Anandtech reviews Sun’s T2000 Coolthreads server, and concludes: “At first sight, Sun has won the performance/watt battle for now, but it cannot rest on its laurels. Low voltage versions of the Xeon ‘Woodcrest’ and Opteron might be able to come very close to the performance/Watt levels that the T1 offers. We also can’t shake the feeling that the number of applications, which will really exhibit the kind of exceptional performance that Sun’s own heavily optimised benchmarks show, will be quite limited. Last, but certainly not least, Sun’s solid engineering has impressed us. Sun’s meticulous attention to detail resulted in a sturdy, well-polished machine.” More benchmarks here.
Neither benchmark mentions disk configuration, which I find curious. In fact the only mention of disks was Colm’s comments about using the noatime option and could use a faster disk setup. Did either system ship with more than one disk, because the system I have has 2 73 GB disks that I have mirrored using Solaris Volume Manager. The output of fstyp -v on the /usr partition on the machine I have shows the maxcontig as 16, and that can be changed with tunefs.
If either site was using SVM, they could tune the disk performance by setting maxphys, md_maxphys in /etc/system and /kernel/drv/sd.conf and see a performance boost.
I wonder. Did or did they not receive a free server for reviewing purposes, with the possibility to keep it? For free. It might not really affect a relatively large site such as Anandtech, but a smaller hardware review website might actually overlook some bad things about the server he gets to keep for free. After all, it’s worth serveral thousand $. Nice bribe.
Edited 2006-03-24 17:35
Here is a comment from Jonathan Schwartz’s blog:
“And just to reiterate – once you receive the system, run your own benchmarks – post them publicly (positive or negative – points earned for thorough and complete), and our marketing team will decide who gets to keep their Niagara systems for free. Free.”
I am sure the marketing department has a couple of engineers on standby to read blog entries just to make sure before they hand out the T2000. Hell, I would like one to replace the E3000 (space heater) I have at home, but I don’t have a blog!
Of course. That is what they say to the public. They would never admit doing such a thing. But I suppose they are not giving away expensive devices just to get negative publicity.
One minute you question if they got it for free (you have no idea…)
“I wonder. Did or did they not receive a free server for reviewing purposes, with the possibility to keep it? For free. It might not really affect a relatively large site such as Anandtech, but a smaller hardware review website might actually overlook some bad things about the server he gets to keep for free. After all, it’s worth serveral thousand $.”
The next:
“Nice bribe. ”
How is it a bribe, if it may not have even been given to them? Not only that, but do you really think a site as big as Anandtech is going to write positive reviews for a 13,000$ server? You’d have to be extremely silly to believe that. They’ve got hardware falling out of their pockets when they wake up in the morning heh.
——————————————————
It’s a good article, a little light on the details though. Not typical Anandtech work as far as I’m concerned. I’ll wait till I see part 2 to pass judgement, but I wasn’t all to impressed with part 1. I certainly wouldn’t base my decisions on a T2000 on this review. Too much is left to guess about, and the choice of software/etc leaves a lot to be desired. I think inexperience (self-admitted) with Solaris hurt the scores a bit too. Oh well, here’s to hoping part 2 is better researched!
I was pretty dissapointed by the lack of details, and I agree with you, the AnandTech piece is definitely not up to their normally high standards.
Well, I suppose there’s a little miscommunication here. What I meant was referring to with bribe, was not the actual bribe itself, but just the offer of a bribe. I don’t know the correct English word, but in Dutch it is called ‘steekpenning’. I.e. money you offer a politician or judge.
Btw, I wrote: “It might not really affect a relatively large site such as Anandtech, but…”
I was not referring to Anandtech or this review. I just wanted to make a generic statement about the potential implications regarding the integrity of smaller reviewers.
And Sun must be expecting that they get more positive reviews than they would otherwise. Why the hell would a capitalistic enterprise give away hardware if they not expect to benefit from it, in the long run? Right. They would never ever do it.
