This blog forms part of the http://performanceGuru.com website which focuses on performance tuning UNIX-like Operating Systems (including Linux).
Monday, December 19, 2005
Redbooks: Simplifying Storage...
have released a new Redbook about Simplifying Storage
. The abstract describes it as:
This IBM Redbook introduces Infrastructure Simplification. Infrastructure Simplification is the methodology of analyzing the complete enterprise: business processes, workflow environment end to end, and IT for simplification. This analysis yields opportunities to save you time and money and eliminate unnecessary complexity that impedes the flow of information. This IBM Redbook discusses Storage Infrastructure Simplification and demonstrates multiple ways that IBM TotalStorage and Infrastructure Simplification can help you reduce complexity, save time and money, and release the flow of information in your business.
Sunday, December 18, 2005
Redbook: Migrating TO NFS Version 4...
have released a good (but lengthy) RedBook about migrating to NFS version 4
. A brief introduction:
The goal of this IBM Redbook is to provide a technical planning reference for IT organizations considering an implementation of Network File System version 4 Protocol (NFSv4) on IBM AIX 5L, either as part of a new installation or as part of a migration from Andrew File System (AFS) or Distributed File Service (DFS).
This book includes sample migrations that can be used as a road map for existing installations of AFS and DFS.
Although ther is mention of AIX most of the content is fairly generic and of help for any platform.
Saturday, December 17, 2005
Threading: Two Good Articles In Technology Review...
The MIT Technology Review has two good articles covering the emergence of threads and core in processors:
Friday, December 16, 2005
Debugging: Signals As A Debugging Tool...
Good article on signals
and their role in debugging. Although the articel focuses on Linux the principles are generally applicable, a summary from the article:
By focusing on the analysis of data captured using signal handlers, you can speed up the most time-consuming part of debugging: finding the bug. This article gives a background on Linux signals with examples specifically tested on PPC Linux, then goes on to show how to design your handlers to output information that lets you quickly home in on failed portions of code.
Performance: Tuning AJAX...
is running an article
about performace tuning AJAX, a good intorduction ot the subject...
Hardware: Details Emerge Of Intel's Forthcoming IO Accelerator...
LWN are carrying an article
that describes the forthcoming Intel IO acceleration architecture based on a patch that Intel has submitted to LKML...
Linux: Zones By Any Other Name...
The OpenVZ Project
has announced its first stable release
which is essentially a zone like patch to Linux. After the initial announcement Andrew Morton requested a high level overview of the patch and it looked like:
On the high-level the system looks like patched Linux Kernel with a
number of user space tools. The kernel itself boots on a usual Linux distribution like RHEL4 and works as usual. But there are extensions which allow to create a new VPS context.
User space OpenVZ tools use these extensions to do the following, e.g. on VPS start:
- turn on and configure quota on VPS file system subtree.
- chroot to this filesystem tree.
- create a UBC context with configured resource limits/guarantees.
- create a VPS context and exec init in this newly created environment.
- newly spawned init executes VPS initscripts as if it was a usual Linux box which has switched power on.
Sounds very similar to zones and is a kernel namespace and resource management framework...
Hardware: Good Posting About Database Scaling On Niagara...
Read a good post
on database scaling on Niagara...
Linux: Mutual Exclusion Primitives...
Jonathan Corbet has dropped his slidedeck
from FOSS.in where he presented on mutual exclusion in the Linux kernel...
Virtualisation: VMWare Architectural Considerations Paper...
Stumbled across this VMWare paper
which looks at various architectural considerations of virtualisation - an interesting read...
Thursday, December 15, 2005
Hardware: Good Post On Determining Utilisation In A CMT World....
Found a good blog posting
on determining utilisation under Niagara. The technique is generally applicable and can be used to look at the efficiency/utilisation of code on x86 platforms, will work out how to do this under Linux using oprofile if I get some spare cycles over Christmas...
Tuesday, December 13, 2005
Hardware: Intel Ready AMD Killer By 2008!
The Register reports
are preparing a hypertransport killer called CSI for 2008, now if only AMD will sit on their hands for two years...
Solaris: Creating x86 Boot Environments...
Sun have released a new blueprint
detailing the x86 boot process. From the introduction:
This article explains the mechanics of the boot process on the Solaris Operating System for x86 platforms so that you understand what is needed to create a customized CD/DVD. It discusses both the hard disk and CD/DVD boot processes, and points out the differences between the two.
If your involved in creating builds, etc for Solaris x86 this is a worthwhile read.
Grid: Patterns For Enterprise Grids...
IBM have released a new RedBook entitled "Patterns For Enterprise Grids"
from the introduction:
The target audience for this redbook are IT architects, consultants, software engineers with a need to use grid computing as a building block to the solution of architectural problems.
In their everyday work, those professionals need to evaluate a business problem and build a solution to solve it. They normally begin by gathering requirements related to the problem, designing a first outline of the solution and taking into consideration any special requirements that must be part of the final solution. After this step, they start the design of the actual solution, which can be comprised of one or more applications, each one requiring its own infrastructure in order to run.
