Huawei to forge big red Itanium iron Intel has added two more server makers to the roster of those who want to build Itanium-based systems: Huawei Technologies and Inspur. That's a 50 per cent growth rate in the Itanium OEM base. At the Intel Developer Forum in Beijing this week, Kirk Skaugen, general manager of the chip maker's Data Center Group, said in his …
Do readers think customers will want a telco-certifiable OS with their Huawei server? How is an enterprise-class OS going to be supported on someone else's tin without severe risk of finger-pointing when something misbehaves (assuming that HP won't be allowed to sublicence their OS qualification processes and tools)?
Based on the article, software doesn't seem to be very important.
The same even goes for the earlier article headed "its' all about the OS software" - once you've read the headline, there's only a couple of mentions of anything other than $$$ or tin.
You might even think people were buying these boxes just for hardware bragging rights, not to actually run applications that run a (big) business.
Odd.
There is one redeeming feature on the Itanic - 5 QPI links.
So if you have enough money to do your own silicon (H does), enough engineers to design a QPI-to-Fabric interconnect (H does), enough developers to churn out bespoke code (H does) and need a platform that can do computing close to the network forwarding plane (H needs) and connect to networking gear at a speed that does not make network gurus chuckle Itanic is actually the right choice.
Would you want to use an expensive (and hot) CPU for this task or ASICs to do the line rate stuff with data management/processing done by cheaper CPU's?
I would have expected the only advantage of using Itanium would have been in the high availability OSes that are already common within telco's.
Well, you seem to be thinking with the Western corporate mindset - who can I sell my end product to in the general marketplace? - when this announcement seems to be wrapped up in a lot of Chinese ego. The Chinese government are (alledgedly) chucking wads of both cash and kudos to Chinese companies that develop solutions that are "free" of Western control. Seeing as one area the Chinese government is very sensitive over is complete control and monitoring of any form of electronic communication inside and crossing China's borders, I'd say there is plenty of financial motivation for a large-scale SMP platform for the Chinese telecoms market. And it will probably run some form of Linux (lots of info on IA64 Linux already out there for the Chinese to use), with custom-written Chinese apps (as in Western apps like MySQL rewritten by Chinese coders). Given that Itanium is a very powerful platform for Linux development, suddenly it makes more sense. Red Hat only dropped development on Itanium because us customers preferred hp-ux, but if the Chinese want full code control then hp-ux is not going to be their choice. Afterwards, I'm sure the Chinese government will be only too happy to sell the solution to other (Third World) countries, probably with plenty of backdoors built in.
You wot?
The Chinese idea of "free of Western control" for hardware is stuff like Loongson (go read about it). Carrier class hardware also needs carrier class software, none of which will be developed for Itanium for much longer. If China wanted "free of Western control" software (and who could blame them), there's always Red Flag Linux (and friends) as a start.
If they want kit to drive automation, comms, etc in their rapidly-developing infrastructure, and they probably will, IA64 is the last place they should be starting.
".....The Chinese idea of "free of Western control" for hardware is stuff like Loongson...." Yes, agreed, I too am curious as to why the Chinese have gone for a Western CPU, but I suspect that Intel have come to some comfie agreement with the Chinese government. I also suspect that the Loongson program just simply won't get the Chinese to the solution they want in the timeframe they need, hence the deal with Intel.
".....Carrier class hardware also needs carrier class software, none of which will be developed for Itanium for much longer...." Yes, which is why there is nothing but carrier grade servers in all telecoms? Oh, hold on a sec - there's loads of big UNIX in all the telecoms I can think of! Go take a look at IBM's or hp's or even Snoreacle's websites and you can see all the telecoms listed as using their SMP, non-carrier servers for those back-end systems. Then take a trip down to Fort Meade or Langley and check out what they're using for sigint and monitoring - what a surpise, not carrier grade servers! Then consider that China has a massive program to develop independent software through Linux, that happens to run very well on IA64, and maybe you'll realise your anti-Itanium prejudices just made you issue a really stupid comment.
"ASICs do not run customer apps well enough"
Well that rather depends on what the app is, a question which is so far unanswered. Perhaps you or Matt could enlighten us, as TPM doesn't seem able to.
Big Iron routers and suchlike don't need IA64, but sometimes they do benefit from special purpose silicon, and there's lots of very affordable, very high performance, remarkably easily programmable FPGA stuff around these days, which suit some classes of application (typically dedicated apps, although an FPGA setup can easily change what it's dedicated to).