“Well, I suppose there’s a little miscommunication here. What I meant was referring to with bribe, was not the actual bribe itself, but just the offer of a bribe. I don’t know the correct English word, but in Dutch it is called ‘steekpenning’. I.e. money you offer a politician or judge. ”
I understand you clearly now. I still don’t feel this is the same though, as they are not GIVING anybody a T2000 for free, or even implying if you write a good review, you’ll get one. They are loaning one for evaluation, with hopes that the majority will purchase. They might give one or two away here and there, but that’s a *promotion*, not a bribe. Unless of course, all the buy-one-get-one-free deals you see at local stores are bribery to you. Bribery to me signifies something unethical (ie. offering money to a politician to get your agenda on his list.)
“I was not referring to Anandtech or this review. I just wanted to make a generic statement about the potential implications regarding the integrity of smaller reviewers. ”
Ok, I admit I did see this disclaimer, it just seemed your post was aimed at them. I am clear on your intention now. I don’t feel that Sun is bribing smaller reviewers though, otherwise you’d see a plethora of very positive glowing reviews of the T2000. I’ve actually only seen one or two, and none were very good.
“And Sun must be expecting that they get more positive reviews than they would otherwise. Why the hell would a capitalistic enterprise give away hardware if they not expect to benefit from it, in the long run? Right. They would never ever do it.”
I think Sun expects most people who do the trial of the T2000 to end up purchasing one (or ten.) I’m sure they’d like positive reviews, and I’m sure they’ll take them when they see them, and maybe even give away a free machine or two. The campaign isn’t targetted at reviewers though, it’s targetted at companies. Most won’t bother publishing some blog entry, they are just going to use and abuse the hell out of the machine, and end up liking it and buying it. That’s the point.
Oh, and about the quote. Here is the updated revision.
“We were also serious about the following: if you write a blog that fairly assesses the machine’s performance (positively or negatively), send us a pointer, we’re likely to let you keep the machine. (And before you ask, the marketing team makes the decision about what qualifies for the promotion, not I – although I know they love drama, charts, and compelling competitive analyses.)”
That absolutely makes it clear their intention.
http://blogs.sun.com/roller/page/jonathan?entry=niagara_benchmarks
I highly doubt that anandtech would keep a T2000 for anything but a novelty. The last time they posted information on their infrastructure it was running on multiple Opteron servers running Windows. They’ve also custom develped most of their own software (ads, forums (through an affiliated company), backend, etc.) so the cost of porting that from Windows/SQL to Solaris and the T2000 threaded architecture would be prohibitive.
I do know wikipedia got a free server and from what I’ve heard its been doing some really nice work there.
It is shame for SUN to have a terrible 20 cycles for MUL instruction in year 2006! I working with u-sparcII-core-based STBs and i still wonder why these devices has a laughable 22 cycle MUL and i486-class memory controller.
Also i wonder why this particular 250 mhz chip can offer only a little fraction of integer performance of comparable ARM core.
So while SUN deserves a respect for pushing it’s own CPU line, their engineers still seems to unable to produce a good low-level cpu design.
Where the hell did you get that information from? its not a 250Mhz chip, and as for the memory controller <shakes head> go get SPARC documentation.
I mean 250mhz sparc chip inside the STB i’m working with, not the T1 =)
BTW some friends of my friend are already working with T1 servers on telecom market here in Russia.
Sorry, my english is bad. It is a known bug, err… feature ^_^
Edited 2006-03-24 21:46
A while ago, I did an informal benchmark of an UltraSPARC II, and it performed about on par clock-for-clock as a Pentium III in integer and anywhere from on-par to _double_ the performance in floating point. This was for things like gzip, ogg encoding, etc. Outside of this, I think the old Ultra workstations did well mainly due to thei UPA bus, which destroyed anything PCs had at the time. It’s too bad Sun didn’t keep that lead in the late 1990s.
I believe call .mul is no longer the perferred way for doing multiplication since v8. Though it is a really neat solution and solves the only thing AFAIK that kept MIPS from being truly RISC/living up to it’s own name. When I was building my own CPU (for school) I definitely wanted to use that approach to keep the longest path down, instead I had to complicate things by making it mutli-cyclic and making my state transition table ugl[y|ier].
Heh, mips lots its own name with MIPS3 (i.e. the R4000)… MIPS actually stood for Microprocessor withough Interlocked Pipeline Stages, and they introduced interlocks after the R3000 (I don’t know if the R6000 had interlocked stages though). 🙂