Every time they can reuse the same set of solutions, devised from the experience they have acquired, the next engagement is simplified, reducing time and costs and increasing the levels of client satisfaction. Capturing, categorizing and providing access to the knowledge gained from each engagement into a repository of information can be beneficial to the overall professional community.
Patterns are great vehicles to capture components with a high degree of commonality among engagements and to express their interrelationships. Although most enterprise grid engagements are typically deployed with solutions that could be categorized as "one-of-a-kind", there is enough information gathered today to allow us to devise a set of common components among them and to derive enterprise grid patterns. The proposed patterns are based on grid solutions designed for enterprise clients over the past couple of years and are therefore representative of the current use of grid technologies in the enterprise today. They may not address all emerging grid technologies or be representative of research grids.
You can use this book as a helping guide for your grid solution design and we also expect that your experience and your feedback may be applied in the improvement of this work.
If like the rest of us, you are slowing moving towards the Grid then this is probably a worthwhile read...
Monday, December 12, 2005
Solaris: Consolidating x86 Servers Onto T1000 Servers...
Sun have published a new blueprint
covering the consolidation of x86 web servers onto T1000 servers.
Solaris: Consolidating On The T2000 Server...
Sun has release a new blueprint
covering consolidation of Sun Store onto the T2000 platform excellent paper that provides an insight into how Sun Architects moved their own applications onto the T2000 platform...
Solaris: Optimising Applications For Niagara...
Not too surprisingly Sun has released a new blueprint about tuning for Niagara
. Looks very good including how to identify candidate applications with cpustat, etc...
Linux: Interview With RedHat CTO...
A short article featuring an interview with Brian Stevens,
who is CTO at RedHat. It mainly covers their push into the virutlisation arena and the forthcoming stateless Linux, which seems ot be Linux with the configuration management information dragged from their newly released directory server...
Hardware: Q & A With AMD General Manager...
DigiTimes.com recently had an opportunity to discuss AMD’s approach
to microprocessor design with Dr. Raghuram Tupuri, general manager, Microprocessor Solutions Sector – Design Engineering, AMD...
Development: Cross Platform Threaded Memory Allocator...
With my current focus on multi threading and its supporting technologies I came across libhoard
. This library claim to outperform libumem, although there is no substantial data to back up this claim. Will look into threading benchmarks and try some benchmarks on the T2000 we are getting in at work to see if these claims hold up...
Virtualisation: Using VMWare With Oracle RAC..
Virtualisation: Financial Services Roundup...
There's a nice round up
of the use of virtualisation in financial services companies in BTN...
Devon Releases An Update For dtrace On FreeBSD...
Devon has posted an update
on his attempts to port dtrace to FreeBSD...
Development: Using dbx To Debug MT Applications...
There's a useful and timely blog posting
on using dbx to debug multi-threaded programs, there's no excuse for not trying Sun's toolchain now that its free...
Sunday, December 11, 2005
Development: Using & Hacking Subclipse...
Development: Good ACM Article On Threading...
As threading seems to be flavour of the month I would recommend this article
in ACM Queue
. The article looks at the locking complexities that arise from preemptive scheduling of threads and introduces a stateless threading library that can be used to avoid the use of mutex structures and other complex locking primitives.
Got some ideas going about Xen, I/O spaces and simplifying the Operating System...
Friday, December 09, 2005
Pervasive Conrtributes Dtrace Probes For PostgreSQL...
From the article:
Pervasive Software Inc., a global value leader in data infrastructure software, announced it plans to contribute source code for DTrace probes to the PostgreSQL community. DTrace probes provide a superior interface for monitoring the PostgreSQL engine and allow for more effective tuning. Pervasive also formally welcomed Sun Microsystems to the PostgreSQL community and applauded new support of PostgreSQL through the Solaris 10 Operating System.
Tuesday, December 06, 2005
Does dtrace Have A New Friend?
When at the latest NC event in London today I attended a breakout session with Marc Tremblay and after all the usual questions at the end, I asked a simple one that interested me:
"You have obviously spent some time characterising workloads to understand them for your Niagra work - what tools did you use for this, were they standard Solaris tools?"
The answer contained the usual, yes we used previous generations of hardware, etc. Then suddenly the word dtrace is mentioned and I cynically expect a further plug of dtrace, however Marc mentions an extension to dtrace which is called dprofile, this allows for the studying of data structures through the software stack and into the hardware, sort of tying up the CPU counters and the dtrace profiling path - even more observability!
Had a quick scrounge around and do not see any indication of the tool anywhere...
The Sun Has Got His Hat On And He's Coming Out To Play...
Have just got back from the Network Computing Presentation in London. First wow, its great to see Sun back in the game, big time. I had the pleasure of watching Jonathan Schwartz extols a social vision for Sun (someone's been reading "The Dream Society" from the Copenhagen Institute of Future Studies) where Sun will not only excel technically but will save the planet in the process - yes it was hyped but there was a lot of smart stuff in there that made both business sense and means we might have a planet left in ten years.
A few sections to highlight the content for you:The Big One
Sun will Open Source the RPL from Niagara - welcome to the world of Open Source hardware. This is a smart move and allowed Jonathan to paint the dream that Africa will hire a fab and produce SPARC chips for local consumption and local growth. The business vision is the smaller chunk of the larger pie story and in terms of competitive advantage - RedHat have no hardware to Open Source and who cares if Microsoft Open Source their mice!Core Wars
The center piece was the T2000 server featuring Niagara. Wow - the price points are down because Sun can now sell chips that have two failed cores as a 6 core Niagara allowing them to keep almost everything the produce for retail - no more expensive SPARC chips, they just beat the volume price point advantage with the usable volume price point advantage.
So that's the start, Niagara has been carefully architected as the first generation of chips that start to share components, in the olden days we called this SOC (system on a chip), the 8 cores share an FPU, great for most workloads, crap for Monte Carlo simulations (but most people run mainstream workloads - I only know what one is because I work at an Investment Bank).
The movement of the memory controllers on board, the memory access latency and the number of threads have all been architected for the most prevalent workloads, loosely called "the Internet" by Sun but the deep dive with Marc Tremblay that I attended seemed to stack up and "throughput computing" became the natural bed-fellow of "the network is the computer".
Other interesting features are the Java offloading and the crypto offloading - both cool. The driver was heat and power though - 7.5Kw give you 960 threads in a rack, so you should be able to chop down the number of racks somewhat, as the throughput paradigm should keep them busy.We Are A Phone Company
Someone loves the mobile phone industry and Sun are moving to the same charging models. Having "given everything away" they now must transition to a service model. The T1000/T2000 will be the first servers that cost less if you buy a service contract, you know like that mobile phone you own. Its actually 'free' if you buy it with a contract but its $400 if you don't. Well servers are capex and one-off costs, if Sun can gain annual service contracts on the back of free hardware then that's their model, if you want a server without a service contract then you can pay for it. Interesting approach, will be very interesting to see how it effects those companies such as RedHat and Microsoft that have no hardware to give you...Conclusions
Aagh it was like the Sun we used to know and love - welcome back - we missed you. I sat with a member of my account staff and he said to me "we did this too late...", I corrected him and proposed he should have said "we should have done this earlier...", if I have any kind of negatives from what I saw, that would be the sum of it, not the worse criticism ever.
Nice Solaris Article on NewsForge...
is running a nice piece on the OpenSolaris distributions and community
. Worth a read, one of the more positive articles I have read in this area...
Xensource Announce The Availability Of Xen 3.0...
In a detailed press release Xensource
have announced the release of Xen 3.0 which has the following features:
* 32-way SMP vServers
* Hot plug CPU's
* Dynamic Relocation of vServers
* Support for PAE in 32-bit vServers
* Support for 1TB of memory in vServers
* Support for Trusted Platform Modules
* Support for x86, x86-64, IA-64 & PowerPC
The next stage in the evolution of Xen is testing, hardening and performance tuning of the core, according to one of Xen's founders.
Monday, December 05, 2005
Comments On Moore's Law & The Future Of Processors...
Greg Papadopoulos' recent blog entry
discusses the misrepresentation of Moore's law and how it will continue in the future. He basically rings in the end of microprocessors (and other discreet components) and predicts that the future will involve microsystems - sounds good and makes an excellent read...
Sunday, December 04, 2005
Hardware: Intel's Roadmap Revealed...
has an article about Intel's rumoured roadmap
. Its a reasonable read and it obvious that the increased competition from AMD
has refocused Intel. A summary from the article:
Intel was surprisingly talkative when it came to future technologies and products this year. As a result, most of the technical audience is up to date regarding the upcoming micro architecture based on the 65 nm Merom design. We discovered that all of these announcements are the top of a hot iceberg only, because the chip firm intends to deliver almost 20 new processor designs within the next eight quarters; all for the sole purpose of dominating the desktop, mobile and enterprise segments.
Life must be tough for Intel currently, in the "Big Iron" chip market the Itanium is starting to look completely irrelevant compared to the SPARC offerings from Sun. Its interesting to compare the content of the Intel article, which seems to be about correcting a flawed vision, with this article from Sun
which seems to be more concerned with continued vision rather than corrected, from the article:
"In the meantime, what I keep telling people is: To truly exploit the value of multicore and multithread, you have to start from scratch. That means a brand new core, a brand new pipeline, and that takes easily four years," Tremblay adds.
"So, fortunately, we started four years ago, through a variety of projects, and we'll see the fruits of that for the next several years, while you're not going to see anything new from the competition until probably the 45-nanometer generation, which is probably another two years away. That gives us a huge window of opportunity."
Anyway that's a quick round-up on chips for now. I have been quite quiet lately, due to a new project I have been working on which is connected (sort of) to this space, so I will be back soon with the results of my research into this new area.