Slashdot Mirror
Adaptec Supporting Ultra160 On IA-64 Linux
GeorgieBoy writes: "Adaptec has announced support for Ultra160 SCSI adapters under Intel 64-bit Linux. Looks like IA-64 Linux will be pretty well supported upon Itanium's arrival." There already are SCSI adapters for the (also 64-bit) Alpha under Linux, but this move sounds like a smart one for Adaptec to tie their name to both Linux and IA-64. Other companies planning pre-emptive hardware support? Step right up, please.
GEEK Humor & satire and rubber chickens!!!@
He's correct. Most systems running linux can't have more then a few gig of ram. Whereas an HP or SUN server can have 4+ gig. For a lot of servers, a very large amount of ram is required for good perforamnce, and the amount of ram you can stick in your average x86 system is just not enough for high end servers. At work, I'm constantly doing large simulations, and this simulations would not have close to enough memory if I only had a 2 gig machine to run them on.
>>Windows NT 4.0 SP5 booting to the login screen in 4 seconds off this HD. Compared to the 30 seconds my CuMine 667 with Intel DMA/66 chipset running NT at work takes, this is a good thing.
I find it VERY unlikely that NT 4.0 on ANY platform could boot to the login screen in 4 seconds. You must have 0 services loaded, with a fresh install.
Come on, use a little imagination. Not everyone just plays games, surfs slashdot, or writes memos on their desktops.
Folks that edit digital video are going to want the fastest raids at their desktop.
cuz it's intel dood. they own you
I don't use 20+ architectures. Who gives a shit if gcc is 239x faster on some obscure outdated and unused platform? Compaq owns you. Give up. Die.
Alphas support up to 2TB thanks to Suse.
Yeah, and IA-64 is *really* going to be the desktop architecture of choice in the beginning? I think not. IA-64 will have an exclusive home in server rooms and possibly on the desks of geeks (hey, I never said *I* don't want an IA-64 machine..) for a while after it comes out. I think that we should be happy with any support that companies are willing to give.
Speaking of Win2K ... Im rolling it out tomorrow. And NO, its not my idea... apparently my MIS got (and wait for it...) a really good deal on licenses!!!!! Bwahaha... good thing im looking for a new job... Anyone need a Unix Sysadmin in Wellington NZ???? **Freakin**
wow....another piece of hardware supported! somebody give Linux a cookie!
Ultra160 SCSI! I hope the Linux drivers have target mode support so I can use them with the Linux IP-Over-SCSI drivers. I'll be able to NFS mount filesystems on a network faster then most of my current drives. :) *drool*
SCSI is not just about hard disks but there is a world out there which requires a wee bit more than IDE on steroids.
what an amazing coincidence, i must have poured at least 160 bowls of hot grits down my pants this morning alone. thank you.
Merced/Itanium is a very troubled chip tht will be expensive to manufacture, burn a huge amount of power, and come a distant second to Alpha EV68 in performance. The Merced is over 300 mm2 in 0.18 um, burns over 140 Watts, and uses a funky 1000+ pin cartridge filled with expensive custom L3 cache SRAMs a la Xeon and Pentium Pro. Performance estimates range in the 40-50 SPECint95, 70-80 SPECfp95 range at 800 MHz (ships at 733 MHz initially)
Compare that with the EV68 that is under 200 mm2 in 0.18 um, burns 65 W max at 1 GHz, will use a 587 pin flip chip package and commodity late write SRAMs for its L3, and will yield performance levels over 60 SPECint95 and 110 SPECfp95. BTW, the EV6 bus runs at up 3.2 GB/s peak compared to 2.1 GB/s peak for the Merced system interface.
Indications are that Compaq and Samsung will bring the price of EV68 down to increase sales volumes as a prelude to bringing out EV7 for high end servers. Combined with second and third generation K7 chipsets, we should see very competive Alpha offerings for Linux in the $2.5k range. OTOH, Merced and its 460GX chipset look very expensive to manufacture and Merced systems will likely not get much below $5k.
