It's a supply and demand thing. One thing that you might also want to to worry about is those "schools" churning out paper MCSEs month after month, advertising big $$$ and life on Easy Street by passing a few tests and getting a few certificates. In an already overcrowded tech market, these places are turning out tons of folks with overblown expectations. Once their dreams are crushed, who knows how cheap they'll be willing to work?
To be fair, colleges and universities all over the US have been doing the same thing for at least 20 years. The degrees aren't high-tech; they are english, art history, etc. I'd love to have spent 4 or 5 years in school only to graduate and find that I can make more money as a waiter than in my "profession," assuming I could find a position.
I have to agree here. My partner and I started a successful business right out of grad school doing consulting work and building clusters. We did it without VC funding and haven't really needed to borrow much along the way except to buy parts for larger projects when we couldn't fund it internally. This will not make you Bezos rich, but it isn't a bad life. Honestly, I don't know how much faster I would want to grow. More money means more headaches. Just being comfortable and busy is good enough for me.
A cheap Duron, hard drive, case, TV-out capable video card, and TV card is going to probably cost more than a stand-alone Tivo, so you are only saving on the "backend." I like the progress that I have seen in MythTV and Freevo, especially the integrated features like emulators and such. That is the appeal for me to build these type of solutions, not price.
FWIW, I am a DirecTV subscriber and the DirecTivo gives you a lot of functionality that you just can't easily replicate. It stores full quality video from the satellite feed on the hard drive. It also allows you to record two shows at the same time. That makes it well worth the price. Of course, mine has a 120 GB hard drive hacked in to give over 100 hours of storage.
Now I just need to figure out how copy the video from the Tivo. I can ftp and telnet into the system, but I haven't investigated the state of the extraction software lately. One of those projects I need to get to...
A "pea" travelling at 90% of the speed of light contains a lot of kinetic energy. Say, 0.01 grams for the pea at 2.7e8 m/s. That works out to 7.3e11 J. That is about the same energy as exploding 175 tons of TNT per pea.
Set aside the issue of engineering the "peashooter" to fire them, you are talking about throwing some potentially destructive material at a neighboring star system. Firing them continuously looks like you intentially want to hit something. I think this might be a bad idea from a "just saying hello" viewpoint.
The problem with sending humans and cargo together is that it is cost effective to use "crappier" rockets to send up fairly replaceable cargo and the "good ones" to send up people. Generally, people are not replaceable, at least in most political terms. The human transport needs to over-engineered and super-certified safe. A BDR can blow up once in a while. It would suck, but at least it wouldn't derail the whole space program for three or four years, like Columba.
In the article, he recommends retooling the boosters and tanks with a new crew module and a separate cargo module. While I can see his point with regard to reuse of existing proven technology, I can't quite get away from the idea that sending people in the same craft as cargo is just a bad idea. Big Dumb Rockets (BRDs) are the only way that we should be putting cargo of any sort into space. We have become very good at orbital rendevous manuevers, so I can't see that separate launch vehicles for people and cargo will present a significant complication.
Perhaps the booster and tanks can be recycled with ONLY a crew module that can actually reach the L1 point. The current shuttles can barely make it to low orbit.
On the whole, he is right. An L1 base would be a nice permanent move into space and is probably something that should have been done in the mid-70's. The establishment of a moon base will be an easier political sell though. Once we hammer out manufacturing techniques, it should be possible to grow a spawling complex on the moon without needing to carry everything from earth. And you know we Americans love to spawl. If we can find water in sufficient quantities and are willing to take nuclear reactors with us to the moon, the fuel for future space flights will probably come from the moon.
I can think of another area where code generation could be very useful...numerically intensive tasks. There are lots of C optimization techniques (register and cache blocking especially) that are architecture, CPU, and memory subsystem dependent. To optimize for different platforms, you normally need to do explicit loop unrolling and hardwire blocking factors in the code.
Code generation allows a "higher level" description of the problem in some abstracted language to minimize replication of common code and also allows adaptive selection of blocking factors, etc. to maximize code performance. The Altas BLAS library is an excellent example of the power of the latter which amounts to per-system adapative code generation. The C Preprocessor and C++ Templates are not sufficiently general to allow this same sort of optimization without some code generation segment.
