I also like Ubuntu quite a lot. It's comfortable to use and really versatile, offers lots of software and good package management. So if you get along with it I can really recommend it.
Ubuntu is a great distro for the desktop. I have it on several of my "antique" machines, and it runs fine on all of them. (In fact, just yesterday I performed the Gutsy Gibbon upgrade on one of my P-III 450's.)
For servers, though, it's a different story. Ubuntu Server would work okay for a small business, but it has a few design flaws that prevent it from being a serious contender for enterprise work. For example, thanks to their use of "Upstart" instead of the normal SysVInit, High Availability clustering just doesn't work. (Unless, that is, that got fixed with the Gutsy Gibbon release.)
For desktops, I heartily recommend Ubuntu. For servers, I would say to go with either one of the Red Hat derivatives, or with Debian 4.0.
Downloaded Mandriva late last night. Loaded it on an old dual Pentium Pro 200, 1GB 50ns EDO ECC Dimms, 12.7GB HDD, onboard LAN, sound, ........ A little playing around and it connected up to the network and the internet. With Ubuntu I could connect to the network, but not the internet. It kept kicking me back to the logon screen every time.
I think what I will do is see if I can find several more HDD trays and load different distro's on each one. They are pretty handy, have a DOS/Win98 SE/Win2k Pro machine setup that way. Simple deal, mount the frame in a 5.25 bay, install HDD in the tray, slide tray into frame, lock, and you are ready to go.
It was very slooooooooow last night. Changed out memory modules this morning and it seems to be doing much better, still a little slow, but much better. At least with this setup I can get online for any updates. Played with it till about 3 AM! At least now I'm making some progress in the right direction!
Mandriva feels sluggish also on newer computers, Kubuntu or Debian KDE is noticeable better if you want KDE. Or Xubuntu, lightweight gui.
Quote:
For servers, I would say to go with either one of the Red Hat derivatives, or with Debian 4.0.
I prefer Debian also for desktop, Debian KDE to be more specific (Debian is gnome by default). Debian 40r1 KDE has been more stable than Kubuntu ever was on my main rig. About Ubuntu, i cant understand why anyone would want Gnome over KDE.
Okay, seems like I was not quite specific enough. ;-)
@Donald Trevault: I quite agree. In my posting I was referring exclusively to desktop use, since that seemed to be what Jim had in mind. Like you I will always choose Debian for a server if I can. Solid, stable, reasonably secure. But on a desktop the newer software packages, easier usability and better hardware support that Ubuntu has to offer really make this OS more attractive to me. And you don't have to learn everything from scratch if you have experience with Debian, since package management and most of the commands are basically the same.
@th3: Yep, I also like KDE quite a lot more. It simply feels more intuitive ( at least to someone who has never owned a Mac) and I also like some of the programs that come with it. Not to mention that, to my eyes, it simply looks better (which is, of course, entirely subjective). So, you are right, I should have referred to Kubuntu here. I simply used the term "Ubuntu" as Jim used that as well, and it is most well-known. Besides, I figured the GUI was really a very subjective choice or a taste question.
Downloaded Debian and Fedora this afternoon. I'll wait to load them until I can pickup 3-4 removable HDD trays. I could load each on a different 'puter but then I'd have to keep moving the damn things around all the time to use them. Space is somewhat limited in this room.
The main thing right now is finding one that works first time out of the box. Once I have one that works, and Mandriva does, then I can play around and find out how it works. If I turn it into toast, well, so what. There isn't going to be anything on there of importance anyway, and I can always reload the OS.
I will be paying closer attention to Linux discussion from now on. :)
The P4 Xeons will all be on Einstein, except for the pair of 3.06Ghz cpu's in the Z Pro. They are still running the SMP app on FAH. The addition of the x255 makes a total of 12 MP Xeons, eight 2.0Ghz MP's and four 1.9Ghz MP's. The two SC450NX's are both running 4 PIII Xeons at 700Mhz and they are max'ed out cpu wise. I have 2 more that I may or may not put back online, they eat a lot of power for the number of WU's done. Their only saving grace is that they do 4 WU's at a time.
If you look at the 450's you'll notice the cpu's are listed at 600Mhz, it's a problem with the BIOS. Intel says they are running at 700Mhz but the BIOS lists them at 600! They were close to EoL when the 700's came out and Intel never bothered to fix it in BIOS.
