Dangerous "Eject this device" in taskbar...?

X

xX_Jack_Carver_Xx

2[H]4U
Joined
Jun 6, 2005
Messages
2,542
On new NF4 motherboards, using the SATA raid array, you end up with WinXPpro (assume all others too) putting the usual "Eject this device" iconin the taskbar notification area. Just like with a USB thumbdrive.


Only this is...how BAD it would be for you to "eject" 1/2 of the raid0 array by accidently clicking one of the SATA drive items.

Anyway to get WinXP from placing this "option", where a user might accidently click it, off in some safe place? Can this be disabled? Esp disabled without losing the normal ability to eject items you really WANT to, like a thumbdrive?
 
I'm not sure if it would corrupt the array or not, but then again, I have to ask.. If your worried about your data, why would you be using RAID0 anyway?
 
The SATA driver is marking those drives as removable. Make sure you have the newest SATA drivers.
 
Isn't that by design, though, to make them hot-swappable?
 
djnes said:
Isn't that by design, though, to make them hot-swappable?
Click to expand...
Yes.

edit: but I think Ranma_Sao is suggesting the most recent drivers for other reasons (new feature to the chipset IIRC).
 
I doubt it'd allow you to "eject" a drive that's in use. The only way I can see anything bad happening is if it's possible for the RAID-drivers not to tag the drives as in use. (Which is unlikely.)
 
HHunt said:
I doubt it'd allow you to "eject" a drive that's in use. The only way I can see anything bad happening is if it's possible for the RAID-drivers not to tag the drives as in use. (Which is unlikely.)
Click to expand...
That's a good point as well. If your using a file from a thumb drive, XP doesn't let you "eject" it.
 
Under no reason should a driver ever think removing one part of raid-0 in the name of hot swappability is a good idea. ;) Raid-1, sure, Raid-5 sure, Raid-0?

The drives should be marked as fixed disks, and the controller should support hot swapibility, windows should be ignorant that the controller allowed you to party on the disks. ;)
 
Heh, OK, so I didn't think that one out... LOL hot swap RAID-0, tehehehe.

The driver probably isn't smart enough to disable hot swap on RAID-0 since it's available for non-RAID drives on that mobo.

While we're on RAID-0 you (OP) realize there is little to no benefit to RAID-0 on most desktops, right? Add to the fact your 1/2ing your MTBF on the HDDs, it's generally not a good idea.

Got backups? ;)
 
Well, it is still the simplest way to get the apperance of one big drive from two smaller ones, to reduce space-juggling.
(Not that I use anything but JBOD. :D )
 
re:Raid0 ....Apparently you guys dont "game" much? Ive done a side by side comparison with the games I play, and level loading. It was well worth implementing the raid array. The 2x Raptor raid0 blew away a single raptor in level transistions in WoW and Farcry and loading AArmy maps.

Yes I have backups, but I dont use my gaming machine for anything but anyway, so if it quit I'd be fine and just reinstall.

I've also been running this NF2 machine on raid0 (2xraptor) for over a year without incident. Frankly I just dont see the whole "fragility" you seem worried about. Sure an enterprise/business install with critical data, raid5 it, but for a simple game machine....Im fine.

I have the latest NVidia NF4 sata raid drivers, they allow this "hotswappable" ability to show up. Wonder if there is a setting somewhere that I can turn it OFF with. Meantime Ill just have to be careful using a thumbdrive.

Heh... lets start a real rowe..... how about using that new Gigabyte iRam board with 1-2G of ddr for the Windows pagefile!!!! Ill jump on that as soon as it is out. NVidia will prolly list THAT as hotswappable too ;-)

In fact the only thing Im really steamed about is why WDigital doesnt make the Raptors SataII drives.
 
Umm no offense, not like you spared us, but we all pretty much game here.

Are you having load races? Or are you really interested in overall performance?

Here's a simple equation. MTBF+RAID 0=MTBF/2. This is a factual equation. HDDs don't fail every year, I'm glad you have been running OK for that long; however; that doesn't change the equation, at all.

