The IT Business

The TLB bug almost ruined the AMD Phenom franchise. But it looks like if Intel has a similar problem with the Nehalem. According to Fudzilla the Nehalem is haunted by a TLB bug too and refers to the errata document of this processor. It looks like history repeating.

At page 16 of the the Intel Core i7 Processor Extreme Edition Series and Intel® Core™ i7 Processor they write:

AAJ1. MCi_Status Overflow Bit May Be Incorrectly Set on a Single Instance
of a DTLB Error
Problem:A single Data Translation Look Aside Buffer (DTLB) error can incorrectly set the Overflow (bit [62]) in the MCi_Status register. A DTLB error is indicated by MCA error code (bits [15:0]) appearing as binary value, 000x 0000 0001 0100, in the MCi_Status register.
Implication: Due to this erratum, the Overflow bit in the MCi_Status register may not be an accurate
indication of multiple occurrences of DTLB errors. There is no other impact to normal
processor functionality.

But in the clarification for errata AAJ1 on page 37 of the same document they state:

In rare instances, improper TLB invalidation may result in unpredictable system behavior, such as system hangs or incorrect data. Developers of operating systems should take this documentation into account when designing TLB invalidation algorithms. For the processors affected, Intel has provided a recommended update to system and BIOS vendors to incorporate into their BIOS to resolve this issue.

Doesn´t sound like There is no other impact to normal processor functionality. I don´t think, that this bug has the scale of the AMB TLB bug, but i think customers are a little bit sensitive today when they hear TLB and bug in the same sentence.

Update: AAJ42 on Page 27 is interesting, too:

Under certain conditions when C6 and two logical processors on the same core are enabled on a processor, an instruction fetch occurring after a logical processor exits from C6 may incorrectly use the translation lookaside buffer (TLB) address mapping belonging to the other logical processor in the processor core.

Funny discussion about boot-up times of pSeries systems:

Hey, I manage 7 p595 squadrons and they are all different. For the most part, the startup time depends on the amount of I/O you have installed. Two of mine are 3 frame p595's with 64 hba, 64 nic, about 20 internal HDD, dual power systems and I think 10 d20 I/O drawers. These servers take about 90 minutes just to initialise the hardware, not including starting any LPAR's. So tell your friend that this is normal and take a good book and a really large coffee with him each time he needs a restart

Hmm, it will be interesting to see where the acquisition of Transitive by IBM will leave Apple (Rosetta is a Transitive technology) or SGI (their ability to execute Irix/MIPS code on their Linux based systems). At least the last system can be viewed as a competition to IBM.

Well, as Transitive delivers one of the foundations of IBMs PowerVM technology, i expected, that they would acquire Transitive at some point in time. Today IBM announced such a move according to this press release: IBM Announces Plans to Acquire Transitive.

The Transitive stuff allows IBM to run Linux x86 code in a Power LPAR. Okay, i don´t really know if it´s sensible idea to use an expensive Power system to virtualize a cheap x86 system. But they need something like that as Linux on Power doesn´t really have a large support by ISVs.

Just did an short look on the new Top500 list:

• Wow ...A Constellation is an own architecture type in the Top 500 list
• The are just 9 Itanium systems left in the list.
• The total sum of Gigaflop per second increased from 11705491,49 on 2415203 to 16958600,19 on 3121579 procs. Thus the yield per CPU increased from 4,85 to 5,43 Gigaflops per proc.
• By the way: Ten year ago, the list sumed up to 29367,61 gigaflops. That´s an performance increas of over 577 times in 10 years.
• When compare the actual increases with Moores law in it´s first derivation (Moore talked about transistor count, not performance), we see some interesting developments.Moores law suggests in it´s first derivation an increase in perfomance very 18 months. Over the 10,5 years period Moore´s law suggested an increase by the factor of 128. So the performance increase of the top 500 clusters was higher than Moores law suggest. The Top500 lists suggests a doubled performance in a little over 12 month instead on the long run.
• But this doubling is just correct when you look at the whole cluster. Let´s look at the CPU level and the list of 06/2007. In this list every CPU yielded 4.05 Gigaflop. The number of 06/2008 ist just increased by the factor of 1.3 in 18 month.
• Back to cluster level: In 06/2007 the TOP500 yielded 6974873 Gigaflop/second. In the following 18 months the the list showed an increase of 2.43 (as already mentioned, the total computing power of the Nov 2008 list was 16958600 GF/s)

Reuters writes in Fujitsu to buy Siemens stake in PC joint venture:

Japanese electronics conglomerate Fujitsu Ltd will buy Siemens AG's 50 percent stake in their computer joint venture for 450 million euros ($567 million), aiming to boost its presence

They just talk about the PC business in this article, so there is no new information about the FSC server business.

There is something really strange in the numbers for the EMC Symmetrix DMX4. A colleague gave me this hint. Just look on the specification sheet. They state on page 2 that this system can house up to 2400 disk (albeit they restrict that in the footnote by writing that more than 1920 drives are just available on RPQ). But that´s not the point.

Just look at the numbers at system capacity on page 3: They state on this whitepaper, that you can plug 1920 1 TB hard disks into the system. Okay .... let´s take the example for RAID 5 with 7+1. You could cut 240 of this RAIDs out of 1920 disks (1920/8). 240 multiplied with 7 Terabyte is 1680 Terabyte. But the maximum system capacity is stated by EMC with 585.91 Terabyte. WTF ?!?!? Where is the missing Petabyte? The first approx. 670 1 TB disks in your DMX4 gives you capacity, the other approx 1250 disks gives just heat dissipation and some nice blinking lights.

I don´t really think, that there is an error in my calculations . Why? When you calculate this for 73 GB you get 122 Terabyte ... almost the system capacity number stated on the specification sheet(you have to deduct some of the storage for internal usage, so it´s slightly less).

I smell a architectural weakness here. Something like "640 Kilobyte is enough for everyone" decision ... just on a larger scale

The third quarter of HP is the one we call the fourth quarter at Sun. So it covers the time from April to June. There is one interesting chart in the earnings presentation of HP - the revenue for Enterprise Servers ans Storage :

The next earnings announcement will be an interesting one. In the next earning announcement they can´t say that they grew year over year in this area. Nevertheless, there is an clear downward trend in this chart. Perhaps this is the reason why they bought EDS. They need that business to cover the decreasing revenue from their server branch.

There is a lot of interesting stuff in the 3Q 2008 Earnings Presentation of IBM (you find the pdf here). Let´s have a look at page 12: Legacy System i is 82% down, i assume the 7% up for converged System p comes from customers buying pSeries instead of iSeries. xSeries is down 18% and storage 3%. IBM Microelectronics is down 27% year to year. In total the System&Technology business lost 10% (at constant currency it´s even 11%).

So yould easily say: Software and Services rescued IBM in the last quarter. But not everything is looking bright there. The signings for Global Services for short-term is up, but the signings for long term deals decreased 19% for strategic outsourcing, 17% down for Global Technology Service and 16% down for Global business services.

By the way: SMB is mentioned often as the big strength of IBM. They make 4,7 billion $ dollars out of this segement. But you have to take into consideration, that IBM has a strong retail equipment department (point-of-sales terminals for example).

I already wrote about the latest benchmarketing trick of IBM in the last blog article, but this article was in german, so i repeat this in english. There was an bold statement in a recent press anouncement of IBM - "IBM Builds on Industry-Leading UNIX Portfolio With New Servers, Software":

The Power 560 can save up companies up to $840,000 and 80-percent in energy by consolidating 13 Sun Fire V490 servers on a single Power 560 server with PowerVM, as compared to consolidating the same number on four Sun SPARC Enterprise M5000 servers with dynamic system domains."

I´ve asked myself, how they get to such numbers. This number of servers couldn´t based on performance. We don´t need 4 M5000 just to substitute 13 V490. But after thinking about after reading the article in the Computerwoche i found out what´s the trick of this comparision. The trick is a cheap one ... even for IBM marketing.

You can partition an M5000 in up to 4 domains. When you just want to consolidate 13 servers, you obviously need 4 systems. This comparision doesn´t compare the compute power of the M5000 with the compute power of the p560. It compares two different virtualisation technologies. So the even the choice of 13 V490 is a really perfidious one. Twelve systems to consolidate would lead to 3 M5000, 13 systems lead to 4 because you have one domain too few. But that´s not the point: You won´t consolidate 13 V490 by using domains. You would use Solaris Containers (perhaps in conjunction with Solaris 9 Containers) for this tasks. By using this Containers you would need only one system, too. And you would need less processing power for it, as Container are a more efficient virtualisation technology in comparison to *PARS.

By the way: The answer "one system" is false for both systems. Independently from the system architecture, virtualisation technology you want at least two systems and a cluster when you consolidate 13 systems on one. Without an additional standby system you are toast in the case of a system failure or maintainance. But that´s a persistent error in every benchmarketing comparision of IBM.

Ein Kollege hat es schon ganz richtig gesagt: Dieser Artikel in der Computerwoche ist eine Denial-of-Service-Attacke auf das Sun-Marketing - IBM zündet Server-Feuerwerk. Denn was da erzaehlt wird, ist auch fuer IBM-Marketing-Verhältnisse echt billig. So billig, das selbst ich ein paar Minuten brauchte, um dahinter zu kommen, was es damit auf sich hat (und ich beschäftige mich häufiger damit) ...

Der "IBM Power 560 Express"-Server verbindet laut Hersteller die Power6-Prozessor-Technologie mit verbesserter Virtualisierungsfähigkeit und Energieeffizienz. Er eigne sich insbesondere als Server-Konsolidierungs-Plattform oder als Datenbank- oder Applikations-Server. Wie der IT-Riese erklärt, können nun Mittelständler durch die Konsolidierung von dreizehn "Sun Fire V490"-Servern auf einem einzigen Power-560-Server mit PowerVM-Technologie bis zu 80 Prozent der Energie sparen im Vergleich zu einer eventuellen Konsolidierung derselben Server auf vier "Sun Sparc Enterprise M5000"-Servern mit Dynamic System Domains.

Ich hätte mir hier von der Computerwoche ein wenig mehr Recherchetätigkeit erwartet, als bloss eine Pressemitteilung von IBM abzuschreiben.

1. Wird hier wieder der uralte Benchmarketing-Trick benutzt eine eine V490(eine Maschine die sich gerade in der Abkündigung befindet) gegen eine gerade angekuendige Maschine zu vergleichen.

2. Ich weiss uebrigens nicht, ob man das als Serverfeuerwerk bezeichnen sollen... eine p560 ist lediglich eine p570, die in ihrem Ausbaupotential halbiert worden ist. Also nix Neues.

3. Rechnen wir doch mal durch, wie man auf vier M5000 gekommen ist. Es geht hier um die Ablöse von 13 Rechnern. Jetzt möchte die IBM diese durch 4 M5000 abloesen. Wie kommt man auf eine solche Anzahl? Es liegt nicht an der Rechenpower, sondern an diesem kleinen Kommentar wegen der Paritionierung. Das ist recht einfach. Eine M5000 laesst sich in 4 Partitionen aufteilen. Nutzt man diese als Paritionierungstechnik, so braucht man natuerlich zur Darstellung von 13 Rechnern 4 Systeme. So ist auch die Anzahl 13 in einer aeusserst perfiden Art gewaehlt. Bei 12 Systemen, kaeme man schon mit 3 Systemen aus, wenn man denn so konsolidieren würde. Das ist eben diese Utilisation-Quatsch (20$ fuer Sun, 60% fuer IBM) nur etwas versteckter.

Dabei muss man anmerken: Diese M5000 sind weit davon ab, auch nur ansatzweise ausgelastet zu sein. Man braucht sie nur wegen der Limitierung der Domains auf 4 pro System. Das ist der ganze Trick dahinter. Macht man das mit Containern - also auf Betriebsystemebene - kommt man auch hier locker mit einem System aus. Wobei Sun wahrscheinlich dann sogar noch mit einer weniger ausgebauten Maschine auskommen wuerde, da wir den Overhead der Virtualisierung via *PARS nicht haben.

Sehr interessant ist auch, das man diese seitens IBM immer noch als Dynamic System Domains bezeichnet. Die hiessen so zu Sun Fire E10 und E20k-Zeiten. Man hat wohl bei IBM noch nicht mitbekommen, das sich die Zeit bei der Konkurrenz auch weitergedreht hat.

Und was hat es jetzt mit dem Kommentar der Denial-of-Service Attacke auf unser Marketing auf sich? Naja, jedes mal, wenn wieder ein solcher Bullshit durch unsere Presselandschaft lanciert wird, muss sich unser Marketing wieder darum kuemmen so einen Mist zu kommentieren.

Das Dumme ist: Man kann IBM dafuer nicht mal für vors Schienbein treten, den technisch ist das durchaus richtig, man kann eine andere Firma nicht für eine Darstellung einer dümmlichen Architektur verklagen.

Whenever you want to dismiss the claims of a competitor or want to set your own or prefered technology in a better light, you should do some research on your topic. Otherwise you may end up with a document that´s outright ridiculous.

I found a really strange piece of "comparison". It´s called "A comparison of virtualization features of HP-UX, Solaris, and AIX". It´s written by Mr. Ken Milberg. And i wasn´t able to stop my shaking the head in disbelive. This text reinforces my personal impression, that this author is just a hired gun to publish claims even IBM doesn´t want to make. But let´s dissect his newest blurb. You shouldn´t read it ... it´s just a really abysmal document. I´ve sacrified my time to do it for you, so don´t waste your own

The Register writes in NetApp faces Sun lawsuit loss:

NetApp was unable to comment immediately on this story. Sun's win - if it is a win - and the PTO decisions potentially turn NetApp's WAFL IP into, well, IP waffle. Sun gets a welcome PR boost to its ZFS and open source credentials, leaving NetApp with a bloody nose and a 22-patent IT infringement case to deal with. Oh, and the economy is going down the tubes too. It isn't so sunny in Sunnyvale right now.

In the last few days the stock of IBM was hit quite hard. Two days with roundabout 5 percent decrease. You can wonder a little bit about this, but there is an interesting article in the "Between the lines" blog: What’s really ailing Big Blue shares? Hint: IBM is part bank. Larry Dignan writes:

Big Blue’s financing unit, which leases hardware and finances projects, is big enough that the Securities and Exchange Commission put IBM on the “do not short” list. This list is designed to get shorts–folks that bet against stocks–off the backs of financial services companies long enough to raise capital or at least survive.

Thus IBM is possibly hit by two trucks: By economic slowdown as a computer ... sorry ... consulting company and by the credit crisis as a financial company. The article is really an interesting read.

Interesting quote of the person who got the stone rolling:

Linux Foundation executive director Jim Zemlin had two words to describe the article - "sensational and regrettable."

Well ... Mr. Zemlin, this sensational article had its start in sensational quotes ...