For whatever it is worth, I have two computers. The older (and slower) one has HyperThreading turned on, so I get four CPUs reported for the two actually present:
Quote:
CPU type GenuineIntel
Intel(R) Xeon(TM) CPU 1.80GHz [x86 Family 15 Model 2 Stepping 7] [fpu tsc sse sse2 mmx]
Number of CPUs 4
Operating System Microsoft Windows XP
Professional Edition, Service Pack 2, (05.01.2600.00)
Memory 1023.01 MB
Cache 976.56 KB
Measured floating point speed 829.12 million ops/sec
Measured integer speed 839.26 million ops/sec
The other one has HyperThreading turned off, and thus reports only the two physical CPUs:
Quote:
CPU type GenuineIntel
Intel(R) Xeon(TM) CPU 2.40GHz [x86 Family 15 Model 2 Stepping 7]
Number of CPUs 2
Operating System Microsoft Windows XP
Professional Edition, Service Pack 2, (05.01.2600.00)
Memory 767.51 MB
Cache 488.28 KB
Measured floating point speed 963.27 million ops/sec
Measured integer speed 1971.64 million ops/sec
Now, from the above it appears that the floating point speed is better optimized with HyperThreading while the integer speed is more evenly divided and thus gains little (or nothing) due to hyperthreading.
So, is the Einstein algorithm more sensitive to floating point speed, thus indicating that HyperThreading would help, or is it more sensitive to integer speed, this indicating that HyperThreading does little other than let you work on more work units simultaneously, but achieving roughly the same daily credits as the non-Hyper mode?
Now, from the above it appears that the floating point speed is better optimized with HyperThreading while the integer speed is more evenly divided and thus gains little (or nothing) due to hyperthreading.
So, is the Einstein algorithm more sensitive to floating point speed, thus indicating that HyperThreading would help, or is it more sensitive to integer speed, this indicating that HyperThreading does little other than let you work on more work units simultaneously, but achieving roughly the same daily credits as the non-Hyper mode?
The difference in integer and floating point benchmark response to hyperthreading says more about the specifics of the benchmarks than about general floating point vs. integer performance. This really is an application-level characteristic.
I had a Gallatin cpu in my main PC for several years, and about a year or so ago did careful controlled testing (_not_ run it for a couple of days and watch the RAC) of hyperthreading impact on total output for the then-current SETI and Einstein aps.
For the aps of that time the general conclusions were:
1. running 2 SETIs in HT was substantially more productive than 1 not HT.
2. running 2 Einsteins of the stock ap in HT was clearly more productive than 1 not HT, but quite a lot less advantage than for SETI
3. running one SETI and 1 Einstein on the two threads was extra-special good.
4. But at a single point in a long series of code efficiency improvement by akosf, the 2 Einsteins on HT case suddenly became appreciably lower total output than running 1 in non-HT.
This last case is the only one I'm aware of for which HT actually hurt productivity, and I doubt it is the case with the current aps. But I don't recall seeing properly controlled comparisons done with the current aps.
In the balance of experience (not only my own experiments, but lots of other reports, you'd most likely get somewhat more total output in HT. But don't expect a lot.
So, is the Einstein algorithm more sensitive to floating point speed, thus indicating that HyperThreading would help, or is it more sensitive to integer speed, this indicating that HyperThreading does little other than let you work on more work units simultaneously, but achieving roughly the same daily credits as the non-Hyper mode?
Inquiring minds want to know... Thanks!
Besides integer and floating point performance, there's another important factor that is not (yet?) benchmarked by BOINC: memory bandwidth.
From profiling the current apps I get the impression that both memory bandwidth and "integer performance" do play a significant role in S5R3, so it might be worthwhile to re-evaluate the effectiveness of Hyperthreading for E@H.
Quite some time ago in this very thread, a number of people were encouraging Tullio to make the leap of faith and plug in a PIII Katmai (up to 600MHz) in place of his 400MHz PII Deschutes. Tullio declined, which of course is fair enough, and the matter essentially ended there. One of the participants in the discussion was Thunder whose parting comment included the following.
Quote:
....
Tullio, I think you're worrying too much about the "hardware" end of things. This should be as simple as unplugging the old processor, plugging in the new one and booting back up. I honestly can't remember if SSE will have to be compiled into the linux kernel or not, but unless you really tweaked your kernel, it's probably in there already. WORST CASE is that you'd have to flash the BIOS on your motherboard, which is as simple as formatting a bootable floppy, putting the new BIOS from the manufacturer on it and following a couple of instructions to make it complete.
