True, pc's are cheap, but thin clients are cheaper still. Having looked at such a system recently, I can say that replacing 25 pc's with new ones or replacing them with 28 terminals and a 4GB dual Xeon 2.8GHz server is about equally expensive.
But after a few years the savings kick in: you won't have to replace failing hd's, power supplies, cdroms, floppy drives and almost no memory. That's savings right there. Add to that the fact that terminals live longer than pc's. You'll have to replace the replacing pc long before you have to replace a terminal.
About the only things you need to replace with thin clients is the server and the monitors, keyboards and mice (the last 3 of which you'll have to do anyway, and the first is vastly cheaper that replacing those pcs).
Now factor in the ease of administration and see savings spiral up even more.
Well, they could handle 3.8 MB/s quite easily:) Having Gbit access to the internet has it's perks, I suppose...
Re:scsi vs. ide: from someone who knows
on
IDE RAID Examined
·
· Score: 1
I've used both as well. While I have no quarrel with your other statements, I must comment on:
Hot swap. Does anyone here know of a hot-swap IDE raid solution? I think not.
I have one running a bonnie right now. 8 disks, 4 channels, connected to the host via LVD SCSI, but the disks are IDE, and they're hot swappable.
I agree with you on speed. I couldn't believe just how fast a 15k rpm SCSI drive really is!
But the whole issue of reliability of individual disks strikes me as less than top priority for a RAID system. You care about the RAID system as a whole, not individual disks. The reliability in that area comes from RAID (5) itself.
(On a side note: I've got a SCSI RAID that's almost 4 years old (7 9GB 10k rpm Seagate Cheetas) and that IDE RAID, almost 1 year old, with conservatively specced 80GB 7200 rpm drives. Guess who's the clear winner by a factor of 2? IDE! It even scores better in seeks! Granted, old SCSI system, but it can't even outperform a single 80GB, 5400 rpm Maxtor drive on any test!)
Well, let's hope that anyone reading this thread doesn't get confused about all this.
I'd be very interested to learn about the unfortunate occasion sparking your ire. And I hope we may agree on your remarks being...outdated, if nothing else;-)
Don't take this wrong, but I saw your comments about RH as (unintentional) FUD, so many people these days like to bash RH for fun without seeming to realise the enormous amount of work they give back to the community.
Because you can't upgrade the kernel from the standard kernel tree without patching the added/modified functionality in. IIRC there isn't another distro which does this.
You must explain to me how I did just that with RH 6.2, 7.0, 7.1, 7.2, 7.3 and 8.0 then. I'm dying to know why I'm the exception. Yes, I used standard kernels (with some added patches, but none from RH).
Yeah, I've started playing with Debian.
I was talking about rpm-based distros and apt...but I agree with your view on debian "GNU/linux".
You must be either joking or trolling with your comment about RH.
How can tweaking a kernel lock you into a distro?
Why don't you install apps from source if you don't like rpms? But I'm curious, how do you uninstall if the 'make install' drops files all over the filesystems? Lots of software doesn't include a 'make uninstall' afaik.
Ever heard of apt? Yes, it works with rpms (and damn fine too, I might add).
I agree with you about MSI boards. When they were hot, I bought 6 Athlon 550's based on them. Out of those 6, 5 were unusable because of instability problems. Having replaced the boards with ASUS ones, these machines are once again usable and stable.
I will NEVER buy MSI boards again (and my supplier won't sell MSI boards as they caused him way too much trouble). I guess "5 times burned, 6 times shy":-)
While I'm an avid AMD supporter, I have to disagree with you about stability. I realise I may be just unlucky, but I've had quite a few bad mb boards for AMD desktops and my experiences with it as a server left me with a sour taste initially.
I have a Tyan Thunder K7 based dual MP 1.2 server with a Broadcom based 3com gbit nic. Apparently, if I understand the comments on the lkml, this board at least seems to reorder some pci reads/write (or something). This "confuses the hardware no end", resulting in an unstable system (going down at least once a week).
The tg3 driver had to be modified to adjust for this hardware bug. There's stability for you...
Now it runs rock-solid, but I expected better from Tyan. So the reputation for instability may have some thruth to it.
But if you do it that way and one of the server areas gets devastated you lose half your capacity. You still lose out from an operational point of view.
It all comes down to: what do you do when you _don't_ have the money to duplicate everything? Sure, I'd like to have everything quadrupled and in multiple buildings, but with university budgets that's not going to happen any time soon (i.e. 100 years =)
I agree completely with your second paragraph, btw.
You never had _any_ hardware or driver bugs? You're one lucky person!;-)
Seeing your dual athlon nfs server go down with a panic because of a bug in hardware + a vulnerable tg3 driver (gigabit nic, broadcom chipset) is not pretty or fun, I assure you.