You don't have to appreciate our experience to listen, you just need to listen with an open mind. Trust me, you, and I don't know everything about the OS. No single person does, not even a single MS employee. That means there is something to learn from almost everyone.

When you understand that you'll listen to "noobs" as well as seasoned members. I've learned a mountian of knowledge from both.

http://www.anandtech.com/storage/showdoc.aspx?i=2101
 
xX_Jack_Carver_Xx said:
re:Raid0 ....Apparently you guys dont "game" much? Ive done a side by side comparison with the games I play, and level loading. It was well worth implementing the raid array. The 2x Raptor raid0 blew away a single raptor in level transistions in WoW and Farcry and loading AArmy maps.
Click to expand...
Maybe you missed the part in school about the Scientific Method? Where experiments that can be duplicated over and over lend themselves to fact?

Many of us on here, and several other sites did actual side by side testing. Mine in fact, was done using a single 36 GB Raptor versus 2 in RAID0. The fact is, the performance game is absolutely negligible. It was repeated over and over by many, and the results all came to the same conclusion. As Phoenix mentioned, your also cutting your MTBF in half, and taking a huge risk with any data on the drives. Argue with factual results if you may, but the argument of "it seems faster" doesn't hold much weight.
 
Well my test was a simple one. Two identical Sil3112 sata based NF7-S boards, same CPU/video/DVD/memory/settings/voltages/powersupplies/WinXPpro where system1 had a single raptor, and system2 had 2x raptor raid0.

Installed the OS (WinXPproSP2)/drivers, installed Farcry, installed WoW, installed EQ.

Then played side by side on the same server (multiplayer)/same level. Whether it was transporting to a new zone, or loading a new level, or loading a multiplayer map in FC, the raid0 was nearly TWICE as fast. Period.

That is side by side, putting my character in the same place, and making the level/zone transition.... ding, raid0 machine arrives in nearly half the time.

I dont know how much more SCIENTIFIC a test I can construct to illustrate the direct effect on the games between single raptor, and 2xraid0 raptors.... all 3 raptors are form the same batch too.

Now YOU may think it just doesnt matter, but Ill give you a direct example of where it DOES matter. Playing FarCry multiplayer, on map transitions, the player who LOADS and spawns fastest gets to grab the BUGGY/HUMMER first. With the raid0 raptors _I_ spawn first.

I really dont get the whole anti-raid bent you folks have but Ill pass on the sage advice and keep on keepin on... BTW, Im moving to 4x raptor raid0 in a couple weeks :D But Im not worried about "data loss", its a gaming PC, takes all of 30min max to install everything all over again, and I havent even bothered to implement Ghost yet. So if a MTBF, god forbid, hit me, I guess I'd just have to suck it up and re-install. Sure a BUSINESS with CRITICAL data might fear raid0, or a very lazy user who never backsup might also. However for typical gaming systems there is little threat of "losing" anything really important, and frankly ANYTHING really important should be BACKED UP anyway, otherwise I guess it wasnt REALLY important.

Ill be transitioning from my DFI Infinity w/Sil3112 PCI card to my new NF4 SLI-DR with 4 raptors on the NF4 sata controller. Ill stick the Gigabyte iRam on the Sil3114, and keep my ATA133 Maxtor on IDE. BTW, the raid0 on the Sil3112 was running for a year on my NF7-S, and I transitioned to the Infinity without a hitch by moving the array to a Sil3112 PCI card first. The OS install has been rock solid and problem free for a very very long time. The Sil3112 PCI card doesnt suffer fromthe write-speed bug the Sil3114 does on the Infinity so I left it installed.

As I said, the only thing that really pisses me off is WD not making raptors a sataII device, since 4x hitachi 80G sataII ($64) drives beats the raptors on a sataII controller in sustained transfer rate. But the raptors kick butt in random access. Slap a sataII controller board on the raptor drives and its off to the races.

To get back to the original question..... So there is no way to get the NVidia drivers to mark the drives FIXED and simply NOT list them in the "ejectable" items icon in the taskbar notification area? Ill just have to BE careful then. I AM using the latest drivers from NV on the NF4. Oh well, thansk for the help anyways guys.
 
I guess we should all alert Anand that he was wrong. :rolleyes: Did you happen to notice that most of the boutique PC makers stopped offering RAID0 configs by default as well? Facts are facts, whether you chose to believe them or not. They do however, remain factual.
 
The only "proven" usefulness for RAID0 has been in digital editing, movie editing, etc which needs to swap rather large files. That is the reason that my home system is setup with RAID0 (for video editing from my camcorder). I also have Norton Ghost running daily backups to a removable HDD. The only data I expect to lose in a HDD crash is some game saves.

Unless Im completely mistaken with my grasp of statistics, and my friend the statistics god, having two MTBF's that are identical leave the sum as identical, not MTBF/2. Thus a RAID0 array has the same MTBF as a single drive.
 
"I guess we should all alert Anand that he was wrong."

Perhaps his test was flawed. What EXACTLY was the system doing during his "timing" period? Do you know? Does HE? Was the CPU utilization of the controller/driver so high that the effect of disk transfer speed was lost in the envelope of system activity?

I think MY comparison experiment was valid and I will stick with my decision to use raid0 for my gaming applications.

As to what Alienware, et al decide to sell..... I am not in the profit making business, I build MY own systems for ME at the lowest $$ and highest performance that the budget allows. I just built my brothers new system (and am about to assemble my own version):

3700+ SanDiego, DFI Ultra-D, 2GB's of VX/UTT ram, X850XTPE, 2x Raptor 36GB, 80GB Maxtor IDE (with WinXPproSP2 installed as backup), 520W OCZ Powerstream PSU, AMS CF1009 (no window) case, mod'd with 120mm intake fan, 120mm memory/mosfet fan internal, NEC 3520A DVD burner..... all for a grand total of $1558. He also got a Samsung 930B 8ms LCD for $340 from Bestbuy. So $2Kish total.

Let Alienware/DELL/etc/etc beat that price/performance.

Its running on stock retail AMD HSF, 250Mhz x11, 1.502Vcore, 3.3Vdimm, 8,2,2,2.0 2T 250Mhz memory/HTT 1:1, purr's like a kitten, StressPrime/SuperPI/3D stable 100% in 90F ambient air temp, CPU peaks at 50C. 2750Mhz was the design goal, and was met with room to spare. Ill be implementing a dangerden WC setup and looking for 3.0Ghz myself. The case is big enough for a 2x120mm BlackIce raddy to fit INSIDE, with seperate airflow chambers/paths for raddy, PSU, and disk/motherboard area.
 
xX_Jack_Carver_Xx said:
"I guess we should all alert Anand that he was wrong."

Perhaps his test was flawed. What EXACTLY was the system doing during his "timing" period? Do you know? Does HE? Was the CPU utilization of the controller/driver so high that the effect of disk transfer speed was lost in the envelope of system activity?

I think MY comparison experiment was valid and I will stick with my decision to use raid0 for my gaming applications.

As to what Alienware, et al decide to sell..... I am not in the profit making business, I build MY own systems for ME at the lowest $$ and highest performance that the budget allows. I just built my brothers new system (and am about to assemble my own version)
Click to expand...
Yeah man, you like what RAID 0 brings to the table and have the knowledge to fix a problem should it ever arise. If nothing absolutely needs to be backed up and you're only worried about speed (even marginal gains) then who's going to tell your differently?

If it's faster to you then it's faster. I also seem to remember people debunking Anand's article sometime last year.
 
Jack, how about you just read the article I linked? Anand even evaluates load times, and yes indeed it was faster loading games. But, you need to read the whole article to understand why it's not a great idea.
 
Okay, in terms of the MTBF, if one drive goes in a two drive RAID0 array, your data is toast on both drives, essentially leaving you in the same boat as if you only had one single drive, and it failed. Since your relying on two drives, your taking twice the chance of loosing your data.

Now if you add more drives to a RAID0 array, if the controller allows it, your decresing your MTBF even more, making your data even more at risk.
 
It has been brought to my attention that there is a debate about RAID-0 going on. Perhaps I can comment on this matter.
DougLite said:
What of small files, which a game typically has many of? As an example, there are 2306 files in my UT2004 directory. 1,100 of them are less than 128KB in size, over 1,000 of them are less than 64KB. If you build a RAID-0 array with a 64KB block size, 40% of the files are smaller than the block size, and will only be loaded by one drive anyway. A 7200RPM drive typically services a read request in 12-13ms (8.5-9.3ms seek depending on brand, plus 4.2ms latency) - a Raptor completes one in less than 8ms (4.5ms seek, 3ms latency).

Presuming no fragmentation and a straight linear transfer occurs after the drive(s) find the beginning of the file, If the RAID-0 array transfers at 120MB/sec and the Raptor transfers at 60MB/sec, then the RAID-0 array gets the 64KB file transferred in less than half a millisecond, while the Raptor takes a whole millisecond - but that's still over 12ms for the 7200 RAID-0 against 9ms for the Raptor. If we load all 1000 of those files that are less than 64KB (not likely) that works out to a 3 second difference - the Raptor hacks through it in 9 seconds, while the RAID-0 array takes 12 seconds. This makes the Raptor 25% faster, and that's even giving the RAID-0 a generous double sustained transfer rate advantage.

Of course, this presumes that the files are scattered all over the disk, making the seeks 'average' out to the drives' actual seek time. Given that the files will be somewhat localized, the Raptor's swifter track to track seeks, solid buffer strategy, and lower latency will give it an even bigger advantage in loading small files. WD specs track to track seek on a Raptor at .6 ms, while the same track to track seek on a Caviar SE is rated at 2.0ms, making the Raptor 3 times faster in track to track seek, a gap that no amount of sustained transfer rate is going to overcome when working with that many small files. Sure, the RAID-0 will gain ground when working with the 346 files that are larger than 8MB in my UT2004 directory, but even if it does get through them twice as fast as the Raptor (optimisticly), it's not going to make up taking longer on the many small files that must be loaded as well. BTW, FarCry is even worse - 825 out of 1163 files are less than 128KB in my FarCry directory. Furthermore, XP start up is under much the same conditions - 6425 files in C:\Windows\system32\, 4,482 of them less than 128KB - more than 5 out of every 8, and only 10 files of more than 8MB.

Sure, RAID-0 will give the Raptor a run for the money if you're doing linear reads and writes, such as with straight file transfers and content creation work, but it falters when loading games. Pure and simple.
Click to expand...
I posted that in this [thread=912316]thread[/thread] when another member asked if two 40GB 7200 drives in RAID-0 were equal to a 73GB Raptor. As you can see, they are not. HOWEVER, in the case of WoW, the vast majority of its data is stored in eight MPQ files, ranging in size from 68MB to right at 1GB. Depending on how WoW is coded, it may very well exhibit the linear access patterns that allow RAID-0 to scale well. Doom3 is much the same way, a handful of huge compressed files that hold the game data. But what of Far Cry, UT2004, and the like, with hundreds of files that are smaller than a typical RAID-0 block size? What happens then? Frequent seeks, and your RAID-0's high sequential transfer rate becomes a NON factor. What of the increased cost of each additional spindle? A 73GB Raptor is $180 - that's the difference between an X800XL and X850XT, or Corsair Value Select and Corsair XMS XL. RAID-0 may scale well in some games. Certainly not all of them, and it certainly is very poor in the return on investment category - double the investment != double the performance, ie load times cut in half.
 
On the file size issue, a couple things.

First, Im using raid0 with raptors, not normal HDisks. Though raid0 with 2x Hatachi sataII drives is pretty damn fast.

Second I use a 16K/16K stripe/cluster, so most of the files are above this.

Third, the complex issue of the HDisks internal buffer ram comes into play, where if a 20K file forces a 128K access, and the next 30K access is sequencial, then its already in the buffer ram, and has near 0ms access time.

The interactions can be quite complex systemwide.

Again, all I can say is Im using it and its quite fast, clearly faster than a single drive, playing the same games side by side on identical systems.
 
First, for the OP:

It used to be, before RAID BIOSes got smart, that allowing one of the drives to be disconnected on the fly was a very bad thing. The BIOS would keep chugging along even with one of the members missing, and as soon as a write went thorugh with a disk missing, that data was toast for sure and you most likely wrecked the whole file system.

Nowadays, if the RAID BIOS detects that a member is missing, it will grind everything to a halt, and give a strongly worded error message to the user with the options of continuing to boot without, shutting down the system to make sure evrything is connected properly, or destroying the array and continuing.

Also, when you issue a stop command, Windows SHOULD force dismount any volumes on the disk and make any handles to those volumes invalid, which is of course impossible on your system drive. Not a big deal. I wouldn't worry about wrecking the data with this, although my saying so is no excuse to slack on backups.

Now continuing the RAID-0 debate:

True. RAID-0 Raptors will definitely be better than 7200 drives in RAID-0.

With the 16K stripe size, you are actually increasing the number of seeks that must be performed when working with the smaller files, as now both disks must work to retrieve any file larger than 16KB. Once again, the goal with RAID-0 is to keep seeks to a minimum and leverage the transfer rate boost as much as possible.

Also, it is true that the buffer(s) and read ahead strategies on the disk are working to increase throughput. However, having RAID-0 actually increases the probability of rotational latency impacting the read service time. On a single drive setup, the latency is either above or below the 3 ms nominal latency of a 10,000RPM spindle - a 50/50 split. However, if you introduce a second spindle in RAID-0, then you increase the chance of a high latency impacting performance - if either drive has a high latency hit, then both drives have to wait, giving you a 75% chance of higher than nominal latency with two spindles in RAID-0 instead of the 50% chance with SLED. You would have to transfer 192KB at 120MB/sec to recover from a 3ms rotational latency hit to draw even with a single spindle at 60MB/sec. That means three 64KB files from your RAID-0 must be transferred at twice the speed of the SLED to make up for one rotational miss where the platter must go all the way around before the read request can be completed. Given that a rotational miss will occur three out of every four files on RAID-0 while only two out of four on a SLED, those aren't very good odds. Those rotational misses can pile up pretty quickly, even when the buffer helps. However, the buffer's impact is somewhat mitigated by the optimistic estimate of doubling sustained transfer rate by going to RAID-0 - actual gains are more in the range of 50 to 75% If you only increase throughput 50%, then you must make six transfers to close the gap - by which time another rotational miss has occurred that the SLED didn't suffer from. The solution to this problem is no secret - Command Queuing - but it introduces additional overhead that takes away from throughput. Command Queuing has been used for years in SCSI drives to mask the impact of rotational latency under random workloads, and it can work similar wonders under highly random desktop loads.

So yes, RAID-0 is an extremely poor choice for frequent localized seeks exhibited in many game loading sequences, and most other desktop tasks.
 
DougLite said:
So yes, RAID-0 is an extremely poor choice for frequent localized seeks exhibited in many game loading sequences, and most other desktop tasks.
Click to expand...
This phrase should be made into a sticky.
 
Still, did anyone notice that he's happy with the performance in WoW, which doesnnot use a lot of small files?
He just might have found a game with an access pattern that works well with RAID0.
Still, I'm not going to benchmark it to check. :D

(I'll do the rest of the 2003 v XP benchmarks when I'm done with my exams, btw.)
 
Ah ha, but now we come to the whole crux of my argument - that RAID-0 does not boost performance in everything. RAID-0 works wonders for content developers that use disk intensive programs such as Photoshop. It works great for users that make disk limited frequent saves and loads when editing video. It even accelerates some desktop usage patterns.

However, many gamers don't want to shell out the cash for an SLI setup, because SLI has limitations. It only helps some programs. It only helps in really intense resolutions and eye candy modes. Two SLI'd mid range cards are slower than one high end card in all but a handful of tasks.

Now, the people that say SLI is a joke need to take a look at RAID-0, and realize it has the exact same limitations. Like SLI, it doubles cost. Like SLI, it only increases performance under intense loads in certain programs. A 10,000RPM Raptor schools 7200RPM drives in RAID-0 in all but a handful of tasks, just like a 6800 Ultra is faster than 6600GT SLI in all but a handful of games and settings. Like SLI, double the investment does not net double the performance.

Now, RAID-0 has some drawbacks that aren't in common with SLI. The biggest is the reduced reliability of the array. The second is that you can't just change a driver profile and make RAID-0 magically scale well. If your programs have access patterns that aren't conducive to RAID-0 scaling, you are stuck with it.

You guys can understand that SLI doesn't make sense for everybody, why can't you understand that RAID-0 is the same way? RAID-0 is suited for only certain usage models, most of them seen in workstation and server environments, rather than on the desktop. Some people have the cash to throw down for SLI/RAID-0/SMP/etc. Many that do feel compelled to justify their investment, but the fact remains that RAID-0 has very real limitations, including performance limitations, that must be considered before you make the decision to double your storage costs.
 
You could also make the very same argument for dual processor machines. I have a few former co-workers who built dual processor Xeon machines thinking it would make a huge differnce in gaming and most other tasks. I mean, two is better than one, right? Little did they know it made absolutely no difference at all. None of their games were coded to utilize a second processor. Neither were any of there applications, so they really saw no benefit of having that second processor, and most definitely, it didn't justify the cost of the extra processor and expensive boards.
 
djnes said:
You could also make the very same argument for dual processor machines. I have a few former co-workers who built dual processor Xeon machines thinking it would make a huge differnce in gaming and most other tasks. I mean, two is better than one, right? Little did they know it made absolutely no difference at all. None of their games were coded to utilize a second processor. Neither were any of there applications, so they really saw no benefit of having that second processor, and most definitely, it didn't justify the cost of the extra processor and expensive boards.
Click to expand...
I love my dual procs. ;) I did have Raid-0 for awhile but busted it for exactly the reasons posted. However, dual procs did make a difference, and still do for me. The thing Dual Proc gave me was life time as well. Dual 1.2's lasted me almost 3 years, where a single 1.2 I would have replaced so fast. ;) But I'm not the average user either, encoding 3GB file to 100MB video file, dual proc is the way. Raid didn't make a bit of difference interestingly enough.

I have been tempted at work to see if RAID-0 SCSI fairs better then RAID-0 SATA (IDE). Espically since SCSI SLED's are so expensive.
 
Ranma_Sao said:
I love my dual procs. ;) I did have Raid-0 for awhile but busted it for exactly the reasons posted. However, dual procs did make a difference, and still do for me. The thing Dual Proc gave me was life time as well. Dual 1.2's lasted me almost 3 years, where a single 1.2 I would have replaced so fast. ;) But I'm not the average user either, encoding 3GB file to 100MB video file, dual proc is the way. Raid didn't make a bit of difference interestingly enough.

I have been tempted at work to see if RAID-0 SCSI fairs better then RAID-0 SATA (IDE). Espically since SCSI SLED's are so expensive.
Click to expand...
I am not doubting you had a reason to go with dual procs. The average user doesn't, and would benefit far more from a single faster proc. Certain tasks and apps that can utilize the second proc are the key.
 
djnes said:
I am not doubting you had a reason to go with dual procs. The average user doesn't, and would benefit far more from a single faster proc. Certain tasks and apps that can utilize the second proc are the key.
Click to expand...
Yeah, but with dual cores, the way of the future is paved! (I'm excited about dual cores, and dual threads at the same time if you can't tell.) More and more programs will take advanatage of multithreading.
 
I agree, the issues are complex, and its not just a casual... do it...for everyone. Im happy with my setup. My only annoyance is the Windows taskbar notification area "eject this" option. I find that scary to have laying around. Hoped for a cure, apparently there is noe. Maybe NVidia will fix it.

I agree on the dualcore, Im sure a 4800+ x2 is in my future, once they are a little less $$. Everything is ready for it, and they seem to OC as well as the SanDiego does. 250Mhz x12 sounds about right ;-)
 
You must log in or register to reply here.
Back
Top