What kind of application am I thinking about for FPGAs? Well pick the random example of deep packet inspection, as one might in a country like China (or even, according to Phorm, Brazil?). You might do basic deep packet inspection with something FPGA-based as front ends. What does that then need behind it that can't be done just as well with something commodity-based rather than something exotic like an IA64 box?
Afaict the only place where IA64 hardware still has any meaningful role is ultra-massive-memory ultra-massive-SMP systems. Everywhere else, AMD64 and its Intel clones are far better value for money and have far better software support in general - but may not currently offer the specific software of choice, if the specific software of choice happens to be based on HP-UX, NSK, or VMS.
So, exactly what application(s) do readers think Huawei and Inspur have in mind?
On a couple of different tracks: Readers are aware that Huawei already do Xeon-based kit are they (the Tecal ones TPM mentioned)? Huawei's Tecal servers are qualified to run Oracle Enterprise Linux. Which is likely to be more important to Huawei prospects, IA64 or Oracle? Larry thinks he knows the answer to that question. Does Matt have a different answer?
Also, obviously Intel aren't the kind of company that likes to StrongARM its big name customers into doing things they might not otherwise do (hello Dell), so I won't suggest that continued access to support on Xeon developments in the face of some real or imaginary US restrictions has been related directly or indirectly to Huawei announcing an IA64 box.
Matt said:
"there's loads of big UNIX in all the telecoms I can think of!"
Obviously; the telco sector has its operational systems and its business systems running on commercial tin as well as core infrastructure boxes which may or may not be on telco-specific tin.
Five or ten years ago, the world was different, but these days why does any of that run better on IA64 than on A N Other chip? If there's any difference in it these days, in many cases it's because of the underlying OS. NSK and VMS have no direct competitors, but there is little uniqueness in HP-UX.
"China has a massive program to develop independent software through Linux"
Indeed so. And how well is Linux doing on IA64 vs other Linux-ready hardware, with AMD64 at the high end, and ARM/MIPS elsewhere, and a bit of SPARC and Power on the commercial UNIX fringes?
"take a trip down to Fort Meade or Langley"
My local equivalent is in Cheltenham. Their approach to the subjects you mentioned when I last looked was based on lots of small but surprisingly powerful boxes. There may have been a few big boxes there too in areas I didn't know. That being said, a handful of sites does not necessarily constitute a profitable market.
"a really stupid comment."
We'll see, won't we.
What would be really really stupid would be for a vendor's organisational stovepipes and limited corporate vision to end up forcing their OS-sensitive customers to pick up the unnecessary cost of continued IA64 chip and system development, when the real differentiation and added value is in the OS software not the underlying chip hardware.
Too long? Have a nice weekend anyway :)
So, you're Huawei and you want to play in the mid-range UNIX for hardware for the Chinese market. You have several choices:
- Wait for x86 to get there.
Almost, but no cigar.
- Wait until Loongson or another home-grown CPU gets there.
Not even close yet.
- Partner with one of the big 3 (IBM, Oracle, HP) to sell their mid-range iron.
Not enough control.
- License someone else's CPU?
Sure, which of them are available with "no strings" - Itanium.
As for OS, my guess is a hardened Linux of some sort, though they could save some time and money by talking to HP. The real question is how long Oracle will stand by and get creamed by this in the China market. And when they decide they *have* to support it, how stupid they'll look.
"Wait for x86 to get there.
Almost, but no cigar."
So be specific. What is the x86 missing that the other mid range competition can offer? IA64 hardware is barely relevant in what most people might call mid range; AMD64 made sure that IA64 was going nowhere as far as "industry standard 64bit" was concerned.
"- License someone else's CPU?
Sure, which of them are available with "no strings" - Itanium"
I'm not at all sure about the "no strings" bit. What can you do with IA64 that you can't do with x86-64, and why? Export controls? munitions? etc. Huawei already have a selection of x86-64 servers, what strings are in the way? Anyway let's put that to one side.
Suppose they build themselves a nice big IA64based box (or "mid range" box).
Now what? They'll need an OS for it.
Which OS? If Linux is the answer, why not just use x86-64? That's where the apps all are, and unless there's a big opportunity that changes the current picture, the high end enterprise class apps are neither available nor supported on Linux on IA64. They are already available and qualified and supported on Huawei's existing x86-64 hardware.
HP-UX or NSK or VMS isn't the answer to the OS question because the vendor (HP not Huawei) (a) wants those OSes dead ASAP (b) the OS vendor won't provide the level of engineering support required to qualify their enterprise-class OS on non-HP hardware.
Where's the beef?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