Not that that counts against the kernel itself of course, but it *wasn't fun*...
Yeah, you could do that, but like a countryman of mine (apparently) remarked, there's a shortage of about 700 million euros in real estate (both maintenance and new) in the higher education echelons.
Spread most of that among 14 universities and you can imagine the space shortages... On top of that: server rooms aren't exactly as cheap as offices.
And if you've got a really expensive piece of equipment that you simply can't split up, what do you do?
But basically, if they're _anything_ like the other Dutch universities, it's the latter. We're doing a pretty thourough job of dismantling our knowledge infrastructure.
Try doing that on a university-wide multi-gigabit capable network on the budget of the average Dutch university. Our universities aren't like M$ in cash, you know. I know, I'm an admin at one myself.
I just hope they're well insured....poor colleagues...
On the upside: they may get a squeaky-clean start when this blows over:-)
Breaking 6 figures is not 'nearly impossible' for single users; I'm about to do so myself (90000+ now). But then, I'm a sysadmin with about 100 pc's doing seti at night, when it's quiet (nice 19, only when load is less than 1.2). I've been doing this since mid 1999. No cheating (former astronomy student here) but plain hard work (on the pc's part, that is).
You can't tell how many cpu's are behind any users' stats (though I agree 5000WU/day is suspicious to say the least). BTW, I'm also a member of seti@nl team and I have my suspicions about the top dogs too.
Linux does support HT. I run a 2GHz Xeon box with HT as a firewall (on 2Gbit lines:) and it runs an smp kernel, using both cores just fine. Blistering fast machine it is, too! (Well, whaddaya expect, with 15000rpm SCSI disks, 2xGbit ethernet, etc, etc:)
I seem to remember a leaked bench about a Hammer that ran at something like 800MHz but outperformed a 2+GHz P4. If that was true, and considering implications are that Hammer will debut at 2GHz or more, it will surely give P4 a run for Intel's money...
Slashdot Mirror
True, pc's are cheap, but thin clients are cheaper still. Having looked at such a system recently, I can say that replacing 25 pc's with new ones or replacing them with 28 terminals and a 4GB dual Xeon 2.8GHz server is about equally expensive.
But after a few years the savings kick in: you won't have to replace failing hd's, power supplies, cdroms, floppy drives and almost no memory. That's savings right there. Add to that the fact that terminals live longer than pc's. You'll have to replace the replacing pc long before you have to replace a terminal.
About the only things you need to replace with thin clients is the server and the monitors, keyboards and mice (the last 3 of which you'll have to do anyway, and the first is vastly cheaper that replacing those pcs).
Now factor in the ease of administration and see savings spiral up even more.
I invite you to take a seat on our 1Gbit link :-)
Daisy-chaining all the way!
Oh boy, do I agree with you! If I had mod points I'd mod you up as insightful in a jiffie :-)
Moderators....?
Well, they could handle 3.8 MB/s quite easily :) Having Gbit access to the internet has it's perks, I suppose...
I've used both as well. While I have no quarrel with your other statements, I must comment on:
Hot swap. Does anyone here know of a hot-swap IDE raid solution? I think not.
I have one running a bonnie right now. 8 disks, 4 channels, connected to the host via LVD SCSI, but the disks are IDE, and they're hot swappable.
I agree with you on speed. I couldn't believe just how fast a 15k rpm SCSI drive really is!
But the whole issue of reliability of individual disks strikes me as less than top priority for a RAID system. You care about the RAID system as a whole, not individual disks. The reliability in that area comes from RAID (5) itself.
(On a side note: I've got a SCSI RAID that's almost 4 years old (7 9GB 10k rpm Seagate Cheetas) and that IDE RAID, almost 1 year old, with conservatively specced 80GB 7200 rpm drives. Guess who's the clear winner by a factor of 2? IDE! It even scores better in seeks! Granted, old SCSI system, but it can't even outperform a single 80GB, 5400 rpm Maxtor drive on any test!)
Well, let's hope that anyone reading this thread doesn't get confused about all this.
;-)
I'd be very interested to learn about the unfortunate occasion sparking your ire. And I hope we may agree on your remarks being...outdated, if nothing else
Don't take this wrong, but I saw your comments about RH as (unintentional) FUD, so many people these days like to bash RH for fun without seeming to realise the enormous amount of work they give back to the community.
Because you can't upgrade the kernel from the standard kernel tree without patching the added/modified functionality in. IIRC there isn't another distro which does this.
You must explain to me how I did just that with RH 6.2, 7.0, 7.1, 7.2, 7.3 and 8.0 then. I'm dying to know why I'm the exception. Yes, I used standard kernels (with some added patches, but none from RH).
Yeah, I've started playing with Debian.
I was talking about rpm-based distros and apt...but I agree with your view on debian "GNU/linux".