It has been "coming" for several years now. Really, it is pretty pathetic. I outted the bucks for an HD projector, the Sony HD DirecTV receiver, and an amplified UHF antenna rig. With a lot of work, I was able to get ABC, CBS, FOX, PBS, and NBC to tune in from the broadcast towers in Pittsburgh.
After all of that work, the *only* channel that has decent HD content is PBS...and there are only a few shows that run all of the time. FOX is completely pathetic. The real networks run only a few shows a week. Even so, the prospect of being able to timeshift those few shows is pretty compelling, so I might need to build a MythHDTV box...Tivo has pretty much ruined me.
There are only a handfull of HDTV channels on direct-broadcast satellite feeds. DirecTV has HDNET, Discovery, ESPN, HBO and SHO in HD. The Dish lineup is similar. There is just not that much Satellite HD content available.
The DVB system that you pointed to is a nice setup, but there are no PC add-in cards that I know of that allow access to DirecTV or Dish digital feeds whether SD or HD. The one exception might be via DirecTivos which can be hacked to allow extraction of the video feeds, but this is as a postprocessing task, not realtime. Perhaps there are add-in cards for Big-Ugly Dishes to decode HD broadcasts, but I am not aware of them.
I believe that Sony is planning to produce an HD DirecTivo receiver shortly for timeshifting HD content. I don't know if they will do a better job of hack-proofing that unit than with other Tivo units, so offline HD extraction is still an open question.
I need to disagree with this statement. In general, most serious engineering design involves many PhD level people, either in managing the MS/BS guys or in solving really hard problems. Walk through GM or Ford or Nvidia or Intel. There are lots of Drs. around.
Now, a PhD in I.T. sounds overly broad. The area of specialization is key with advanced degrees. My advanced degree is not in engineering or even mechanical engineering...it is in computational fluid dynamics. A PhD in security or networking or algorithm design could be highly useful and lead to well paying positions doing that sort of work. A PhD for someone who is changing network cards and installing Windows service packs is a complete waste...
It is called bi-linear filtering...instead of making pixels larger when zooming, it interpolates to fill in the in-between pixels using the four nearest "real" pixels. There is a fair amount of linear algebra involved and that made it impossible for CPUs of that era to do it in real time. 3DFX was noteworthy for being the first hardware manufacturer to implement bi-linear filtering in hardware (Voodoo 1). </useless-info>
You are right; it is not possible to do computation without SOME losses even if we use completely reversible phenomena. The entropy increase comes when we "forget" information,.i.e., clearing a register.
That is different from what they are talking about in the article. Their goal is to move to essentially reversible reactions using spin rather than current-type electronic phenomena that contain Ohmic irreversibilities. The Ohmic losses dominate the heat generation in current ICs. The next on the list of energy loss for an IC is probably RF radiation. Entropy production from information loss is pretty far down the scale, but it is the one that cannot be "engineered away" so that is why it is always included in the "how big can a computer get" calculation.
What we need is an open-source distribution of software for Windows. It would install Mozilla, OpenOffice, GIMP, and whatever else seems useful. Wrap it up with a nice installer that makes the "open" nature of the software clear. Configure everything so that a mindless XP user can just run the installer and get all of the best that the opensource folks have to offer. It could become a very popular thing if it was all in one place and easy to find.
I can see a market for retail sales as well, so it could be worthwhile for a for-profit company to pursue this. I would love to be able to hand windows users a CD with an Open Source bundle of software and know that they will probably be able to explore and use the software without much handholding.
There is a BIG reason to roll your own that the article misses. WEP is pretty pathetic from a security viewpoint. To my knowledge, the cheap WAPs do not support IPSEC or other encryption layers natively on top of the wireless layer.
Last year, I was investigating "hackable" WAPs that I could reprogram using Linux or *BSD to enforce a layer of encryption at the first entry point into the network. Otherwise I would need to string a separate wired LAN (or VLAN) for the unsecured network connecting access points and then funnel everything to a firewall for the encryption. Unfortunately, this would still allow someone to hack on the other wireless clients, etc.
I am a little surprised that with war-driving, -flying, -chalking, ad nauseum, wireless security still seems to be a non-issue for most people.
BIOS chip swaps...if you hose a bios chip and need to "hotswap" and flash in another system, those pullers come in handy. But, to be fair, I have only done that once since my first 286 (about 20 computers ago) Now, I did solder an extra 256 KBs of ram into my C64 memory expansion module, but that was back in the day.
I would echo this endorsement. The 1.1RC release is significantly more polished. Font handling seems to be much cleaner, and the package is much faster overall.
Though I do agree with you in principle, ICBMs were designed with a warhead size and weight that are smaller than typical convential munitions. Modern mirvs are well under 1000 lbs.
Rigging a single ICBM even with 8 mirvs in the ~600 to ~800 lb range is pretty ineffective use of single-use extra-atmosphere delivery system. That would be less than 4 tons of munitions on target. I assume the that bomber would be a real bomber and carry several times that capacity.
Dual P3 motherboards with onboard eepro100. Serverworks chipset. I think they were Supermicro 370DLE. Those were the only dual chipsets with decent memory bandwidth for 933 MHz chips. The BIOS that came on them was exceptionally buggy. I wasted almost an entire week trying to get PXE to work...it makes me angry just thinking about it. But that is good...it just reminds me of important lessons learned: No More Diskless Clusters and Tyan instead of Supermicro.
It is automatic? Try it sometime with a four or five dozen systems and looming deadline. Try flashing a new BIOS onto 60 motherboards because there is a bug in the PXE code that shipped. The list of potential problems is pretty long.
Boot hard drives are cheap (~$50) and uses the boot mechanism that is most stable and well validated. People building ~1000 CPU clusters may well justify the cost savings and the additional setup work to get all of the kinks out of setup. People building "normal" clusters (10 CPU_COUNT 100) probably can't.
These boxes had one of the extra-buggy version of PXE that just would not work at all. Ended up using a etherboot floppy in each system to get it up. I built shared, read-only/usr,/sbin,/bin, etc and/var,/etc, and/tmp came off of independent nfs shares. Maybe using ram drives for these would have helped.
The tftpboot thing was a bit of a mess to set up. Then Redhat simply did not want to let the normal boot process work, so I had to rewrite rc.sysinit basically from scratch.
Plus, I had to hack in a sleep command into the Etherboot floppy to keep all of the systems from booting simulateously. Doing so thrashed the NFS server and deadlocked the whole startup.
All together, I probably spent two full weeks getting that working correctly. Compare that to a netcat-based disk-cloning installation on an 80-node cluster that took me about 3 hours.
There are no doubt canned boot installs now, but when I did this two and half years ago, there were none that would work with software we use on the cluster. I got to see the "inside" of how all of this works and with 500+ nodes set up over 5 years, I can say with some certainty that I never want to do it again.
I built a 60-node cluster about two years ago that originally had boot drives in each node. Since the gubment bought the boxes from the lowest bidder, we got very poor quality hardware to work with and about 75% of the hard drives died within a few months. To get things working again, I had to do the diskless boot thing. It was not at all fun. I can't see any advantage to that approach at all and I certainly wouldn't do it again by choice.
Re:What's ultrasonic communitation?
on
Reflections
·
· Score: 1
"Time-reversal" probably refers to the deconvolution that must be done to isolate the different channels. The essential idea of multipath transmission is that each signal will reach the different antennas at slightly different times. So, each antenna gets all of the signals, but there will be a slight time shift for each part. Ti find the original signals, you have to sort out the time shifts. Hence, "time reversal."
On-the-fly recoding of an 720p or 1080i HD stream to an equivalent resolution mpeg4 file will take a lot of CPU power. I don't think that custom hardware to do even NTSC resolution realtime transcoding to mpeg4 exists. I know that you can just barely do it in software with the fastest P4s and Athlons.
I would bet that the tivo will just be dumping the OTA and DirecTV HD streams right to the hard drive. Doing anything else will require alot of custom hardware.
Re:But you should use one anyway...
on
IDE RAID Examined
·
· Score: 2
That is really the key point. Hardware does not have some huge advantage over software. Many of the low-end "hardware" solutions are just software RAID buried in the drivers.
Good hardware RAID cards will have NVRAM to hold changes that may have occurred prior to a power outage or system crash. This is widely touted as an advantage, but drives all have huge caches...WDs have 8 MBs now. NVRAM is not going to save that 8 MBs per drive once it is in the hard drive cache. For an OS crash, there isn't a difference. The drive should finish the write even if the OS is fubar. BUT, when someone unplugs your raid system from UPS, the writes will be incomplete no matter what the controller.
Slashdot Mirror
It's a supply and demand thing. One thing that you might also want to to worry about is those "schools" churning out paper MCSEs month after month, advertising big $$$ and life on Easy Street by passing a few tests and getting a few certificates. In an already overcrowded tech market, these places are turning out tons of folks with overblown expectations. Once their dreams are crushed, who knows how cheap they'll be willing to work?
To be fair, colleges and universities all over the US have been doing the same thing for at least 20 years. The degrees aren't high-tech; they are english, art history, etc. I'd love to have spent 4 or 5 years in school only to graduate and find that I can make more money as a waiter than in my "profession," assuming I could find a position.
I have to agree here. My partner and I started a successful business right out of grad school doing consulting work and building clusters. We did it without VC funding and haven't really needed to borrow much along the way except to buy parts for larger projects when we couldn't fund it internally. This will not make you Bezos rich, but it isn't a bad life. Honestly, I don't know how much faster I would want to grow. More money means more headaches. Just being comfortable and busy is good enough for me.
A cheap Duron, hard drive, case, TV-out capable video card, and TV card is going to probably cost more than a stand-alone Tivo, so you are only saving on the "backend." I like the progress that I have seen in MythTV and Freevo, especially the integrated features like emulators and such. That is the appeal for me to build these type of solutions, not price.
FWIW, I am a DirecTV subscriber and the DirecTivo gives you a lot of functionality that you just can't easily replicate. It stores full quality video from the satellite feed on the hard drive. It also allows you to record two shows at the same time. That makes it well worth the price. Of course, mine has a 120 GB hard drive hacked in to give over 100 hours of storage.
Now I just need to figure out how copy the video from the Tivo. I can ftp and telnet into the system, but I haven't investigated the state of the extraction software lately. One of those projects I need to get to...
A "pea" travelling at 90% of the speed of light contains a lot of kinetic energy. Say, 0.01 grams for the pea at 2.7e8 m/s. That works out to 7.3e11 J. That is about the same energy as exploding 175 tons of TNT per pea.
Set aside the issue of engineering the "peashooter" to fire them, you are talking about throwing some potentially destructive material at a neighboring star system. Firing them continuously looks like you intentially want to hit something. I think this might be a bad idea from a "just saying hello" viewpoint.
The problem with sending humans and cargo together is that it is cost effective to use "crappier" rockets to send up fairly replaceable cargo and the "good ones" to send up people. Generally, people are not replaceable, at least in most political terms. The human transport needs to over-engineered and super-certified safe. A BDR can blow up once in a while. It would suck, but at least it wouldn't derail the whole space program for three or four years, like Columba.
In the article, he recommends retooling the boosters and tanks with a new crew module and a separate cargo module. While I can see his point with regard to reuse of existing proven technology, I can't quite get away from the idea that sending people in the same craft as cargo is just a bad idea. Big Dumb Rockets (BRDs) are the only way that we should be putting cargo of any sort into space. We have become very good at orbital rendevous manuevers, so I can't see that separate launch vehicles for people and cargo will present a significant complication.
Perhaps the booster and tanks can be recycled with ONLY a crew module that can actually reach the L1 point. The current shuttles can barely make it to low orbit.
On the whole, he is right. An L1 base would be a nice permanent move into space and is probably something that should have been done in the mid-70's. The establishment of a moon base will be an easier political sell though. Once we hammer out manufacturing techniques, it should be possible to grow a spawling complex on the moon without needing to carry everything from earth. And you know we Americans love to spawl. If we can find water in sufficient quantities and are willing to take nuclear reactors with us to the moon, the fuel for future space flights will probably come from the moon.
I can think of another area where code generation could be very useful...numerically intensive tasks. There are lots of C optimization techniques (register and cache blocking especially) that are architecture, CPU, and memory subsystem dependent. To optimize for different platforms, you normally need to do explicit loop unrolling and hardwire blocking factors in the code.
Code generation allows a "higher level" description of the problem in some abstracted language to minimize replication of common code and also allows adaptive selection of blocking factors, etc. to maximize code performance. The Altas BLAS library is an excellent example of the power of the latter which amounts to per-system adapative code generation. The C Preprocessor and C++ Templates are not sufficiently general to allow this same sort of optimization without some code generation segment.
It has been "coming" for several years now. Really, it is pretty pathetic. I outted the bucks for an HD projector, the Sony HD DirecTV receiver, and an amplified UHF antenna rig. With a lot of work, I was able to get ABC, CBS, FOX, PBS, and NBC to tune in from the broadcast towers in Pittsburgh.
After all of that work, the *only* channel that has decent HD content is PBS...and there are only a few shows that run all of the time. FOX is completely pathetic. The real networks run only a few shows a week. Even so, the prospect of being able to timeshift those few shows is pretty compelling, so I might need to build a MythHDTV box...Tivo has pretty much ruined me.
There are only a handfull of HDTV channels on direct-broadcast satellite feeds. DirecTV has HDNET, Discovery, ESPN, HBO and SHO in HD. The Dish lineup is similar. There is just not that much Satellite HD content available.
The DVB system that you pointed to is a nice setup, but there are no PC add-in cards that I know of that allow access to DirecTV or Dish digital feeds whether SD or HD. The one exception might be via DirecTivos which can be hacked to allow extraction of the video feeds, but this is as a postprocessing task, not realtime. Perhaps there are add-in cards for Big-Ugly Dishes to decode HD broadcasts, but I am not aware of them.
I believe that Sony is planning to produce an HD DirecTivo receiver shortly for timeshifting HD content. I don't know if they will do a better job of hack-proofing that unit than with other Tivo units, so offline HD extraction is still an open question.
Advanced degrees are for purely academic fields.
I need to disagree with this statement. In general, most serious engineering design involves many PhD level people, either in managing the MS/BS guys or in solving really hard problems. Walk through GM or Ford or Nvidia or Intel. There are lots of Drs. around.
Now, a PhD in I.T. sounds overly broad. The area of specialization is key with advanced degrees. My advanced degree is not in engineering or even mechanical engineering...it is in computational fluid dynamics. A PhD in security or networking or algorithm design could be highly useful and lead to well paying positions doing that sort of work. A PhD for someone who is changing network cards and installing Windows service packs is a complete waste...
It is called bi-linear filtering...instead of making pixels larger when zooming, it interpolates to fill in the in-between pixels using the four nearest "real" pixels. There is a fair amount of linear algebra involved and that made it impossible for CPUs of that era to do it in real time. 3DFX was noteworthy for being the first hardware manufacturer to implement bi-linear filtering in hardware (Voodoo 1).
</useless-info>
There is still good TV around, but it costs...HBO & Showtime. Dead Like Me on Showtime is my new personal favorite.
You are right; it is not possible to do computation without SOME losses even if we use completely reversible phenomena. The entropy increase comes when we "forget" information, .i.e., clearing a register.
That is different from what they are talking about in the article. Their goal is to move to essentially reversible reactions using spin rather than current-type electronic phenomena that contain Ohmic irreversibilities. The Ohmic losses dominate the heat generation in current ICs. The next on the list of energy loss for an IC is probably RF radiation. Entropy production from information loss is pretty far down the scale, but it is the one that cannot be "engineered away" so that is why it is always included in the "how big can a computer get" calculation.
What we need is an open-source distribution of software for Windows. It would install Mozilla, OpenOffice, GIMP, and whatever else seems useful. Wrap it up with a nice installer that makes the "open" nature of the software clear. Configure everything so that a mindless XP user can just run the installer and get all of the best that the opensource folks have to offer. It could become a very popular thing if it was all in one place and easy to find.
I can see a market for retail sales as well, so it could be worthwhile for a for-profit company to pursue this. I would love to be able to hand windows users a CD with an Open Source bundle of software and know that they will probably be able to explore and use the software without much handholding.
There is a BIG reason to roll your own that the article misses. WEP is pretty pathetic from a security viewpoint. To my knowledge, the cheap WAPs do not support IPSEC or other encryption layers natively on top of the wireless layer.
Last year, I was investigating "hackable" WAPs that I could reprogram using Linux or *BSD to enforce a layer of encryption at the first entry point into the network. Otherwise I would need to string a separate wired LAN (or VLAN) for the unsecured network connecting access points and then funnel everything to a firewall for the encryption. Unfortunately, this would still allow someone to hack on the other wireless clients, etc.
I am a little surprised that with war-driving, -flying, -chalking, ad nauseum, wireless security still seems to be a non-issue for most people.
IC extractors. I mean, who uses this stuff?
BIOS chip swaps...if you hose a bios chip and need to "hotswap" and flash in another system, those pullers come in handy. But, to be fair, I have only done that once since my first 286 (about 20 computers ago) Now, I did solder an extra 256 KBs of ram into my C64 memory expansion module, but that was back in the day.
I would echo this endorsement. The 1.1RC release is significantly more polished. Font handling seems to be much cleaner, and the package is much faster overall.
Though I do agree with you in principle, ICBMs were designed with a warhead size and weight that are smaller than typical convential munitions. Modern mirvs are well under 1000 lbs.
Rigging a single ICBM even with 8 mirvs in the ~600 to ~800 lb range is pretty ineffective use of single-use extra-atmosphere delivery system. That would be less than 4 tons of munitions on target. I assume the that bomber would be a real bomber and carry several times that capacity.
Dual P3 motherboards with onboard eepro100. Serverworks chipset. I think they were Supermicro 370DLE. Those were the only dual chipsets with decent memory bandwidth for 933 MHz chips. The BIOS that came on them was exceptionally buggy. I wasted almost an entire week trying to get PXE to work...it makes me angry just thinking about it. But that is good...it just reminds me of important lessons learned: No More Diskless Clusters and Tyan instead of Supermicro.
It is automatic? Try it sometime with a four or five dozen systems and looming deadline. Try flashing a new BIOS onto 60 motherboards because there is a bug in the PXE code that shipped. The list of potential problems is pretty long.
Boot hard drives are cheap (~$50) and uses the boot mechanism that is most stable and well validated. People building ~1000 CPU clusters may well justify the cost savings and the additional setup work to get all of the kinks out of setup. People building "normal" clusters (10 CPU_COUNT 100) probably can't.
These boxes had one of the extra-buggy version of PXE that just would not work at all. Ended up using a etherboot floppy in each system to get it up. I built shared, read-only /usr, /sbin, /bin, etc and /var, /etc, and /tmp came off of independent nfs shares. Maybe using ram drives for these would have helped.
The tftpboot thing was a bit of a mess to set up. Then Redhat simply did not want to let the normal boot process work, so I had to rewrite rc.sysinit basically from scratch.
Plus, I had to hack in a sleep command into the Etherboot floppy to keep all of the systems from booting simulateously. Doing so thrashed the NFS server and deadlocked the whole startup.
All together, I probably spent two full weeks getting that working correctly. Compare that to a netcat-based disk-cloning installation on an 80-node cluster that took me about 3 hours.
There are no doubt canned boot installs now, but when I did this two and half years ago, there were none that would work with software we use on the cluster. I got to see the "inside" of how all of this works and with 500+ nodes set up over 5 years, I can say with some certainty that I never want to do it again.
I built a 60-node cluster about two years ago that originally had boot drives in each node. Since the gubment bought the boxes from the lowest bidder, we got very poor quality hardware to work with and about 75% of the hard drives died within a few months. To get things working again, I had to do the diskless boot thing. It was not at all fun. I can't see any advantage to that approach at all and I certainly wouldn't do it again by choice.
"Time-reversal" probably refers to the deconvolution that must be done to isolate the different channels. The essential idea of multipath transmission is that each signal will reach the different antennas at slightly different times. So, each antenna gets all of the signals, but there will be a slight time shift for each part. Ti find the original signals, you have to sort out the time shifts. Hence, "time reversal."
On-the-fly recoding of an 720p or 1080i HD stream to an equivalent resolution mpeg4 file will take a lot of CPU power. I don't think that custom hardware to do even NTSC resolution realtime transcoding to mpeg4 exists. I know that you can just barely do it in software with the fastest P4s and Athlons.
I would bet that the tivo will just be dumping the OTA and DirecTV HD streams right to the hard drive. Doing anything else will require alot of custom hardware.
That is really the key point. Hardware does not have some huge advantage over software. Many of the low-end "hardware" solutions are just software RAID buried in the drivers.
Good hardware RAID cards will have NVRAM to hold changes that may have occurred prior to a power outage or system crash. This is widely touted as an advantage, but drives all have huge caches...WDs have 8 MBs now. NVRAM is not going to save that 8 MBs per drive once it is in the hard drive cache. For an OS crash, there isn't a difference. The drive should finish the write even if the OS is fubar. BUT, when someone unplugs your raid system from UPS, the writes will be incomplete no matter what the controller.