhere over 150,000 work units on RC5, whereas
on Seti I've got 1500.

[l008com]

20 MB??? What version are you one? My client is 400K

------------------
Make Money With Your Macintosh!

[Scotttheking]

It uses 20MB of RAM.

[l008com]

oooooooooooooooooooooooh I get it.
I have teh dnet client, and it's really horrible. Going without 'pretty' graphics is one thing, but a text based interface, selecting items by typeing numbers and hitting return. That ancient history. Is it so hard toad a button?

------------------
Make Money With Your Macintosh!

[Todd Madson]

Pretty buttons and graphics cause processor overhead. Seti is the thing that's horrible, that graphic display is cute but you can do so much more by not displaying it. The OS X command line client is a lot better.

[l008com]

You can set seti to shut off the visual completely in as little as one minute. Having pretty visuals when needed just makes people more likely to use it, and make it easier to use. I'm sure many truley use it as a screen saver, something you cannot do with dnet, adn even if you did, how much is a nonchanging text window goona save your screen?

------------------
Make Money With Your Macintosh!