Of course I know that! But look at the price diff for same clockspeed chips! In cases like that, ALWAYS go for the Athlon.
You must be either joking or trolling with your comment about RH.
How can tweaking a kernel lock you into a distro?
Why don't you install apps from source if you don't like rpms? But I'm curious, how do you uninstall if the 'make install' drops files all over the filesystems? Lots of software doesn't include a 'make uninstall' afaik.
Ever heard of apt? Yes, it works with rpms (and damn fine too, I might add).
Wow, XP 2100 and P4 1.7 run at the same clock - look at the diff in price (well, the XP runs 33MHz faster, even)
I agree with you about MSI boards. When they were hot, I bought 6 Athlon 550's based on them. Out of those 6, 5 were unusable because of instability problems. Having replaced the boards with ASUS ones, these machines are once again usable and stable.
:-)
I will NEVER buy MSI boards again (and my supplier won't sell MSI boards as they caused him way too much trouble). I guess "5 times burned, 6 times shy"
While I'm an avid AMD supporter, I have to disagree with you about stability. I realise I may be just unlucky, but I've had quite a few bad mb boards for AMD desktops and my experiences with it as a server left me with a sour taste initially.
I have a Tyan Thunder K7 based dual MP 1.2 server with a Broadcom based 3com gbit nic. Apparently, if I understand the comments on the lkml, this board at least seems to reorder some pci reads/write (or something). This "confuses the hardware no end", resulting in an unstable system (going down at least once a week).
The tg3 driver had to be modified to adjust for this hardware bug. There's stability for you...
Now it runs rock-solid, but I expected better from Tyan. So the reputation for instability may have some thruth to it.
Why don't you get a water-based cooling kit then? That should keep it running like greased ice :-)
And it would be very quiet as well.
But if you do it that way and one of the server areas gets devastated you lose half your capacity. You still lose out from an operational point of view.
It all comes down to: what do you do when you _don't_ have the money to duplicate everything? Sure, I'd like to have everything quadrupled and in multiple buildings, but with university budgets that's not going to happen any time soon (i.e. 100 years =)
I agree completely with your second paragraph, btw.
You never had _any_ hardware or driver bugs? You're one lucky person! ;-)
Seeing your dual athlon nfs server go down with a panic because of a bug in hardware + a vulnerable tg3 driver (gigabit nic, broadcom chipset) is not pretty or fun, I assure you.
Not that that counts against the kernel itself of course, but it *wasn't fun*...
Yeah, you could do that, but like a countryman of mine (apparently) remarked, there's a shortage of about 700 million euros in real estate (both maintenance and new) in the higher education echelons.
Spread most of that among 14 universities and you can imagine the space shortages...
On top of that: server rooms aren't exactly as cheap as offices.
And if you've got a really expensive piece of equipment that you simply can't split up, what do you do?
That's what I thought as well :-)
But basically, if they're _anything_ like the other Dutch universities, it's the latter. We're doing a pretty thourough job of dismantling our knowledge infrastructure.
Try doing that on a university-wide multi-gigabit capable network on the budget of the average Dutch university. Our universities aren't like M$ in cash, you know. I know, I'm an admin at one myself.
:-)
I just hope they're well insured....poor colleagues...
On the upside: they may get a squeaky-clean start when this blows over
Under the name Polinux perhaps? :-)
(stupid 20 second rule, grmbl, grmbl)
Breaking 6 figures is not 'nearly impossible' for single users; I'm about to do so myself (90000+ now). But then, I'm a sysadmin with about 100 pc's doing seti at night, when it's quiet (nice 19, only when load is less than 1.2). I've been doing this since mid 1999. No cheating (former astronomy student here) but plain hard work (on the pc's part, that is).
You can't tell how many cpu's are behind any users' stats (though I agree 5000WU/day is suspicious to say the least). BTW, I'm also a member of seti@nl team and I have my suspicions about the top dogs too.
Linux does support HT. I run a 2GHz Xeon box with HT as a firewall (on 2Gbit lines :) and it runs an smp kernel, using both cores just fine. Blistering fast machine it is, too! (Well, whaddaya expect, with 15000rpm SCSI disks, 2xGbit ethernet, etc, etc :)
Amen. Damn, I wish I had the money to buy all the researchers I support good machines!
Ah well, universities will never be rich enough for that (not the one *I* work at, anyway).
Well, that's probably because your gfx card couldn't figure out that weird 1200x1600 resolution then ;-)
I seem to remember a leaked bench about a Hammer that ran at something like 800MHz but outperformed a 2+GHz P4. If that was true, and considering implications are that Hammer will debut at 2GHz or more, it will surely give P4 a run for Intel's money...
How about the Dutch government from about 1600 to about 1800? That was a republican form of government (although not everyone could vote of course).