So Alpha gives higher performance, consumes much less power, has more robust compilers, and can potentially ride K7 chipset volumes to much lower system costs than Merced. I think the 64 bit platform of choice for Linux is quite obvious!
this is sick dude. They caught that guy in Florida and he is up on capital charges. You need to get some help soon.
Unfortunately for Microsoft, they are only shipping SQL Server on IA64, not the whole 'BackOffice'.
So, the scorecard for Itanium will look like:
Microsoft) - Windows 2000, Visual C++, SQL Server
Linux) - Anything that will compile.
Adaptec probably wrote that driver for Microsoft. I know they wrote the base set of drivers included in NT4.
Irrelevant news
About expensive hardware
Frequent on Slashdot
I'll have you know that Scooby snacks are high in fiber and have 12 essential minerals and vitamins. And they come in 6 yummy flavors.
Trolling for Scooby doo!
Linux on IA/64
Faa - ATA looks good on paper, but try a box under load. Seen it, and I'll take SCSI-2 over DMA/33, thanks. (Can't imagine 66 is much better.)
Yes ... its big fat and ugly... it dumbs everything down ... all the bad stuff from Win98 bundled with the inflexibility of NT4 ive grown to hate it. **freakin**
They just haven't announced the fact that they also apparently have Athlon-specific drivers for the Ultra160 SCSI chipset yet.
I have heard through one of the AMD beta testers about the raw speed that the onboard Adaptec Ultra160 on the new AMD motherboards designed for the Athlon/Athlon Ultra (this one was a test one running RAMBUS) provides. Yes, it may have custom Athlon-optimized drivers. No, you won't see them until the actual hardware ships.
Windows NT 4.0 SP5 booting to the login screen in 4 seconds off this HD. Compared to the 30 seconds my CuMine 667 with Intel DMA/66 chipset running NT at work takes, this is a good thing.
This is a test machine that had a 1.2GHz copper-based Athlon that was running a chip from a test run at Dresden. It was definately running a test Adaptec chipset. This was in January.
I'm just waiting for the announcement. I want one of these for home. Anything running optimized drivers like that with on-board integration is going to smoke whatever Intel has that will be available this year, since they have been rushing their chips out without testing them like they should.
Remember the AMD 5x86, NexGen nx586, Cyrix 5x86 and MediaGX, as well as the AMD 486/120 and 133? Intel rushed their chips out the door to beat those releases. I remember the PII/300 getting announced and showing up in decent quantities 6 months later, and don't even bring up the early Pentiums. Itanium is essentially a test chip, and I'll expect major changes between now and McKinley.
Meanwhile, Adaptec chooses to not announce a product that will be available for a reasonable price in June. Itanium is going to cost as much as the 450 Xeons with 2MB cache did, and a lot more for the higher end models. For the price of the chip alone, I'll be able to buy that Athlon running the same SCSI and actually run applications wel.
Their ftp site has kernel drivers and boot disk images for their scsi controllers. Not to mention they support FreeBSD, Linux, NetBSD, SCO Unix, Solaris, Unixware and Win2k.
ftp://ftp.tekram.com/SCSI/390X/
Only the State obtains its revenue by coercion. - Murray Rothbard
The only people excited about Itanium are the ones who think they want to run enntee on it, and they are bound for disappointment because it'll be at least a year before m$ can demo a working 64-bit enntee. Linux on IA64 will help, but it won't save them. People who want to run Linux on a 64-bit platform are already doing so and probably won't buy into the Intel hype. If Intel's 64-bit offering were substantially cheaper than others, they might claim this market anyway, in the same way and for the same reason as they claimed most of the 32-bit market. Alas, Itanium is looking at least as expensive as competing options, and the first versions are unlikely to match up in performance or scalability.
And really, hype is all it is. The hard truth here is that Intel is between two and six years behind its competitors and is trying to make up for it with slick marketing hype. While Intel has called press conferences, their traditional RISC workstation competitors have marched on, crushing Intel's new processor in performance before it's even shipped. Within the next two to three years, we can expect to see more Alpha 21264s, UltraSPARC IIIs, MIPS R14k, and Itanium. As it looks, Itanium will be the weakest of the four. Ironically, they could probably have avoided this if they hadn't made the idiotic decision to ship Itanium with an x86 compatibility mode. That extra die space could have been used to take the performance lead away from their competitors. Instead, they've simply added support for an architecture that everyone agrees should have died 10 years ago.
