Alan Hargreaves' Blog

The ramblings of an Australian SaND TSC* Principal Field Technologist

Wow, even HP is advertising the T1 (Niagara) based boxes

I was recently pointed at The Real Story about Sun’s Niagara.

You have to believe you are on a good thing when a competitor feels they have to talk about your product.

It’s an interesting read, they get some things right, put others into the wrong context and actually generate a little FUD in the way of unfounded speculation, but that’s par for the course for marketing type folks. I’m an engineer, I try to read past that kind of thing. Let’s have a look at what they have to say.

November 2005 – Sun’s upcoming Niagara processor is a departure from the traditional SPARC designs embodied by Sun’s UltraSPARC and Fujitsu’s SPARC64. In fact, Sun has characterized Niagara’s approach as “radical”.

So far, so good. It is a radical design. Thank you for recognising this.

Sun has stated that Niagara will be binary compliant with previous SPARC designs but this fact does not tell the entire story regarding how well a SPARC binary compatible program will run on Niagara. It’s not enough to simply run – the program must also run well to be of actual value. Significant software optimization may be required to ensure that software will work well with Sun’s Niagara.

In fact, to be called SPARC, requires this binary compatibility.

As with any hardware, the best benchmark is the application. You should choose your hardware in the full knowledge of whether or not it is appropriate to what you are going to put on it.

As with any new hardware, optimisation will always (well ignoring the sins committed in the name of optimsation) improve things, given that you know the strengths and weaknesses of teh machine.

Anyway, let’s go into the Facts

Fact 1: Sun’s Niagara chip is made up of individual cores that have much slower single thread performance when compared to the higher performing cores of the Intel® Xeon[TM], Itanium® , AMD Opteron® or even current UltraSPARC processors.

Applications that require high single thread performance or do not perform well in a threaded environment will perform better on traditional processor models like the current Intel or AMD based servers. Sun’s CIO has said that customers would be “disappointed” in Niagara’s single thread performance when compared to AMD Opteron.

Absolutely correct. The strength of machines based on this chip are in the multi-threaded/multi-processed realm, thank you for pointing it out.

Machines built around the T1 are not designed for the kind of load that HP is alluding to here. The kind of load that these machines will excel at are those that are multithreaded or, or for that matter, multi-processed (eg apache). This machine is aimed at the edge of the network. If you need to run a load that does not fit this machine, then common sense would dictate that you choose a machine more suited. Sun sells machines with our traditional UltraSPARC and some kick-ass Opteron based boxes.

Fact 2: It takes more than binary compatibility for previous software to work well. Sun has admitted that software must be optimized to work well within Sun’s Niagara architecture.

For example, Solaris/SPARC applications that scale to just 1 or 2 threads and rely on improvements in single stream processor core for performance gains are likely to run considerably slower on Niagara without any software optimization.

That’s just common sense. You will note that we actually tell our customers things like this. Let’s have a look at the reference made here.

Today’s software will work on CMT chips, but must be optimized to work well, Yen said. “For example, internally in our system software, there are places where we may have limited the parallelism only to four simultaneous threads or eight,” he said. “Now, knowing that even one processor can support 32 simultaneous threads, certainly we’d like to further enhance the parallelism.”

The T1 based boxes released today run at 1.0GHz, so yes, the applications that run on a smaller number of threads, might run slower. Might? Most of the “edge of the net” type applications are not cpu bound, they spend a lot of their time waiting on disk and network I/O. It may even be that you are not going to notice a significant speed difference, as cpu is not necessarily your bottleneck. If cpu is your bottleneck (and you can’t parallelise the application), then you should probably choose a different platform.

Fact 3: To fully exploit Sun’s Niagara systems, developers may have to change how applications are architected. Sun has stated that Niagara changes the minimum application scalability demands from 1-4 threads to 32 concurrent threads.

Server applications which worked well on entry level systems which scaled to 1-4 threads may have to be reworked to scale properly to 32 threads. Will ISVs be willing to rearchitect a program that works well today on the lion share of entry level servers in use, just so it can perform better on Niagara? How many ISV’s can support yet another platform – yet alone one with very special needs?

Again, thank you for pointing out a strength. We actually offer a platform that will allow developers to make use of much better parallelism along with a shared cache.

Let’s have a look at the references for this statement.

They are both from Richard McDougal (one of the authors of the Solaris Internals book and a performance guru).

