In tests that i've done on solaris boxes (a very small 2 cpu UltraSPARC III server, not the big iron) I often see 'strange' results like this from benchmarking utilities like iozone. (as in 600+MB/s read and write speeds)
In general, if the file fits into memory, read and write speeds can often exceed the theoretical limits of the underlying array. This is because the OS buffers the hell out of the I/O (unless you use something like UFS's 'forcedirectio' mount flag to avoid the OS buffers). This is good and bad. Good for throughput but bad if the buffer is allowed to grow unchecked and forces applications into swap as one tries to write a huge file that is larger than the RAM on the system. (think DVD iso's on a box with only 4G of ram).
In the case of the 2.4 vs 2.6 shootout, I'd be much more interested in seeing a full iozone graph showing block size vs file size vs performance. ( If you do this, make sure that you test for file sizes at least twice as large as the RAM on the box to see what happens when the system runs out of space for a buffer. )
All of that said, the best I've seen so far (same 2 cpu sparc box) in actual disk performance is 180MB/s (and that was with a NexSAN ATABeast that had 42 300G IDE drives). The load was spread over two 1Gb fibre channel links, so i was starting to approach the max throughput of the HBA's at that point:-)
Unfortunately, it is a bit harder than a simple zip code based lookup since zip codes can span towns and counties. It really requires a full gis based system that uses the tax boundaries (which can change at a moments notice) to figure out how much to charge. It is possible that your next door neighbor pays a different tax rate than you.
All of that said, if the states fail in their quest, they will prob just move toward a higher property tax/income rate. (in my part of the world, the local city has a budget shortfall due to the lack of tourists spending money at local stores. a property tax would have avoided the problem, although at the expense of the local population)
The University of Wisconsin has a site at http://genchem.chem.wisc.edu/demonstrations/ that contains links to dozens of demos (with raitings) for various categories. I'm sure there is something for each age/interest level there
this worked before in a chemistry demo i gave in high school. It gives an unexpected result and it is colorful to boot:-) http://jchemed.chem.wisc.edu/JCESoft/CCA/CCA2/MAIN/AMFOUNT1/CD2R1.HTM for more details
I dont think that intel ever really had a question for Sun to answer. At 800MHz, Intel's 64it chip is slow in the all important MHz rating (sun has had 900's out for a while now) and still has a few years of compiler design ahead of it before it makes any sense. And this is Sun's 3+rd generation of 64 bit chips, vs Intel's 1st.
As for HP, they helped intel build their 64 bit chip, so the PA-RISC is more or less dead.
those are list prices from store.oracle.com. University's get a huge discount.
note that standard edition is much cheaper than Enterprise Edition. (like 1/10th the cost) However, if you want to use RAC (which larry is indicating as the replacement for a single large system) then you must use EE. Also, you are not permitted to put standard edition on any system with more than 4 cpu's. Also, this is just oracle's cost, you still need to add in your own cluster software to make sure that if an instance goes down, you dont loose that data (RAC is not a 'shared nothing' cluster)
now, if you want to manually partition your data, you could get away with using standard edition on a ton of boxes. you are simply trading cost in one area for extra time and expense to implement in another The actual advantage in ones particular environment is dependant on the cost of time (students are cheap;-) and the ongoing cost of managing the extra added complexity.
for starters, you are comparing a dual cpu box with a quad cpu box. The quad's always cost more per cpu. Simple matter of the fact that it is harder to get 4 cpus' to talk together compared to only 2 (why do you think intel has yet to produce a 64 way smp server.... sun did it 5 years ago, cray did it before them.) You also have to look at things like backplane contention (are all of those cpu's on the same bus? sucks to be you if they are)
anyway, yes, sun boxes cost more than their intel counterparts in the low to mid range. That said, I have yet to find an intel box that does what the X1 can do for the low end, and once you get to 8 ways systems, intel starts to disappear from the map (and the sun boxes are the same cost or cheaper).
Now, we got the hardware price argument out of the way.
when making a decision, there are 3 major areas to consider: Price, Performance and Reliability. Only an idiot would focus on price when the cost of downtime is a million an hour.
The real reason i purchase sun boxes is not because they are the fastest. You want fast cpu's? Go get an intel box.
here are the main reasons I continue to purchase sun boxes:
#1) Sun's support organization. It is second to none. period, end of story. You have a problem, they fix it. I had a failed disk earlier this week, the support rep's first response was to send a tech on site that day.
#2) When they boast about binary compatibility from $1,000 to $10,000,000, they are not kidding. I can give the developers a low end box and know that the app will still work on a mid to high end box
#3) It just works. I dont get the "what glib are you using", "is that rev XYZ of that nic?" or any of that other crap.
#4) the hardware seems to last forever and ever and ever. And sun supports the stuff for a long time. Every try and get dell to support a six year old box? yeah, good luck.
#5) did i mention the support?
#6) it was built to be managed from a serial port and live on a network from day 1. I love the fact that i can put all of my servers in a colo, walk out, and do the OS install from home. I know that PC's are now beginning to get to the point where you can hook a serial cable up and get them to boot from the net and do an os install. lets face it, there are whole books on how to use jumpstart in the sun environment and do 100% hands off installs. It just works, and it is fully supported.
So, as you can see, there is more to the decision than just cost. In the world that i work in, time is money, and the hardware cost is a very small percentage of the TCO.
Actually, it is licensed to run based on the number of cpu's. RAC (real application clusters) cost an extra 50%. Last I checked, you could download 9i for linux intel (just watch those system requirements very carefully, your favorite distro is most likely not covered)
All of that said, if oracle can get you to get rid of your 72CPU SunFire 15K and replace it with a 128 single cpu intel boxes..... (extra intel boxes to make up for the lost ram and system bandwidth in the 15K)
Lets see what that would cost ya...
list price for a 72 cpu license for a single box is 2.9 million. List price for those 128 cpu's w/ RAC will cost ya 7.6 million.
Lots of smaller, "cheaper" systems can often cost less overall. This is not the case here, where the price delta more than makes up for the cost of the big sun box, and we dont even have to get into the argument over the extra cost involved in managing 128 different systems. (besides, RAC is not a 'shared nothing' cluster, so management of large clusters is a real pain)
anyway, larry is always going to need sun to produce the big boxes for its big clients.
No, a beowulf cluster is the last thing that one would use for nuclear simulation.
While great at highly parallel tasks that require very little synchronization between threads (think code cracking), nuclear testing (and almost all other fluid dynamic problems) generally requires all of the cpu's to have high speed access to all of the memory. So one needs a huge shared memory system (think Cray or Sun StarCat).
And for this reason, I find the top 500 list to be a bit misleading in these days of massively parallel systems. Its great as a test of how many flops the system can crank out, but it does not take into account the memory bandwidth between the cpu's, and that is often more important than raw cpu horsepower.
#4 -- Most of the time you are in class only b/c you have to be. Honestly 60% of the time spent in class is down time. Not learning time. Who cares what they are doing.
Dude, what type of community college are you going to? I'm sorry, but there was very little downtime in the college classes that I attended. If you were not paying attention to half of the lecture, you were in trouble. It was as simple as that. It was full speed ahead, hope you can keep up with the rest of us.
Then again, the freshman intro classes were designed to make sure that the first year's didn't have time to go to frat parties (and it also helped the departments weed out the less capable).
So, my suggestion to you is this: if you spend 60% of your time in class bored and not learning anything, you should find a new college to go to. After all, you're paying for it, you may as well get your money's worth.
The SGI origin has a ccNUMA architecture, which makes it great for some tasks, ok for others, and awful for yet others. (the trick is to make sure that your particular app falls under the 'great' category)
The sun system is an smp based system, everything connects to a common backplane and each board has equal access to all of the other boards. With the sgi, the speed of accessing memory on the local board or boards in the same cabinet is much faster than hits to memory in remote cabinets.
From what I can tell, Sun is planing on producing a special system board that goes into one of those 18 slots. Thus, with 19 StarCats you can create one big system with 1836 cpu's and 9.7TB of ram. (think of a system in the middle that acts as the center of a star) it will most likely be based on a COMA architecture rather than a ccNUMA. Like the SGI, memory access will depend on the distance between the requesting cpu and the storage location. The difference is that under COMA, if a cpu requests a particular bit of memory a lot, that page is either migrated or copied to a memory bank on that cpu's memory board (so if 5 cpu's all need read only access to the same bit of memory, then they can each have their own copy in a local memory bank. write updates are what make the system a pain in the ass to manage ).
There are 18 "cpu/memory" boards that hold 4 cpu's each. This brings the system up to a total of 72 cpu's and 576GB of ram.
Now, if you want an server that just does number crunching and dont care about I/O, you can then add 'MaxCPU" modules. Each module holds two additional cpu's (no memory) and occupies the hPCI module slot (a hot swap PCI case that can hold what looks like two to four pci cards). You can use up to 17 of the hPCI module slots to hold MaxCPU modules. (there are 18 pci channels on the system, and at least one must be used for accessing the boot disk).
So there ya have it, 106 cpu's and half a terabyte of ram. I think that in most cases, folks will opt to not use the MaxCPU modules and just stick to the 72 cpu limit.
Lets remember, that this system is not intended to replace a beowolf cluster of cheap pc's. It is intended to do something that most beowolf clusters can never do: present a single OS image with half a terabyte of memory that any cpu can access at very high speed.
This is a system that is very good at things like fluid dynamics and massive database operations. It is not a good idea if all you want to do is get to the top of the list for the SETI@Home project
It also varies from state to state. My case involved colorado. California has their own laws on the topic. YMMV, so se a real lawyer. In the end, you should at least get paid before any investors (if there is anything left).
As an interesting side note, it seems that only salary and accrued vacation (not pto, vacation) is protected in colorado. So, while I did get paid, the policy premiums for our life and medical insurance were not covered for the last month (actually just the last 8 days of employment). (this is why i carry my own life insurance in addition to my employers). that said, my employer was nice enough to send me a check for $14.95 to cover the cost of a perscription that my wife had filled 3 days before I got caned.
Having recently gone through a dot bomb, here is the order of payout:
1) Employees get paid first. Period. If the company does not have the cash to cover payroll, they are in big trouble.
2) Creditors get paid second, usually in order of size or importance. This means that the bank gets their money and contractors get their take after the bank.
3) VC's get whatever is left (if anything). They put their money at risk, they knew the risk, and they stood the most to gain.
Now, in some states (I'm in colorado), if the employer does not pay in 15 days or so, you can send them a nice little form letter (available at the colorado department of labor's website) that basically says that if they dont pay in 15 days that they owe you triple.
Now, here is the real kicker: if they still dont pay, you can go after the company and then select officers of the company and the (yes, the ceo himself and usually the head of the board of trustees). Like I mentioned before, they are required to be able to meet payroll, and if they can't, They must lay you off before they run out of money, not after.
Anyway, that is the way it worked in my case. IANAL, but I play one on slashdot.
Most banks (at least in the US) will wave most fees if you manage to have a minimum balance or meet other criteria. Some (mostly smaller credit unions) will even cover the cost of the ATM fees when you use a 3rd party atm.
Now, a starving college student may have a hard time trying to manage the ~$1,500 minimum balance to avoid all fees, but there are other ways around the problem of service fees:
1) Only use your banks ATM's. Lets face it, if you purchase on impulse and need to use the atm that is conveniently located where you are, then you really dont manage your money that well anyway and should just count the atm fee as a stupidity tax. Plan your purchases in advance and hit your banks atm on the way and you can avoid the problem.
2) Use a credit card with no annual fee. You should be able to find one with very little hassle, just dont expect a 9.99% interest rate to go along with it.
3) pay that credit card off each month. You don't care if you have a card with a high interest rate if you pay it off in full each month.
4) get a checking account with no monthly fees (my bank will wave the fee if you manage a minimum balance or use direct deposit). There may be a few hoops, but it is usually worth your time and effort to jump through those hoops to save the fees (mine used to charge a $2 "transaction fee" for every transaction that I did at the bank that I could have done through an ATM (deposits, withdraws, etc). Solution? simple, use the atm.
In the end, I have not paid a banking fee (interest, service charges, etc, etc) in several years. It's all part of a good money management solution.
The british failed because the british troops had no real reason to fight other than the glory of england. (let me tell ya, that does not get you very far against anybody who is fighting for their homeland).
The russians failed because the US saw an excellent opportunity to accelerate the collapse of the soviet empire. After all, do you really think that the afghans found that many AK-47's on the battlefield? Didn't think so. The US gave the resistance lots of weapons. This is the same group that would become the taliban a few years later. We also gave them billion US dollars in aid to make sure that they had the supplies that they needed.
Of course, the big error made by the US. After the war, we just got up and walked away. We created a vacuum, the taliban filled that vacuum. And now that you have a country of folks who only know how to do one thing well (other shoot a US made stinger missile), it is hard to get them to suddenly do something else (like rebuild a nation).
Had the US poured a few of those billion into the country in the form of humanitarian aid, we would not have this problem today. Then again, we expected the local countries to do their part in assisting the afghans. (lesson #2, The Saudi's don't give a shit about the afghans)
So, how do we get out of this mess?
We can start by using special ops to get at a few of the more popular targets (bin laden, a few of the camps). However, the real victories will not be shown on CNN in an arcade like display: finding the money trail, arresting folks before they get a chance to carry out their plans, etc, etc. The last thing that we should do is launch a land war and try and occupy the country. Unless we want to throw a few million americans at the problem, we don't have a chance in hell of making it work.
In the end, the US will have to take a long term view of the area, something that we have never done and try and get ourselves out of the area as soon as possible. After all, we judge a company based on their quarterly results, what makes you think that we will start to ask the question "but what happens in 10 years"
Lets hope that the US gets a clue this time and takes care of the problem and not just the symptom. After all, I'm a capitalist pig who likes all of the things that a healthy economy can get me and that economy does not like having large buildings blown up.
Ethernet starts having problems due to collisions at 40% or so - depending on the number of nodes.
Oh fergawdsakes, will this urban legend ever die!
It simply isn't true.
Actually, it is true... Under ethernet, the more nodes you put in a collision domain (this is layer 2, not to be confused with a layer 3 _broadcast_ domain), the more likely a collision, the more time is spent recovering from collisions. The exact 40% depends on how many nodes are on the segment and how much info they have to send and how often, but 40% is in the ballpark for when problems start to appear
There is a very simple (and not very expensive nowadays) way to get around this problem: use switches in place of hubs. A switch creates a two node collision domain (the switch and the end node). In fact, when full duplex is enabled (only possible with switches for an obvious reason) the collision detection is turned off.
The result: no collisions and thus you can get even more efficient than token ring (no token to wait for).
I have to admit I don't understand why Sun is resisting the switch to Linux. I'm not saying they should dump Solaris over night, but a two or three year transition plan would make a lot of sense
(warning: Sarcasm ahead) Ya know, I dont know why all of these BSD guys keep resisting the switch to linux. I'm not saying that they should dump *BSD overnight, but a two or three year transition plan would make a lot of sense
That said, I run all of my production systems off of SPARC. When it comes to cheap cpu that only takes up 1 rack unit of space, you just can't beat a Sun X1 ($995 to start, and it takes PC133 DIMM's to boot).
Solaris has a huge installed base and is rock solid. I've been running solaris systems for 5 years now, and they only crash if there is a hardware problem. IMNSHO, their support group is second to none.
Besides, can you imagine the mods that would be required to allow linux to dynamically add cpu and memory modules? (not that one can do that with pc hardware)
Lets face it, you choose the right tool for the job. For my environment, that tool is solaris. For yours, it is prob Win2K.
The article mentions that the end node are where the intelligence is. that is quite wrong.
the pc's at the end don't have a clue as to how to get a packet from point a to b, they just send it to the default router and hope that all goes well.
the routers are where the intelligence resides on the internet, all 100,000+ of them. (some more intelligent than others)
if anything, the internet is an example of a decentralized intelligence. There is no single point of control. there is no single point of failure. there is no single person who can thus guarantee that stuff will work since they dont control everything from point a to b.
now, ATT was able to have 99.999% uptime because they controlled the entire thing. Many of the larger ISP's also have what is very close to that same level of reliability _within their own network_. Once it leaves their network it is out of their hands. ATT does not take responsibility for QOS of calls to china and ISP's do not offer any SLA's for packets that leave their network
The backbone can support all that various business wants it to, they just dont want to pay for it. Think about it, a single long distance call from New York to San Francisco cost about $0.05 a minute (us) That gets you a dedicated 64Kbps link from point a to b (assuming the old uncompressed telco data rates where a T1 carried 28 voice channels). Now, what does a typical dialup line cost? $20 a month? That would buy you 400 minutes or a little under 7 hours. How many hours of surfing does the average person do per month on their 56K line? I bet that it is a good bit over 7 hours, prob closer to 30+ (not counting slashdot users. I racked up over 550 hours in one month once when i was telecommuting and only had a dialup. the isp was not happy with me).
So, using the telco networks price as a guide, if we all want dedicated, guaranteed access, we should be willing to pay for it. Thus, our 30 hours per month of internet access should a) not exceed 64Kbps and b) cost about $90. Want to do video @1.5Mbps? That will cost you a bit more:)
Having interviewed more folks than I care to remember for a bunch of positions over the years, I've given up on asking almost anything that can be answered with a quick glance at the man page.
Lets face it, a sysadmin almost always has the man pages at their fingertips, so who cares if the person knows what flag to tar will produce a listing of what is in the archive.
What I am looking for is a conceptual understanding of the system. How do they work? what is their problem solving method? Are they a 'hacker' in their problem solving methodology or do they take a more scientific approach (and change only one variable at a time and retest)? (give you a hint, hackers are great in the development lab and the last thing that you want in a production environment)
I also look for the ability to understand how a group of systems are put together and why one would or would not use certain applications based on a given environment. For example, in our env the dev systems all use NIS and are generally wide open (I work in a mostly sun env with a few linux boxen) but production systems never have nis and never run inetd. ssh and the server application itself are the only things on those systems. If some sysadmin walked in and installed package xyz on the production systems cuz it is 'cool', they'd be shown the door real quick.
I tend to be a bit biased toward folks with a degree simply because it tends to show the ability to organize large amounts of information and stick with a task (that said, I'll substitute 4 years of real world experience for that degree). I also tend to prefer folks with degree's in the practical hard sciences (chem, physics). Comp sci degrees are nice but not of any importance if the person knows at least 3 other computer languages.
overall, I look for somebody with good organization and problem solving skils. If they have those then the rest they can learn on the job fast enough.
if they used a simple usb (passthrough of course) hardware dongle in lieu of the registration process?
I mean, which is cheaper, tossing in a $1.00 usb dongle or having a team of folks staff the phone lines to handle all of the requests?
Of course, the usual folks will cry about the pain of dongles, but such is life. get over it. (for a good time, try being a license admin with unix boxes that tie everything to the hostid, great fun when you upgrade your license server and have to contact 20 different vendors for new keys to half a million worth of software:)
Slashdot Mirror
In tests that i've done on solaris boxes (a very small 2 cpu UltraSPARC III server, not the big iron) I often see 'strange' results like this from benchmarking utilities like iozone. (as in 600+MB/s read and write speeds)
:-)
In general, if the file fits into memory, read and write speeds can often exceed the theoretical limits of the underlying array. This is because the OS buffers the hell out of the I/O (unless you use something like UFS's 'forcedirectio' mount flag to avoid the OS buffers). This is good and bad. Good for throughput but bad if the buffer is allowed to grow unchecked and forces applications into swap as one tries to write a huge file that is larger than the RAM on the system. (think DVD iso's on a box with only 4G of ram).
In the case of the 2.4 vs 2.6 shootout, I'd be much more interested in seeing a full iozone graph showing block size vs file size vs performance. ( If you do this, make sure that you test for file sizes at least twice as large as the RAM on the box to see what happens when the system runs out of space for a buffer. )
All of that said, the best I've seen so far (same 2 cpu sparc box) in actual disk performance is 180MB/s (and that was with a NexSAN ATABeast that had 42 300G IDE drives). The load was spread over two 1Gb fibre channel links, so i was starting to approach the max throughput of the HBA's at that point
you may want to check out products like the GreenLight Trigger. Its basically a magnet under the bike that trips the sensor.
Did ya ever see the stereo's in those POS's owned by most teens? It often increases the value of the car by an order of magnitude :-)
Unfortunately, it is a bit harder than a simple zip code based lookup since zip codes can span towns and counties. It really requires a full gis based system that uses the tax boundaries (which can change at a moments notice) to figure out how much to charge. It is possible that your next door neighbor pays a different tax rate than you.
All of that said, if the states fail in their quest, they will prob just move toward a higher property tax/income rate. (in my part of the world, the local city has a budget shortfall due to the lack of tourists spending money at local stores. a property tax would have avoided the problem, although at the expense of the local population)
# rm / -r
/' ? (i know, the '/ -r' syntax workes on linux, but not some of the other unixes out there, like solaris)
rm: / is a directory
-r: No such file or directory
Hmmmm, dont you mean 'rm -r
The University of Wisconsin has a site at http://genchem.chem.wisc.edu/demonstrations/ that contains links to dozens of demos (with raitings) for various categories. I'm sure there is something for each age/interest level there
this worked before in a chemistry demo i gave in high school. It gives an unexpected result and it is colorful to boot :-) http://jchemed.chem.wisc.edu/JCESoft/CCA/CCA2/MAIN /AMFOUNT1/CD2R1.HTM for more details
I dont think that intel ever really had a question for Sun to answer. At 800MHz, Intel's 64it chip is slow in the all important MHz rating (sun has had 900's out for a while now) and still has a few years of compiler design ahead of it before it makes any sense. And this is Sun's 3+rd generation of 64 bit chips, vs Intel's 1st.
As for HP, they helped intel build their 64 bit chip, so the PA-RISC is more or less dead.
those are list prices from store.oracle.com. University's get a huge discount.
;-) and the ongoing cost of managing the extra added complexity.
note that standard edition is much cheaper than Enterprise Edition. (like 1/10th the cost) However, if you want to use RAC (which larry is indicating as the replacement for a single large system) then you must use EE. Also, you are not permitted to put standard edition on any system with more than 4 cpu's. Also, this is just oracle's cost, you still need to add in your own cluster software to make sure that if an instance goes down, you dont loose that data (RAC is not a 'shared nothing' cluster)
now, if you want to manually partition your data, you could get away with using standard edition on a ton of boxes. you are simply trading cost in one area for extra time and expense to implement in another The actual advantage in ones particular environment is dependant on the cost of time (students are cheap
Ok, i'll bite:
for starters, you are comparing a dual cpu box with a quad cpu box. The quad's always cost more per cpu. Simple matter of the fact that it is harder to get 4 cpus' to talk together compared to only 2 (why do you think intel has yet to produce a 64 way smp server.... sun did it 5 years ago, cray did it before them.) You also have to look at things like backplane contention (are all of those cpu's on the same bus? sucks to be you if they are)
anyway, yes, sun boxes cost more than their intel counterparts in the low to mid range. That said, I have yet to find an intel box that does what the X1 can do for the low end, and once you get to 8 ways systems, intel starts to disappear from the map (and the sun boxes are the same cost or cheaper).
Now, we got the hardware price argument out of the way.
when making a decision, there are 3 major areas to consider: Price, Performance and Reliability. Only an idiot would focus on price when the cost of downtime is a million an hour.
The real reason i purchase sun boxes is not because they are the fastest. You want fast cpu's? Go get an intel box.
here are the main reasons I continue to purchase sun boxes:
#1) Sun's support organization. It is second to none. period, end of story. You have a problem, they fix it. I had a failed disk earlier this week, the support rep's first response was to send a tech on site that day.
#2) When they boast about binary compatibility from $1,000 to $10,000,000, they are not kidding. I can give the developers a low end box and know that the app will still work on a mid to high end box
#3) It just works. I dont get the "what glib are you using", "is that rev XYZ of that nic?" or any of that other crap.
#4) the hardware seems to last forever and ever and ever. And sun supports the stuff for a long time. Every try and get dell to support a six year old box? yeah, good luck.
#5) did i mention the support?
#6) it was built to be managed from a serial port and live on a network from day 1. I love the fact that i can put all of my servers in a colo, walk out, and do the OS install from home. I know that PC's are now beginning to get to the point where you can hook a serial cable up and get them to boot from the net and do an os install. lets face it, there are whole books on how to use jumpstart in the sun environment and do 100% hands off installs. It just works, and it is fully supported.
So, as you can see, there is more to the decision than just cost. In the world that i work in, time is money, and the hardware cost is a very small percentage of the TCO.
Actually, it is licensed to run based on the number of cpu's. RAC (real application clusters) cost an extra 50%. Last I checked, you could download 9i for linux intel (just watch those system requirements very carefully, your favorite distro is most likely not covered)
All of that said, if oracle can get you to get rid of your 72CPU SunFire 15K and replace it with a 128 single cpu intel boxes..... (extra intel boxes to make up for the lost ram and system bandwidth in the 15K)
Lets see what that would cost ya...
list price for a 72 cpu license for a single box is 2.9 million. List price for those 128 cpu's w/ RAC will cost ya 7.6 million.
Lots of smaller, "cheaper" systems can often cost less overall. This is not the case here, where the price delta more than makes up for the cost of the big sun box, and we dont even have to get into the argument over the extra cost involved in managing 128 different systems. (besides, RAC is not a 'shared nothing' cluster, so management of large clusters is a real pain)
anyway, larry is always going to need sun to produce the big boxes for its big clients.
No, a beowulf cluster is the last thing that one would use for nuclear simulation.
While great at highly parallel tasks that require very little synchronization between threads (think code cracking), nuclear testing (and almost all other fluid dynamic problems) generally requires all of the cpu's to have high speed access to all of the memory. So one needs a huge shared memory system (think Cray or Sun StarCat).
And for this reason, I find the top 500 list to be a bit misleading in these days of massively parallel systems. Its great as a test of how many flops the system can crank out, but it does not take into account the memory bandwidth between the cpu's, and that is often more important than raw cpu horsepower.
Dude, what type of community college are you going to? I'm sorry, but there was very little downtime in the college classes that I attended. If you were not paying attention to half of the lecture, you were in trouble. It was as simple as that. It was full speed ahead, hope you can keep up with the rest of us.
Then again, the freshman intro classes were designed to make sure that the first year's didn't have time to go to frat parties (and it also helped the departments weed out the less capable).
So, my suggestion to you is this: if you spend 60% of your time in class bored and not learning anything, you should find a new college to go to. After all, you're paying for it, you may as well get your money's worth.
The SGI origin has a ccNUMA architecture, which makes it great for some tasks, ok for others, and awful for yet others. (the trick is to make sure that your particular app falls under the 'great' category)
The sun system is an smp based system, everything connects to a common backplane and each board has equal access to all of the other boards. With the sgi, the speed of accessing memory on the local board or boards in the same cabinet is much faster than hits to memory in remote cabinets.
From what I can tell, Sun is planing on producing a special system board that goes into one of those 18 slots. Thus, with 19 StarCats you can create one big system with 1836 cpu's and 9.7TB of ram. (think of a system in the middle that acts as the center of a star) it will most likely be based on a COMA architecture rather than a ccNUMA. Like the SGI, memory access will depend on the distance between the requesting cpu and the storage location. The difference is that under COMA, if a cpu requests a particular bit of memory a lot, that page is either migrated or copied to a memory bank on that cpu's memory board (so if 5 cpu's all need read only access to the same bit of memory, then they can each have their own copy in a local memory bank. write updates are what make the system a pain in the ass to manage ).
The system grows to 106 in the following way:
There are 18 "cpu/memory" boards that hold 4 cpu's each. This brings the system up to a total of 72 cpu's and 576GB of ram.
Now, if you want an server that just does number crunching and dont care about I/O, you can then add 'MaxCPU" modules. Each module holds two additional cpu's (no memory) and occupies the hPCI module slot (a hot swap PCI case that can hold what looks like two to four pci cards). You can use up to 17 of the hPCI module slots to hold MaxCPU modules. (there are 18 pci channels on the system, and at least one must be used for accessing the boot disk).
So there ya have it, 106 cpu's and half a terabyte of ram. I think that in most cases, folks will opt to not use the MaxCPU modules and just stick to the 72 cpu limit.
Lets remember, that this system is not intended to replace a beowolf cluster of cheap pc's. It is intended to do something that most beowolf clusters can never do: present a single OS image with half a terabyte of memory that any cpu can access at very high speed.
This is a system that is very good at things like fluid dynamics and massive database operations. It is not a good idea if all you want to do is get to the top of the list for the SETI@Home project
It also varies from state to state. My case involved colorado. California has their own laws on the topic. YMMV, so se a real lawyer. In the end, you should at least get paid before any investors (if there is anything left).
As an interesting side note, it seems that only salary and accrued vacation (not pto, vacation) is protected in colorado. So, while I did get paid, the policy premiums for our life and medical insurance were not covered for the last month (actually just the last 8 days of employment). (this is why i carry my own life insurance in addition to my employers). that said, my employer was nice enough to send me a check for $14.95 to cover the cost of a perscription that my wife had filled 3 days before I got caned.
Having recently gone through a dot bomb, here is the order of payout:
1) Employees get paid first. Period. If the company does not have the cash to cover payroll, they are in big trouble.
2) Creditors get paid second, usually in order of size or importance. This means that the bank gets their money and contractors get their take after the bank.
3) VC's get whatever is left (if anything). They put their money at risk, they knew the risk, and they stood the most to gain.
Now, in some states (I'm in colorado), if the employer does not pay in 15 days or so, you can send them a nice little form letter (available at the colorado department of labor's website) that basically says that if they dont pay in 15 days that they owe you triple.
Now, here is the real kicker: if they still dont pay, you can go after the company and then select officers of the company and the (yes, the ceo himself and usually the head of the board of trustees). Like I mentioned before, they are required to be able to meet payroll, and if they can't, They must lay you off before they run out of money, not after.
Anyway, that is the way it worked in my case. IANAL, but I play one on slashdot.
Most banks (at least in the US) will wave most fees if you manage to have a minimum balance or meet other criteria. Some (mostly smaller credit unions) will even cover the cost of the ATM fees when you use a 3rd party atm.
Now, a starving college student may have a hard time trying to manage the ~$1,500 minimum balance to avoid all fees, but there are other ways around the problem of service fees:
1) Only use your banks ATM's. Lets face it, if you purchase on impulse and need to use the atm that is conveniently located where you are, then you really dont manage your money that well anyway and should just count the atm fee as a stupidity tax. Plan your purchases in advance and hit your banks atm on the way and you can avoid the problem.
2) Use a credit card with no annual fee. You should be able to find one with very little hassle, just dont expect a 9.99% interest rate to go along with it.
3) pay that credit card off each month. You don't care if you have a card with a high interest rate if you pay it off in full each month.
4) get a checking account with no monthly fees (my bank will wave the fee if you manage a minimum balance or use direct deposit). There may be a few hoops, but it is usually worth your time and effort to jump through those hoops to save the fees (mine used to charge a $2 "transaction fee" for every transaction that I did at the bank that I could have done through an ATM (deposits, withdraws, etc). Solution? simple, use the atm.
In the end, I have not paid a banking fee (interest, service charges, etc, etc) in several years. It's all part of a good money management solution.
Ok, lets look this one over:
The british failed because the british troops had no real reason to fight other than the glory of england. (let me tell ya, that does not get you very far against anybody who is fighting for their homeland).
The russians failed because the US saw an excellent opportunity to accelerate the collapse of the soviet empire. After all, do you really think that the afghans found that many AK-47's on the battlefield? Didn't think so. The US gave the resistance lots of weapons. This is the same group that would become the taliban a few years later. We also gave them billion US dollars in aid to make sure that they had the supplies that they needed.
Of course, the big error made by the US. After the war, we just got up and walked away. We created a vacuum, the taliban filled that vacuum. And now that you have a country of folks who only know how to do one thing well (other shoot a US made stinger missile), it is hard to get them to suddenly do something else (like rebuild a nation).
Had the US poured a few of those billion into the country in the form of humanitarian aid, we would not have this problem today. Then again, we expected the local countries to do their part in assisting the afghans. (lesson #2, The Saudi's don't give a shit about the afghans)
So, how do we get out of this mess?
We can start by using special ops to get at a few of the more popular targets (bin laden, a few of the camps). However, the real victories will not be shown on CNN in an arcade like display: finding the money trail, arresting folks before they get a chance to carry out their plans, etc, etc. The last thing that we should do is launch a land war and try and occupy the country. Unless we want to throw a few million americans at the problem, we don't have a chance in hell of making it work.
In the end, the US will have to take a long term view of the area, something that we have never done and try and get ourselves out of the area as soon as possible. After all, we judge a company based on their quarterly results, what makes you think that we will start to ask the question "but what happens in 10 years"
Lets hope that the US gets a clue this time and takes care of the problem and not just the symptom. After all, I'm a capitalist pig who likes all of the things that a healthy economy can get me and that economy does not like having large buildings blown up.
Oh fergawdsakes, will this urban legend ever die!
It simply isn't true.
Actually, it is true... Under ethernet, the more nodes you put in a collision domain (this is layer 2, not to be confused with a layer 3 _broadcast_ domain), the more likely a collision, the more time is spent recovering from collisions. The exact 40% depends on how many nodes are on the segment and how much info they have to send and how often, but 40% is in the ballpark for when problems start to appear
There is a very simple (and not very expensive nowadays) way to get around this problem: use switches in place of hubs. A switch creates a two node collision domain (the switch and the end node). In fact, when full duplex is enabled (only possible with switches for an obvious reason) the collision detection is turned off.
The result: no collisions and thus you can get even more efficient than token ring (no token to wait for).
(warning: Sarcasm ahead) Ya know, I dont know why all of these BSD guys keep resisting the switch to linux. I'm not saying that they should dump *BSD overnight, but a two or three year transition plan would make a lot of sense
That said, I run all of my production systems off of SPARC. When it comes to cheap cpu that only takes up 1 rack unit of space, you just can't beat a Sun X1 ($995 to start, and it takes PC133 DIMM's to boot).
Solaris has a huge installed base and is rock solid. I've been running solaris systems for 5 years now, and they only crash if there is a hardware problem. IMNSHO, their support group is second to none.
Besides, can you imagine the mods that would be required to allow linux to dynamically add cpu and memory modules? (not that one can do that with pc hardware)
Lets face it, you choose the right tool for the job. For my environment, that tool is solaris. For yours, it is prob Win2K.
exactly.
:)
The article mentions that the end node are where the intelligence is. that is quite wrong.
the pc's at the end don't have a clue as to how to get a packet from point a to b, they just send it to the default router and hope that all goes well.
the routers are where the intelligence resides on the internet, all 100,000+ of them. (some more intelligent than others)
if anything, the internet is an example of a decentralized intelligence. There is no single point of control. there is no single point of failure. there is no single person who can thus guarantee that stuff will work since they dont control everything from point a to b.
now, ATT was able to have 99.999% uptime because they controlled the entire thing. Many of the larger ISP's also have what is very close to that same level of reliability _within their own network_. Once it leaves their network it is out of their hands. ATT does not take responsibility for QOS of calls to china and ISP's do not offer any SLA's for packets that leave their network
The backbone can support all that various business wants it to, they just dont want to pay for it. Think about it, a single long distance call from New York to San Francisco cost about $0.05 a minute (us) That gets you a dedicated 64Kbps link from point a to b (assuming the old uncompressed telco data rates where a T1 carried 28 voice channels). Now, what does a typical dialup line cost? $20 a month? That would buy you 400 minutes or a little under 7 hours. How many hours of surfing does the average person do per month on their 56K line? I bet that it is a good bit over 7 hours, prob closer to 30+ (not counting slashdot users. I racked up over 550 hours in one month once when i was telecommuting and only had a dialup. the isp was not happy with me).
So, using the telco networks price as a guide, if we all want dedicated, guaranteed access, we should be willing to pay for it. Thus, our 30 hours per month of internet access should a) not exceed 64Kbps and b) cost about $90. Want to do video @1.5Mbps? That will cost you a bit more
Having interviewed more folks than I care to remember for a bunch of positions over the years, I've given up on asking almost anything that can be answered with a quick glance at the man page.
Lets face it, a sysadmin almost always has the man pages at their fingertips, so who cares if the person knows what flag to tar will produce a listing of what is in the archive.
What I am looking for is a conceptual understanding of the system. How do they work? what is their problem solving method? Are they a 'hacker' in their problem solving methodology or do they take a more scientific approach (and change only one variable at a time and retest)? (give you a hint, hackers are great in the development lab and the last thing that you want in a production environment)
I also look for the ability to understand how a group of systems are put together and why one would or would not use certain applications based on a given environment. For example, in our env the dev systems all use NIS and are generally wide open (I work in a mostly sun env with a few linux boxen) but production systems never have nis and never run inetd. ssh and the server application itself are the only things on those systems. If some sysadmin walked in and installed package xyz on the production systems cuz it is 'cool', they'd be shown the door real quick.
I tend to be a bit biased toward folks with a degree simply because it tends to show the ability to organize large amounts of information and stick with a task (that said, I'll substitute 4 years of real world experience for that degree). I also tend to prefer folks with degree's in the practical hard sciences (chem, physics). Comp sci degrees are nice but not of any importance if the person knows at least 3 other computer languages.
overall, I look for somebody with good organization and problem solving skils. If they have those then the rest they can learn on the job fast enough.
if they used a simple usb (passthrough of course) hardware dongle in lieu of the registration process?
:)
I mean, which is cheaper, tossing in a $1.00 usb dongle or having a team of folks staff the phone lines to handle all of the requests?
Of course, the usual folks will cry about the pain of dongles, but such is life. get over it. (for a good time, try being a license admin with unix boxes that tie everything to the hostid, great fun when you upgrade your license server and have to contact 20 different vendors for new keys to half a million worth of software