So let's drink one final toast to Intel. Here's to you for holding back the computing industry for 20 years. Farewell, you shall not be missed.
I'm pretty sure tux likes scooby snacks. That belly of his tells me he likes them a lot!
_damnit_
_damnit_
It's my job to freeze you. -- Logan's Run
Posting to the same story that you have moderated removes the moderation. Strangely enough he is still at +1 Offtopic though :).
:)
I do pity you however. They are gona lynch you in metamoderation
--= Isn't it surprising how badly I spell ?
I mean, the prices for IA64 systems probably
will be the same as or worse than an equiv
Alpha system, and Alpha is a much cleaner
design anyhow... What's the point in being
excited about IA64?
For every problem, there is at least one solution that is simple, neat, and wrong.
Hmm. It should also be noted that mostly nobody is going to use an IA-64 desktop computer during the first months anyway.
--Moo
--
If the only OS available for the first systems would be Linux?
:-)
It would be almost as good if W2K will be available, but has to run ix x86 compatibility mode.
I bet Redmond is pulling a whole bucketload of all nighters to prevent either of these scenarios. And we all know how much quality that provides
--
Infuriate left and right
Hmmm, while this is somewhat boring news to those of us that run FreeBSD and have had Ultra 160 support for the new adaptech drivers since January, I must say I'm impressed with their 19160 board. While crippled in software under Windows, it isn't crippled on FreeBSD and runs like a bat out of ****. I got mine white box from MicroExpress for $160.00. Great price, and very competitive with all the other cards.
BTW, the Linux driver is just a port of Justin Gibb's FreeBSD driver. Justin does good work, which is why I went for the 19160 board when I was putting together a system recently.
My understanding is that the Alpha port still exists internally. They're using it to build 64-bit Windows.
The real problem is not that NT wasn't designed to be portable -- it's that NT was only designed to be portable to 32-bit archs. Microsoft had been promising a 64-bit port of NT4 for Alpha for years, and couldn't deliver.
I agree that with 2000 running on i386 only, there's probably a huge temptation for hack programmers to shove some non-portable code in there. Hopefully Microsoft will show a bit of discipline here. (I like Sun's policy of shipping Solaris on i386 even though it's a money-loser. It ensures that they stuff is portable, and will probably save them tons of time and money in the long run.)
--
Business. Numbers. Money. People. Computer World.
I Wish gcc was as fast as ccc. But it isn't. I need fast float. Alpha has it. Compaq made it available for free to us on Linux. I am not complaining. I would guess that gcc will catch up. For now I would like to Thank Compaq for making ccc available to me on Alpha Linux for free. And, in kind, I am sure Compaq is thankfull for my purchases...
Improving gcc Is probably enticing for the hardware manufactures - to sell their architecture. If I can get a 2x improvement in performance through hardware optimization in ccc then it is in their best interest to make gcc fast as possible on their architecture. I wish the whole distro was compiled with ccc's speed.
Adults are obsolete children. - Dr. Seuss
I grabbed Compaq's ccc compiler for Linux/Alpha just 2 days ago. It was indeed 2x faster float than what gcc could give. Integer performance will be 10 - 30% better. Get the Compaq Linux Alpha compiler Here.
Adults are obsolete children. - Dr. Seuss
Well at least we got both cygnus and SGI working on compilers, sgi had in mind a release of the source so at least we can hope for 'good enough' support.
But I since we havent got the chip yet it must be quite hard to come up with some benchmarks or stuff...
De lyckliga slavarna är frihetens bittraste fiender, legalisera!!!
Yeah right. They dont even have support for their existing products under Windows 2000. You have to use the generic Microsoft Driver. Can you say crash? Pete -- So I'm being a little pessimistic..
What's a sig? Pete Brubaker
Do you have any links to IP-over-SCSI implementations? Sounds neat.
"It's tough to be bilingual when you get hit in the head."
The whole point of all this is that Intel won't make the big release until MS are ready with win2k.