The first is titled CMT is coming: Is your application ready?. In this article Richard references the work that a number of major ISVs are undertaking to get even better performance on the T1 platform. He also notes that the Association for Computing Machinery (ACM) is running a special feature on the impact of CMT on software. These articles also look like great reading.

The second reference is of one of the papers that Richard submitted to the aformentioned feature. In this paper Richard talks about the opportunities provided by chip multiprocessing. Another great read.

OK, now we get into the HP sales pitch.

Be careful — there is a waterfall of change lurking around the bend.

Sun’s Niagara approach will require a new paradigm for software development. Sun’s highly touted binary compatibility with Niagara doesn’t tell the real story when applications may have to be optimized or even rearchitected to work well.

I’ve already spoken about this. In order to be called UltraSPARC, binary compatibility is a given. I like to hope that our customers actually think about what kind of machine would run a proposed load well before committing to a purchase.

So, if you are a Sun customer, you are heading for decision time. Sun’s roadmap is definitely taking you in a direction requiring you to change.

Utter rubbish.

The T2000 is a box designed for a particlar market, just because we have released this one does not mean that we do not continue to develop for the other areas. The roadmap for our other servers is looking pretty good. We have some pretty amazing Opteron based hardware now availble (The x4100 and x4200 look damned impressive) and we continue to provide out highly scalable SMP boxes, as well as working with Fujitsu for our next generation of these boxes, and the further development of Rock.

Unlike many of our competitors, Sun takes our backwards compatibility very seriously.

At this fork in the river remember that you have other options, including a move away from Sun to HP.

During your evaluation please consider HP for industry standard options ranging from HP ProLiant to Integrity and NonStop servers. The time is right to make a move.

Of course, it is always the decision of the customer as to which hardware is appropriate for their workload. It would be remiss of HP’s marketing not to make this statement. After all, they are trying to sell HP equipment.


I’d like to thank HP for pointing their customers to a whole host of articles that discuss the strengths of our new hardware.

One thing that does disappoint me though is that the article completely fails to mention the other big strength of the new platform. The power budget. As a former System Admin (16 years before I came to Sun a bit over 6 years ago), I like the idea of a cooler machine room that draws a lot less power.

Technorati tags: , ,


It occurs to me that the list of references praising T1/Niagara may
disappear from the page. In case that happens, I have listed them

  1. Sun Microsystems, FAQ,
    , #12:

    (for some reason this is listed twice)
  3. Sun Microsystems presentation,
    , page 39
  4. “Sun CIO hints at Niagara’s future”. See
  5. CNET, “Sun burnishes next-gen Sparc chips” (


    “CMT is coming: Is your application ready?” Oct, 31, 2005
  7. Sun Microsystems, Richard McDougall, “Extreme Software Scaling”, Ref:

Written by Alan

December 6, 2005 at 6:37 pm

Posted in General

5 Responses

Subscribe to comments with RSS.

  1. Who are they to talk about drastic unnecessary change when they went ballistic with Itanium and killing off HP/UX?


    December 6, 2005 at 7:03 pm

  2. Interesting 🙂

    Rayson (via proxy)

    December 6, 2005 at 8:03 pm

  3. All that from a company that refers to their ‘line doubling’ technology in their HDTV’s as “Wobulation”. Talk about FUD.
    Interesting writeup.


    December 6, 2005 at 8:15 pm

  4. This struck me from their writeup:
    “Applications that require high single thread performance or do not perform well in a threaded environment will perform better on traditional processor models like the current Intel or AMD based servers. Sun’s CIO has said that customers would be “disappointed” in Niagara’s single thread performance when compared to AMD Opteron.”
    Honestly, if you were looking for single thread performance, especially compared to Opteron, you weren’t using SPARC anyway.
    The real question for me is: I’m looking at replacing a V480 (two 900MHz USIII chips). Will the T2000 be a good upgrade for the appserver workloads on this machine (the workload includes some SPARC only software that isn’t availble for x86, so that rules out the x4200)? My guess is yes, and I’m waiting on a quote and availability.
    OT/PS: How’s the brewing going Alan? I have a yummy all-grain English Pale Ale in the secondary fermenter right now, made with homegrown Goldings hops too… 🙂

    Sean Ross

    December 7, 2005 at 9:07 am

  5. Having had a rather frank discussion with HP about the PA-RISC/Itanium decision point they are forcing, I’m surprised that the words “pot”, “kettle” and “black” weren’t present here somewhere 🙂

    Chris Saunderson

    January 15, 2006 at 2:18 pm

Comments are closed.

%d bloggers like this: