Sunday July 23, 2006 
- Blogroll
- Adrian Cockcroft
-
Bart Smaalders
- Brendan Gregg
-
Darren
-
Jim Mauro
-
John Gardner
-
Jon Haslam
-
Joost Pronk
-
Nicolai Kosche
-
Phil Harman
-
Richard McDougall
- Links
- OpenSolaris
Today's Page Hits: 92
Sunday July 23, 2006 Here is an interesting podcast
or you can read a related article
the author Jaron Lanier doesn't think much of wikipedia - his bio is on wikipedia.
To put it in a nutshell this colourful podcast raises some really good reasons why a collection of differing points of view can be far more useful than a single source. I found this more interesting than the recent wikipedia vs britannica debate.
[ T: Wikipedia ]
To put it in a nutshell this colourful podcast raises some really good reasons why a collection of differing points of view can be far more useful than a single source. I found this more interesting than the recent wikipedia vs britannica debate.
[ T: Wikipedia ]
Friday February 17, 2006 It only been 8 days since the project was
announced Now Dave has a machine
booting.
[ T: NiagaraCMT CoolThreads Linux ]
[ T: NiagaraCMT CoolThreads Linux ]
Thursday February 16, 2006 I mentioned on the OpenSolaris
site that commercial NAS devices charge
between $1.60 to $3.00 per GigaByte without redundancy. That is 0.33 to
0.62 GigaBytes per dollar. From two web sites I got prices in
January 2006 for 30 hard
drives.
A more interesting way to measure Hard Disk value is GigaBytes per Dollar
Assuming that the GB per $ performance of hard drives can be represented as a quadratic we get
the red dotted line.
lm(formula = Size/Price ~ Size + I(Size^2), data = mytable)
Residuals:
Min 1Q Median 3Q Max
-0.36294 -0.14051 0.02323 0.13648 0.38674
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 6.514e-01 1.517e-01 4.295 0.000202 ***
Size 1.173e-02 1.169e-03 10.030 1.33e-10 ***
I(Size^2) -2.052e-05 2.030e-06 -10.107 1.13e-10 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.1907 on 27 degrees of freedom
Multiple R-Squared: 0.7923, Adjusted R-squared: 0.7769
F-statistic: 51.5 on 2 and 27 DF, p-value: 6.093e-10
Solving the quadratic we get the best price performing hard drive would be 286GB, sadly nobody makes such a drive but by eyeballing the graph the sweet spot is between 250GB to 320GB with a price performance above 2GB per dollar. Giving us a reasonable margin to build our OpenSolaris Appliance and still beat the comercial products 0.33 to 0.62 GigaBytes per dollar.
[ T: OpenSolaris Qube ]
A more interesting way to measure Hard Disk value is GigaBytes per Dollar
Assuming that the GB per $ performance of hard drives can be represented as a quadratic we get
the red dotted line.
lm(formula = Size/Price ~ Size + I(Size^2), data = mytable)
Residuals:
Min 1Q Median 3Q Max
-0.36294 -0.14051 0.02323 0.13648 0.38674
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 6.514e-01 1.517e-01 4.295 0.000202 ***
Size 1.173e-02 1.169e-03 10.030 1.33e-10 ***
I(Size^2) -2.052e-05 2.030e-06 -10.107 1.13e-10 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.1907 on 27 degrees of freedom
Multiple R-Squared: 0.7923, Adjusted R-squared: 0.7769
F-statistic: 51.5 on 2 and 27 DF, p-value: 6.093e-10
Solving the quadratic we get the best price performing hard drive would be 286GB, sadly nobody makes such a drive but by eyeballing the graph the sweet spot is between 250GB to 320GB with a price performance above 2GB per dollar. Giving us a reasonable margin to build our OpenSolaris Appliance and still beat the comercial products 0.33 to 0.62 GigaBytes per dollar.
[ T: OpenSolaris Qube ]
Tuesday February 14, 2006 Talking to folks about building a new home computer after the
initial tricky but quickly resolved problem of which CPU do you want to
use. Comes storage, I'm talking internal hard drives. There is so
much choice in the marketplace at the moment. Do you choose PATA (new
name for ATA) or SATA? What about MTBF, power consumption, noise etc.
What size is best 80GB up to 500GB all seem plausible sizes but where
is the sweet spot?
So my goal is have fun putting together an OpenSolaris high performance low cost NAS appliance and finding way to measure to see if it really is high performance and low cost.
Economics of PC Storage part 2 - Hard Disks
[ T: OpenSolaris Qube ]
So my goal is have fun putting together an OpenSolaris high performance low cost NAS appliance and finding way to measure to see if it really is high performance and low cost.
Economics of PC Storage part 2 - Hard Disks
[ T: OpenSolaris Qube ]
Friday February 10, 2006 This comment came from a blog I'm not sure if it was deliberate FUD or the author was just not aware that in today's world the old rules of thumb are no longer relevant.
Today almost all business applications are either multi-threaded or multi-process [firefox, apache, oracle, mysql, postgress, java, ...] So you are more interested in the performance of many threads than a single thread. The old thinking is cache misses are bad do anything to reduce cache misses build a bigger cache, keep other cores away from this cache because when we have a cache miss the processor will stall and that will reduce performance. The real problem is not cache misses but processor stalls. The T1 has 4 hardware threads per core so when we have a cache miss it can work on another thread, the processor doesn't stall. So does the T1 have more cache misses than other processors – maybe, does it reduce the performance of the T1 – No. More theory here more practice here.
[ T: NiagaraCMT
CoolThreads CPU
]
Thursday February 09, 2006
Wednesday February 08, 2006 A new piece of hardware appeared on a local street light pole.
It looks to be part of google's plan to setup free WiFi. I went looking for a google wifi faq but couldn't find one, but I did find this blog. The first questions that come to mind is will it be better than DSL or cable, what bandwidth is available both up and down. What are the economics of this for google and for the city. Will people throw away their mobile phones and use skype devices while in silicon valley?
[ T: WiFi ]
It looks to be part of google's plan to setup free WiFi. I went looking for a google wifi faq but couldn't find one, but I did find this blog. The first questions that come to mind is will it be better than DSL or cable, what bandwidth is available both up and down. What are the economics of this for google and for the city. Will people throw away their mobile phones and use skype devices while in silicon valley?
[ T: WiFi ]
Tuesday February 07, 2006 Well I don't speak for AMD, Intel or Sun for that matter but I have
some comments regarding the questions here.
I'm sure it possible to make a lower electrical power Ultrasparc with less cores or slower clock or make a higher compute power version with even more cores or faster clock, If you wanted to model or make one yourself the chip design tools and specs to do it are here.
While The UltraSparc T1 has 8 cores and 4 threads per core giving 32 hardware threads and no other commercially available chip comes close to it. It is more important to think of the whole system it will used in like I blogged before here, So once you see the final product you can see the reasons why the other choices were made.
If your interested in the AMD space we've used the same design criteria, and as Marc Andreessen discussed here (the numbers are here) its the most cost efficient solution.
[ T: OpenSolaris AMD Intel Linux Solaris ]
I'm sure it possible to make a lower electrical power Ultrasparc with less cores or slower clock or make a higher compute power version with even more cores or faster clock, If you wanted to model or make one yourself the chip design tools and specs to do it are here.
While The UltraSparc T1 has 8 cores and 4 threads per core giving 32 hardware threads and no other commercially available chip comes close to it. It is more important to think of the whole system it will used in like I blogged before here, So once you see the final product you can see the reasons why the other choices were made.
If your interested in the AMD space we've used the same design criteria, and as Marc Andreessen discussed here (the numbers are here) its the most cost efficient solution.
[ T: OpenSolaris AMD Intel Linux Solaris ]
Wednesday January 25, 2006 Well the OpenSolaris Appliances community is beginning to take shape. Now the fun part is deciding what it is :) A few wish lists are starting to appear in blogs and on the discussion pages. We have ideas around Qube or Media PC replacement. To Me the media PC is not that interesting without first creating the “Qube” with 2005 technology.
My requirements for OpenSolaris Appliances 1.0
NAS server SMB and NFS
High performance and low cost
My requirements for OpenSolaris Appliances 2.0
NAS server SMB, CIFS, NFS
DHCP Server
Firewall
Name Server DNS and LDAP
Print Server
Nice to have in OpenSolaris Appliances 2.0
Web Server
Mail server
Simple & easy management
Easy interoperability with other home network devices
So for Me the problem I want to solve first is sharing and securing of several hundred GB of data, this is the key area where I think OpenSolaris can outperform all other appliances. Once We have that done then all the other features should just be software upgrades.
[ T: OpenSolaris Qube Solaris Appliances ]
Wednesday January 11, 2006 I've just been googling news about
Oracle and Sun after today's Sun/Oracle Town Hall meeting at Oracle
Headquarters, a mini Oracle World keynote was my first impression -
lights/music - If you've been to an Oracle World keynote you know what
I mean. Then Scott and Larry on stage for an hour, 10
minutes of jokes mostly
around acquisitions, a 40 minute presentation, 10 minutes of questions.
Its interesting to see how a one hour event gets compressed into a paragraph or two by different news organizations and the points that they decide to focus on. If you read 5 or 6 articles you'll get a good overview of what when on.
Most of the news articles picked up on the Oracle preferred Sun, then Oracle preferred Linux, now Oracle prefers Sun again theme. You could just as easily interpret history as Oracle has always preferred Sun except for the time when the only easy way to use cheap x86/x64 machines was with Linux, when given the choice between cheap x64 boxes running Solaris or Linux Oracle choose Solaris.
There was one phone question "So what does this mean for DBAs?" which didn't get the follow up deserved
We could follow this train of thought - going forward you will be able to purchase a Sun server with Oracle pre-installed. This gives us an opportunity to do some configuration of the server while at the factory so you don't have to. There are some easy setup decisions since we know you are going to be running oracle you will need a dba group and an oracle user. We don't have to modify
I suppose what I really want to know is what setup would most System Administrations and DBAs agree on so we can make them the default. Less customization means less configuration issues.
Its interesting to see how a one hour event gets compressed into a paragraph or two by different news organizations and the points that they decide to focus on. If you read 5 or 6 articles you'll get a good overview of what when on.
Most of the news articles picked up on the Oracle preferred Sun, then Oracle preferred Linux, now Oracle prefers Sun again theme. You could just as easily interpret history as Oracle has always preferred Sun except for the time when the only easy way to use cheap x86/x64 machines was with Linux, when given the choice between cheap x64 boxes running Solaris or Linux Oracle choose Solaris.
There was one phone question "So what does this mean for DBAs?" which didn't get the follow up deserved
We could follow this train of thought - going forward you will be able to purchase a Sun server with Oracle pre-installed. This gives us an opportunity to do some configuration of the server while at the factory so you don't have to. There are some easy setup decisions since we know you are going to be running oracle you will need a dba group and an oracle user. We don't have to modify
/etc/system
since in Solaris 10 we can configure shared memory, semaphores on the
fly in /etc/project we could put oracle in the FX
scheduling class, use the fair share scheduler, configure a zone(s)
for oracle etc, all of these things will improve performance or ease of
use, we can do more but where do we stop? I suppose what I really want to know is what setup would most System Administrations and DBAs agree on so we can make them the default. Less customization means less configuration issues.
[ T: Solaris Oracle ]
Monday January 02, 2006 I just saw this here
Note:
- Many DBAs feel that if the data is already contained within the buffer cache the query should be efficient. This could not be further from the truth. Retrieving more data than needed, even from the buffer cache, requires CPU cycles and interprocess IO. Generally speaking, the cost of physical IO is not 10,000 times more expensive. It actually is in the neighborhood of 67 times and actually almost zero if the data is stored in the UNIX buffer cache.
One must always be careful when using the UNIX buffer cache with Oracle
The default Unix behavior is to comply with the POSIX standard for reading and writing files
So the default file system behavior is not optimal for oracle, since oracle can manage its' file accesses e.g. not reading a block while that same block is being written. The extra layer of protection that POSIX gives is not needed. You can improve your IO concurrency in Solaris by adding the directio mount option to the database partitions in /etc/vfstab, or you could set the
You may now want to increase the size of the buffer cache since oracle is now bypassing the buffer cache and some of that memory can now allocated directly for oracle buffers.
[ T: Oracle Solaris ]
Note:
- Many DBAs feel that if the data is already contained within the buffer cache the query should be efficient. This could not be further from the truth. Retrieving more data than needed, even from the buffer cache, requires CPU cycles and interprocess IO. Generally speaking, the cost of physical IO is not 10,000 times more expensive. It actually is in the neighborhood of 67 times and actually almost zero if the data is stored in the UNIX buffer cache.
One must always be careful when using the UNIX buffer cache with Oracle
The default Unix behavior is to comply with the POSIX standard for reading and writing files
Read-Write Locks and Attributes
Read-write locks (also known as readers-writer locks) allow a thread to exclusively lock some shared data while updating that data, or allow any number of threads to have simultaneous read-only access to the data.
So the default file system behavior is not optimal for oracle, since oracle can manage its' file accesses e.g. not reading a block while that same block is being written. The extra layer of protection that POSIX gives is not needed. You can improve your IO concurrency in Solaris by adding the directio mount option to the database partitions in /etc/vfstab, or you could set the
Oracle
Parameter FILESYTEMIO_OPTIONS to SETALLYou may now want to increase the size of the buffer cache since oracle is now bypassing the buffer cache and some of that memory can now allocated directly for oracle buffers.
[ T: Oracle Solaris ]
Tuesday December 20, 2005 It's finally official you can read about it here,
page 8 has Oracle's description of a processor for licensing
calculations.
So to summarize Oracle Enterprise license is $40000 or ($60000 with
RAC) per processor.
When a chip has multiple cores the rules change
A dual Xeon server 2 * 1 = 2 oracle licenses = $80000
[ T: NiagaraCMT Solaris Oracle ]
When a chip has multiple cores the rules change
- UltraSparc T1 core factor .25
- AMD or Intel core factor .5
- Everthing else core factor .75
A dual Xeon server 2 * 1 = 2 oracle licenses = $80000
[ T: NiagaraCMT Solaris Oracle ]
Monday December 19, 2005 NOP for technorati
Friday December 16, 2005 I just got a Dr Dobbs Update email, The first
article that caught my eye was Thoughts on Language Design
by Guy Steele.
In a nutshell a programming language dictate how we think about
problems and how we solve them. Most programming today is built around
the single thread of control with binary choices if-then-else
etc. The article raises some good ideas around rethinking this programming model.
My vote for a completely different model would be this.
My vote for a completely different model would be this.
Thursday December 15, 2005 I've been meaning to look at Ruby / Ruby on Rails for some time
now, well ever since I heard there was a DTrace
provider for it. After
seeing that RoR version 1.0 had been released that made now a good time to
take a look.
What is RoR 1.0
I followed the install instructions it all went very smoothly all the gnu tools required ship with Solaris 10 and OpenSolaris in /usr/sfw/bin (make sure it's in your PATH)
Now it's time to see what RoR can do!
[ T: Ruby OpenSolaris Solaris ]
What is RoR 1.0
- Ruby 1.8.2 an interpreted language - a 3.6M tarball compile
yourself
- RubyGems 0.8.11 - ruby package manager (written in Ruby) - 160K tarball
- Rails - web application development framework (written in Ruby picked up from http://www.rubyonrails.org and installed on your machine)
I followed the install instructions it all went very smoothly all the gnu tools required ship with Solaris 10 and OpenSolaris in /usr/sfw/bin (make sure it's in your PATH)
Now it's time to see what RoR can do!
[ T: Ruby OpenSolaris Solaris ]