-0-0-0-0-0-0-0-0-0-
Laptop006
laptop006@netexecutive.com
Vic, Australia
/* FUCK - The F-word is here so that you can grep for it */
I'm all for more FLOPS and its great that Compaq has given us their compiler. I just meant to suggest that the true value of gcc is greater than its raw performance numbers, without it its hard to imagine what the Free Software movement would have looked like. To make gcc better one has to do it in the context of making other platforms better, which is why I didn't think that contributing in a real way would be interesting to a hardware vendor. They would rather give their platform an advantage, as Compaq has done with their compiler.
You make a good point about having a whole distro compiled with a Compaq optimized compiler. That might be the carrot that causes them to submit a patch that has their improvements to gcc. Right now the linux kernel is tied to gcc and I can't count the number of times I've tried to use the vendor's optimized compiler to build some sourceware and given up in frustration and had it build with no problem with gcc. So making gnu/linux systems better overall might be what convinces Compaq its in its own interest to contribute to gcc, apparently they don't think that's the case now.
But can Compaq produce a compiler that beats gcc on 20+ architectures? Or even better, can Compaq help to improve gcc to reach that level of performance? Improving gcc is probably not as enticingly sexy to the vendors that are making noises around Free Software as improving Linux itself, so I would not expect Compaq, or the other big vendors to do that soon.
That said, of course it would be great if gcc performed better. Cygnus has ported gcc to VLIW architectures before, so I would expect that will help. Also, hopefully now that Redhat has bought Cygnus some of that IPO money will go towards engineers working on improving gcc.
Now that they have dumped the Alpha from the 2000 family, 2000 will, inevitiably, grow to be Intel 32 bit achitecture specific. So when the IA64 trundles along there will be a fearsome amount of work getting the codebase compatible. So it's going to be a bodge.
I (heart, linked to an external RAID array and a couple of CD-ROM drives) SCSI.
--
Are you saying that the average computer running Linux is a cheap Taiwanese motherboard and K6-2? That could very well be true. I've never seen a study or poll done. If people tend to avoid the x86 platform when building enterprise-class servers, it's not Linux's fault.
If we can support as much high-end server hardware as possible, I'm sure the kernel as a whole will prosper as a result of it. Linus would never accept a patch that broke every other architecture besides IA64.
I don't think your comment is flamebait, but you could have worded it more gracefully.
file:
Are yoy sure? Don't forget AMD, they push Intel do a log of thing they don't whant to do.... like bring down IA-64 in the K8 space.
It's clear to me that the advantage in the second machine is 5 disk configuration not SCSI Bus.
In my opinion it make sense to have a SCSI machine only if you need more disk that your ide channel(1 disk per channel). Take in mind that a Ide RAID + 18GBx2 disk cost less than one 9 GB SCSI disk, in Italy of course:). Look at promise site for benck(ok it's not the best source;).
More yummy hot hardware to play with at work under linux, it's really nice to see that no longer is linux fully supporting current and old hardware, but also *future* hardware, how much better can it get? =)
---
Just because you disagree doesn't make it offtopic or flamebait.
Come to think of it, if you stick a 64Bit processor in a 32 bit cluster, will it even work ?
I don't know why it shouldn't. If I'm not mistaken Beowulf clusters exist are made up of independent computers conected via network connections. As long as your code can run the software you should be able to run a Beowulf.
On the other haand, I could be just running my mouth, pretending I know something when in reality I have no idea what I'm doing.
Jay
-- polish ccs mirror
I accidentally just moderetated this down.... would someone PLEASE moderate this up....
So while we're on the SCSI/DMA issue, has anyone out there seen/played with/set up a RAID 0 DMA/66 box with fileserve performance super-tweaked? How much can you milk from it, how does it compare to SCSI solutions?
Ok, so, Question. In the 75G Deskstar Article posted yesterday, a "maximum media data rate of 444 megabits per second (Mb/s)" is quoted. So what exactly does this mean again? Doesn't this mean that there's approx a 3:1 burst transfer rate bottleneck?
Does anyone have the answer? Are the technological hurdles high? How is it that we have gigabit ethernet and yet drives can be three times as fast as their bus?
When Oligarchy strikes, it's time to be a scab.
I wonder what would happen if I stuck an itanium to the end of my home-built beowulf cluster ?
Come to think of it, if you stick a 64Bit processor in a 32 bit cluster, will it even work ?
With linux as the big-boy OS here for a change, I'll bet were gonna be seeing a lot of comercial 64Bit itanium based beowulf clusters. Just imagine how much phun you could have with one of those. Even if you only cluster 4 processors.
Now all we need is for someone to build a version of the dumb 3D engin with support for multiprocessing.
So I can play DOOM supercomputer style.
"Semper in excretum set alta variant"
There is no doubt that the limiting factor is the bandwidth limitations not the limitations of our hard drives. I mean we operate off of a T1 and 80Mbits/sec is more than fast enough for us. However, for someone who needs a high performance workstation I could see the advantages.
Nathaniel P. Wilkerson
NPS Internet Solutions, LLC
www.npsis.com
Nathaniel P. Wilkerson
www.haidacarver.com
I would be suprised if IDE ever surplants SCSI for servers, it just isn't designed for it - case in point, running MS SQL server (6.5) on NT Server 4, which of these two machines do you think performed best:
P2 350MHz, 128Mb RAM, 2 x 6 Gb IDE DMA/33 (striped)
Pentium 166MHz, 80Mb RAM, 5 x 2Gb Wide SCSI (striped w/parity)
I'll give you a clue, the faster machine with more RAM ran like a total dog whenever any major database work was done.
This is nothing new. Remember it was many YEARS after the release of the 386 MS came out with a version of Windows that used any 32bit features. For those who don't remember it was Windows 3.1 and the features were performance tweaks for virtual memory. It would still run on a 286 if you knew how to detune the setup.
It did take MS a few years to take advantage of special 386 modes. Quarterdeck's Desqview/Qemm and Qualitas' 386MAX beat them to the punch by a large margin, and made it clear to even the average DOS user that MS was shipping things that barely deserved the title "Operating System". I consider NT to be the real first attempt, since previous versions didn't do much with the capabilities of the 386 -- even Win98 doesn't do much to isolate potentially errant processes.
Nit: Windows 386 (v.2.11??) required a 386 and did have some basic DOS VM support. It didn't have DPMI support, though, so running other 386-aware programs was verboten.
Windows 386 was, of course, horrid compaired to Desqview though it did pre-date Win 3.1x.
A firewall can not protect you from yourself. Turn off what you do not need. Do not use the firewall to do your work.
Unfortunatly, most systems running linux are not for high-performance environments...
Please define "high-performance." Standing alone, this comment may appear at first glance to be flamebait.
"Wouldn't the time be better spent bring a real _server_ OS up to IA-64 spec?"
To what end? So that high-end hardware becomes useable only by "high-end" (READ: rediculously expensive) OS users? Umm...or would the time perhaps be better spent bringing high-performance features to Linux? As a sysadmin who has worked with both Linux and some of those other "real _server_ OS"'s you speak of, I'd much rather see Linux brought up to the level of Solaris and some of the other enterprise-level, high-performance OS's out there.
I'd just rather deal with Linux on a daily basis. I've learned it's a much more pleasant experience than wrangling with some of the commercial UNIX offerings.
I'm glad to see more development of this nature. Congrats to Adaptec for taking this first, small step in the right direction. Next step: Open-source drivers, open hardware and development specs.
Back then, it was the desktop, and there was no competition.
Right now, this is for servers, M$ is trying to make W2K the be-all, and there's some fierce competition. This will be a real black eye for M$ marketing. It will put paid to a lt of their boasting as to being ready for the enterprise server market. It will be DELICIOUS!
--
Infuriate left and right
I wonder if gcc is up to the task for compiling programs for IA64.
Compaq is saying that its compilers is generating code whichs runs 2 times faster than code generated by gcc.
For the Alpha, a chip for which code generation is much more easier than the IA64 (VLIW compilers are complicated beasts!).
So Linux will run first on the IA64, yes, but will it have good performances ? I'm not sure at all.
Does anyone have more informations ?