P.S. If Gary asks, I might just get those 600's 'topless' and take some risque pictures of that pair for him! ;)
The comment was directed at Tullio but included a reference to the fact that I'd never come across a 600MHz Katmai and the highest frequency that I'd actually come across was 550MHz. As Thunder actually possessed a pair of those rare 600MHz beasts and I didn't, his final comment was quite understandable :).
At a local IT auction recently, there were some interesting servers, mixed in with some older ones. There was a compaq DL380 G2 which had dual Tualatin 1400 processors that I was keen to buy. As sometimes happens, the auctioneer, knowing there will be interest in the better spec offerings and little interest in the dinosaurs, will often combine lots and ask for a single price for all.
So, I ended up winning the DL380 for quite a cheap price anyway, with this other old compaq thrown in, essentially for free. When I got them home and had a look, I found that the old dinosaur actually contained a pair of 600MHz Katmais, 512MB RAM and an internal (but not connected and not documented) 73G SCSI drive along with the two 9G drives that were listed. Everything works fine and so I now no longer need Thunder to risk pornography charges with his two little beauties :). Here is its results list and as you can see it's no worldbeater but I couldn't just not give it the opportunity to crunch just because of a bit of electricity, could I? :). For comparison, here is the results list for the DL380 which you can see is way more than twice as fast.
I was interested in reading a bit of history about Katmais in general and the 600MHz version in particular so I did a bit of googling and came up with a local site which has some amazing information on cpus, main boards, hard drives, etc of that era and earlier - right back to the year dot with all manufacturers included. The stuff is very readable and goes into quite a lot of very interesting detail. It's very easy to navigate and quite addictive if you start reading :). There is nothing on recent cpu history - the information essentially ends with the Athlon XP 1800+ in October 2001. The oldest entry is for the Intel 4004 released in 1971. If anyone is interested in cpu history, check out the pages listed in the cpu index. The stuff on Katmais (as well as other cpus of that period) is in the link called "mainstream 1999". Alinator will probably be very interested in this page because the opening entry is for the AMD K6-2/550 and the opening sentence is, "A dog of a chip." :).
A dual Tualatin 1400 in 19" for 100EUR? That's really a good deal.
Yup, and so I ordered another one, so I now have four Tualatins crunching for E@H Home. You even get a 12 month warranty.
Bikeman, a belated congratulations on your choice of the excellent dual Tualatin 1400 as your "legacy crunching farm" mainstay :). My own particular favourite is the Compaq DL380 G2 of which I've now acquired quite a number. In fact I bought two more yesterday, one with 2G RAM which cost 65EUR and the second with 4G RAM which cost 85 EUR. I also bought a pallet containing 4 old HP Netservers plus a few other bits for 3.5EUR ($AUD5). Those 4 have so far yielded 2 x 1.0G cpus, 2 x 933MHz cpus, 2 x 733MHz cpus and a total of 2G of PC133 registered server RAM. There was nobody else even interested in making a bid so my $5 bid prevailed.
Quote:
Quote:
512MB RAM are a little short, if ECC registered SDRAM prices went down as much as the Tualatin prices, you should consider adding some more (you might need low profile)
Currently I'm planning to use those boxes almost exclusively for E@H, and for this 512 MB of RAM under Linux is more than enough even on a dual CPU system.
If you want to upgrade your RAM, just seek out your local IT auction house and go and inspect almost any old 1999/2000/2001 vintage servers to see if any still have RAM inside them. Any with coppermine PIII processor(s) would be perfect. The more battered they look, the better as you will probably be the only person bidding :). This is exactly how the person selling to you on ebay has got their stock and you may as well cut out the middleman :).
Quote:
I found a place in my apartment where the noise doesn't bother anyone, so I guess I'll leave the noisy and mostly redundant fans as they are for the moment.
If the noise of one or two servers becomes a problem then I guess a stack of six or so would be out of the question :). I've taken a picture of just such a stack and would like to post it here but (never having done this before) am not sure of the possible ways to do it (and what to avoid, etc). Perhaps someone might be kind enough to give me a brief run through on "How best to post images - for dummies 101" thanks :).
RE: I also like Ubuntu
)
Ubuntu is a great distro for the desktop. I have it on several of my "antique" machines, and it runs fine on all of them. (In fact, just yesterday I performed the Gutsy Gibbon upgrade on one of my P-III 450's.)
For servers, though, it's a different story. Ubuntu Server would work okay for a small business, but it has a few design flaws that prevent it from being a serious contender for enterprise work. For example, thanks to their use of "Upstart" instead of the normal SysVInit, High Availability clustering just doesn't work. (Unless, that is, that got fixed with the Gutsy Gibbon release.)
For desktops, I heartily recommend Ubuntu. For servers, I would say to go with either one of the Red Hat derivatives, or with Debian 4.0.
Downloaded Mandriva late last
)
Downloaded Mandriva late last night. Loaded it on an old dual Pentium Pro 200, 1GB 50ns EDO ECC Dimms, 12.7GB HDD, onboard LAN, sound, ........ A little playing around and it connected up to the network and the internet. With Ubuntu I could connect to the network, but not the internet. It kept kicking me back to the logon screen every time.
I think what I will do is see if I can find several more HDD trays and load different distro's on each one. They are pretty handy, have a DOS/Win98 SE/Win2k Pro machine setup that way. Simple deal, mount the frame in a 5.25 bay, install HDD in the tray, slide tray into frame, lock, and you are ready to go.
It was very slooooooooow last night. Changed out memory modules this morning and it seems to be doing much better, still a little slow, but much better. At least with this setup I can get online for any updates. Played with it till about 3 AM! At least now I'm making some progress in the right direction!
Mandriva feels sluggish also
)
Mandriva feels sluggish also on newer computers, Kubuntu or Debian KDE is noticeable better if you want KDE. Or Xubuntu, lightweight gui.
I prefer Debian also for desktop, Debian KDE to be more specific (Debian is gnome by default). Debian 40r1 KDE has been more stable than Kubuntu ever was on my main rig. About Ubuntu, i cant understand why anyone would want Gnome over KDE.
Team Philippines
Okay, seems like I was not
)
Okay, seems like I was not quite specific enough. ;-)
@Donald Trevault: I quite agree. In my posting I was referring exclusively to desktop use, since that seemed to be what Jim had in mind. Like you I will always choose Debian for a server if I can. Solid, stable, reasonably secure. But on a desktop the newer software packages, easier usability and better hardware support that Ubuntu has to offer really make this OS more attractive to me. And you don't have to learn everything from scratch if you have experience with Debian, since package management and most of the commands are basically the same.
@th3: Yep, I also like KDE quite a lot more. It simply feels more intuitive ( at least to someone who has never owned a Mac) and I also like some of the programs that come with it. Not to mention that, to my eyes, it simply looks better (which is, of course, entirely subjective). So, you are right, I should have referred to Kubuntu here. I simply used the term "Ubuntu" as Jim used that as well, and it is most well-known. Besides, I figured the GUI was really a very subjective choice or a taste question.
Downloaded Debian and Fedora
)
Downloaded Debian and Fedora this afternoon. I'll wait to load them until I can pickup 3-4 removable HDD trays. I could load each on a different 'puter but then I'd have to keep moving the damn things around all the time to use them. Space is somewhat limited in this room.
The main thing right now is finding one that works first time out of the box. Once I have one that works, and Mandriva does, then I can play around and find out how it works. If I turn it into toast, well, so what. There isn't going to be anything on there of importance anyway, and I can always reload the OS.
I will be paying closer attention to Linux discussion from now on. :)
RE: The P4 Xeons will all
)
For whatever it is worth, I have two computers. The older (and slower) one has HyperThreading turned on, so I get four CPUs reported for the two actually present:
The other one has HyperThreading turned off, and thus reports only the two physical CPUs:
Now, from the above it appears that the floating point speed is better optimized with HyperThreading while the integer speed is more evenly divided and thus gains little (or nothing) due to hyperthreading.
So, is the Einstein algorithm more sensitive to floating point speed, thus indicating that HyperThreading would help, or is it more sensitive to integer speed, this indicating that HyperThreading does little other than let you work on more work units simultaneously, but achieving roughly the same daily credits as the non-Hyper mode?
Inquiring minds want to know... Thanks!
RE: Now, from the above it
)
The difference in integer and floating point benchmark response to hyperthreading says more about the specifics of the benchmarks than about general floating point vs. integer performance. This really is an application-level characteristic.
I had a Gallatin cpu in my main PC for several years, and about a year or so ago did careful controlled testing (_not_ run it for a couple of days and watch the RAC) of hyperthreading impact on total output for the then-current SETI and Einstein aps.
For the aps of that time the general conclusions were:
1. running 2 SETIs in HT was substantially more productive than 1 not HT.
2. running 2 Einsteins of the stock ap in HT was clearly more productive than 1 not HT, but quite a lot less advantage than for SETI
3. running one SETI and 1 Einstein on the two threads was extra-special good.
4. But at a single point in a long series of code efficiency improvement by akosf, the 2 Einsteins on HT case suddenly became appreciably lower total output than running 1 in non-HT.
This last case is the only one I'm aware of for which HT actually hurt productivity, and I doubt it is the case with the current aps. But I don't recall seeing properly controlled comparisons done with the current aps.
In the balance of experience (not only my own experiments, but lots of other reports, you'd most likely get somewhat more total output in HT. But don't expect a lot.
RE: So, is the Einstein
)
Besides integer and floating point performance, there's another important factor that is not (yet?) benchmarked by BOINC: memory bandwidth.
From profiling the current apps I get the impression that both memory bandwidth and "integer performance" do play a significant role in S5R3, so it might be worthwhile to re-evaluate the effectiveness of Hyperthreading for E@H.
CU
Bikeman
Quite some time ago in this
)
Quite some time ago in this very thread, a number of people were encouraging Tullio to make the leap of faith and plug in a PIII Katmai (up to 600MHz) in place of his 400MHz PII Deschutes. Tullio declined, which of course is fair enough, and the matter essentially ended there. One of the participants in the discussion was Thunder whose parting comment included the following.
The comment was directed at Tullio but included a reference to the fact that I'd never come across a 600MHz Katmai and the highest frequency that I'd actually come across was 550MHz. As Thunder actually possessed a pair of those rare 600MHz beasts and I didn't, his final comment was quite understandable :).
At a local IT auction recently, there were some interesting servers, mixed in with some older ones. There was a compaq DL380 G2 which had dual Tualatin 1400 processors that I was keen to buy. As sometimes happens, the auctioneer, knowing there will be interest in the better spec offerings and little interest in the dinosaurs, will often combine lots and ask for a single price for all.
So, I ended up winning the DL380 for quite a cheap price anyway, with this other old compaq thrown in, essentially for free. When I got them home and had a look, I found that the old dinosaur actually contained a pair of 600MHz Katmais, 512MB RAM and an internal (but not connected and not documented) 73G SCSI drive along with the two 9G drives that were listed. Everything works fine and so I now no longer need Thunder to risk pornography charges with his two little beauties :). Here is its results list and as you can see it's no worldbeater but I couldn't just not give it the opportunity to crunch just because of a bit of electricity, could I? :). For comparison, here is the results list for the DL380 which you can see is way more than twice as fast.
I was interested in reading a bit of history about Katmais in general and the 600MHz version in particular so I did a bit of googling and came up with a local site which has some amazing information on cpus, main boards, hard drives, etc of that era and earlier - right back to the year dot with all manufacturers included. The stuff is very readable and goes into quite a lot of very interesting detail. It's very easy to navigate and quite addictive if you start reading :). There is nothing on recent cpu history - the information essentially ends with the Athlon XP 1800+ in October 2001. The oldest entry is for the Intel 4004 released in 1971. If anyone is interested in cpu history, check out the pages listed in the cpu index. The stuff on Katmais (as well as other cpus of that period) is in the link called "mainstream 1999". Alinator will probably be very interested in this page because the opening entry is for the AMD K6-2/550 and the opening sentence is, "A dog of a chip." :).
Cheers,
Gary.
RE: RE: A dual Tualatin
)
Bikeman, a belated congratulations on your choice of the excellent dual Tualatin 1400 as your "legacy crunching farm" mainstay :). My own particular favourite is the Compaq DL380 G2 of which I've now acquired quite a number. In fact I bought two more yesterday, one with 2G RAM which cost 65EUR and the second with 4G RAM which cost 85 EUR. I also bought a pallet containing 4 old HP Netservers plus a few other bits for 3.5EUR ($AUD5). Those 4 have so far yielded 2 x 1.0G cpus, 2 x 933MHz cpus, 2 x 733MHz cpus and a total of 2G of PC133 registered server RAM. There was nobody else even interested in making a bid so my $5 bid prevailed.
If you want to upgrade your RAM, just seek out your local IT auction house and go and inspect almost any old 1999/2000/2001 vintage servers to see if any still have RAM inside them. Any with coppermine PIII processor(s) would be perfect. The more battered they look, the better as you will probably be the only person bidding :). This is exactly how the person selling to you on ebay has got their stock and you may as well cut out the middleman :).
If the noise of one or two servers becomes a problem then I guess a stack of six or so would be out of the question :). I've taken a picture of just such a stack and would like to post it here but (never having done this before) am not sure of the possible ways to do it (and what to avoid, etc). Perhaps someone might be kind enough to give me a brief run through on "How best to post images - for dummies 101" thanks :).
Cheers,
Gary.