Another good question: How many months until ATA 200? The IBM 75G Deskstar article talks about ATA 100 support. Yeah, SCSI has it's place, yeah it has low CPU over head, etc, but is adaptec not getting a little worried? Can you imagine the situation if ATA transfer rates exceed SCSI? SCSI drives are targeted for server market, which makes their prices "unnaturally" high (inelastic demand) but at some point might not it become cheaper to build, say, a HD subsystem blackbox of ATA drives with it's own CPU, microkernel feeding a host through gigabit ethernet or something? If the software tools were in place, you might get a full ATA host with the disk subsystems priced at half the cost of SCSI flavor.
Take home prophecy: Linux (i.e. geeks) kill the SCSI oligarchy.
Go Geeks Go
Unfortunatly, most systems running linux are not for high-performance environments. Wouldn't the time be better spent bring a real _server_ OS up to IA-64 spec?
Desperation is a stinky cologne
It seems to me that this is just a tacit admission by a major hardware manufacturer that Linux is well ahead of Doze, Solaris and any other OSes that are being developed for IA64. Based on Microsoft's previous record, we might see a 64 bit IA64 windows sometime in 2002, and Solaris is for the time being on hold, after a rather childish slagging match between Intel and Sun with each one claiming that the other was not pulling their weight by doing a fair share of the porting effort. As a result, when Itanium arrives, there will be only one viable choice - Linux. So if you are a hardware manufacturer that would like to be able to take advantage of the new architecture - then you can wait 2 years for MS, or you can just bite the bullet (jump on the bandwagon?) now. It's a question of simple, obvious business sense.
ø`ø,,ø`ø,,ø`ø,,ø`ø,,ø`ø,,ø`ø,ø`ø
Adaptec has high bandwidth Hard drive controllers to sell on the x86 market. They believe iNTEL's claim that this will be Itanium in just a few short months. Therefore they must make this available for Itanium and support it as much as possible.
Since there is only one complete OS that's correctly capable of doing real work on Itanium today; they really have no choice. See this story for details about Turbo Linux on Itanium. This isn't just a kernel or a compiler but a full, functional Linux distribution. The closest thing anyone has to a Linux distribution ( in terms of included Apps ) is Microsoft's "Back Office suite".
--= Isn't it surprising how badly I spell ?
This is nothing new. Remember it was many YEARS after the release of the 386 MS came out with a version of Windows that used any 32bit features. For those who don't remember it was Windows 3.1 and the features were performance tweaks for virtual memory. It would still run on a 286 if you knew how to detune the setup.
All the apps for the platform were 16 bit however. It wasn't until yet more years that they actually released a 32 bit OS. It was Windows NT and though it had some 16 Bit stuff in there it qualified as a 32 bit OS even in early release.
MS will sell a 32 bit OS or even that hybrid 16/32 bit Win98 or WinME on Itanium and the market will lap it up. Perhaps not as quickly or in the volumes that Linux will enjoy but enough to be noticed.
As for Monterey. I have herd a lot of posturing from the people producing it and some talk from people who promise to ship it on boxes but I have yet to here customers who say "We want to use Monterey on Itanium to run our business". Why would they ? It's a compliantly unknown quantity. In the Jargon file it qualifies as Vaporware. I.e. Nobody has seen even an Alpha build.
--= Isn't it surprising how badly I spell ?
Even with a pair of Quantum Atlas V drives (27 MB/sec write rate), I doubt that any real performance gains will be measured running at 160 MB/sec burst rate over 80 MB/sec (U2W).
Now, when you've got a RAID 0 stripe 4 drives deep, that might show some potential for improvement.
So this one is definitely for the server room.
Those new IBM drives with up to a 16 MB cache - if you have 16 MB cache on each drive, plus 64 MB cache on the RAID controller, then the 80 MB/sec is potentially rate-limiting.
So - how many months until Ultra 4 320/m SCSI?
Paul
.
They should think about becoming more like Advansys who actually provide kernel tuning advice.
Or perhaps Symbios who have programming guides and real datasheets for much of their stuff.
free login required
Basically Adaptec should spend some of it's time thinking about the customer now, not the the customer in a year that they are trying to create.
--
dronf!