pros / cons of stripe raid with more than 2 drives?

Graham

Hello :-)
Veteran
Supporter
I've just got my new machine working, and I have to say, I'm blown away by the difference a decent hdd setup makes. I've set up two 72gb 10k sata raptors, and I'm quite seriously considering going to 4 of them... effectivly 40k rpm... mmm

between the windows loading logo appearing and the login screen is currently 4 seconds. :p :LOL:

I will be using another (slower) hdd for automated backup of critical stuff btw (ie, code).

I'm severly tempted... But I do wonder if it would really be that benificial? Is there anything I may be overlooking?

yay! new pc = much fun-ness


(in short, would it be worth it? temptation is an evil thing...)
 
Last edited by a moderator:
I don't think 4 disks will give you nearly the increase that 2 gave you (especially if some of that increase was just going from 7,200 to 10,000 rpm drives), but it should still be worth it, especially since you double your capacity as well. Make sure you're not putting more than two drives on the same controller. Ideally each drive would be on separate controllers on a good quality expansion controller card... but I can understand not wanting to drop a few hundred dollars on that in addition to the cost of the drives.

Just make sure you keep up to date with those backups :)
 
What do you need the increased drive performance for? Surely it cannot be because you think you spend too much of your valuable time waiting for Windows to load.

4 drive RAID-0 might make sense if you are doing low seek, high streaming operations like digital video. Another reason it makes sense for this sometimes is because the life span of those files are very short. With 4 drives you are effectively reducing your MTBF by a factor of 4 and this is getting to the yuck-yuckity-yuck-yuck (technical term) end of the spectrum.

Perhaps you were thinking of two independent RAID-0 arrays for a grand total of 4 drives? Still, unless you need it, and then you would know, I suspect you could find better areas of your PC to improve upon. 4GB memory...a (second) huge and lovely display...I dunno...think of something! :p

Wrong forum BTW. See: Beyond Forums > PC Forums > Hardware Talk
 
er.. 40K rpm:LOL:
You remind me of a friend who thought that dual G4s= effectivly more mhz.
raid does nothing for seek time and everything for sequential reads/writes (depding and which level).
The problem with raid is that typical desktop applications (including games) don't really benefit from raw sequential speed.
However video editing is great use of raid 0:!:
 
righteo

well the reasoning is pretty simple really, my other machines have 7200 rpm drives, and I also also have a performance monitor showing both cpu and disk usage. An awful lot of the time when I was doing my usual stuff, disk usage would be 100%, cpu ~30%.

I guess I'm just trying to build the fastest pc I can, within reason. One that I will not get frustrated waiting on when developing and doing the various things I do (which usually isn't gaming, btw). I could have easily spent the same amount of money getting the 4800+ instead of the 4200+ I bought, yet I feel this by far makes a far larger overall performance impact. Thats why I ask if it's worth while going a step further.
I realise that raid 0 with 4 10k drives isn't 40k rpm :) but if it feels like it, well then maybe it's worth it. A pc is usually only as fast as the slowest part after all, and hdds arn't exactly quick.

The reason I put it in this forum was that originally, I felt like writing a 'hell yeah' post because I've only just got the thing running. But I kinda scrapped that. Also I am massivly jet lagged at the moment so I'm not exactly thinking clearly. :D
 
Before you go out and buy another two raptors, original poster, you should know the Hitachi Deskstar 7K500 is currently the overall fastest IDE drive in existence, as measured by Storagereview.com. The Raptor is aging badly right now, it's hardly been updated for what, one and a half years? Two? Its capacity is pitiful by today's standards, and buying two to get some decent room is seriously cost-inefficient.
 
radeonic2 said:
raid does nothing for seek time and everything for sequential reads/writes (depding and which level).

Wrong wrong wrong. Oh so very wrong.... :)

Optimizing seek time is actually the single most important reason why high-end servers use Raid configs.

When your first disk is busy reading file1, your second disk can already start reading file2 and your third and fourth disk can read file3 and file4.

You'll have to make sure that you make your stripe size as big as possible, so that there is a good chance that a small file will fit on a single disk.

In the time it takes to read a single file from 1 disk, you can now read 4 files from 4 disks. This will effectively drop seek time by 75%.

But of course this has not effect is you only need to read 1 file, so you need a situation where you have lots of simultaneous file access. (like on servers, or desktops with applications that are really heavy on the disks)
And if the file is so large that it is already spread on 4 disks, it also won't help seek time of course.

If you want to see proof that this works you can have a look at an (old) test on anandtech. If you at the graphs you will see that the best performance is always with a large stripe size.
http://www.anandtech.com/storage/showdoc.aspx?i=1491&p=18

A short but clear explanation here: http://www.pcguide.com/ref/hdd/perf/raid/concepts/perf_Stripe.htm
 
Graham said:
I've just got my new machine working, and I have to say, I'm blown away by the difference a decent hdd setup makes. I've set up two 72gb 10k sata raptors, and I'm quite seriously considering going to 4 of them... effectivly 40k rpm... mmm

between the windows loading logo appearing and the login screen is currently 4 seconds. :p :LOL:

I will be using another (slower) hdd for automated backup of critical stuff btw (ie, code).

I'm severly tempted... But I do wonder if it would really be that benificial? Is there anything I may be overlooking?

yay! new pc = much fun-ness


(in short, would it be worth it? temptation is an evil thing...)
It depend on a few things, like how many heads, platters, controllers, etc you can get working all at once. In a past job we used to put together massive arrays (in the terabyte plus range) for continuous processing, and when you were looking at disc i/o, the two biggest things were disc speed, and bandwith to the disc (ie the path through the controller).

What you're suggesting will probably work if you have the bandwidth through the controllers to run that many discs. For instance if you have two controllers and they can each run two discs at full speed, then you will see a benefit from going from two to four discs. If each controller can only handle one disc each at full bandwidth, then adding more than two discs will simply bottleneck at the controller. At that point you would either need to add more controllers or use controllers with adequate bandwidth.

Any speed increases are also dependent on the type of disc i/o you are performing. Some tasks are better suited to raid setups than others. For instance, in a RAID5 setup small reads can be very fast, but lots of small writes won't benefit as much because of all the extra parity data that needs to be written. So for example, your database app might benefit if it's doing lots of reading and not so much writing (ignoring the other advantages of RAID5 such as redundancy).
 
Last edited by a moderator:
mjtdevries said:
Wrong wrong wrong. Oh so very wrong.... :)

Optimizing seek time is actually the single most important reason why high-end servers use Raid configs.

When your first disk is busy reading file1, your second disk can already start reading file2 and your third and fourth disk can read file3 and file4.

You'll have to make sure that you make your stripe size as big as possible, so that there is a good chance that a small file will fit on a single disk.

In the time it takes to read a single file from 1 disk, you can now read 4 files from 4 disks. This will effectively drop seek time by 75%.

But of course this has not effect is you only need to read 1 file, so you need a situation where you have lots of simultaneous file access. (like on servers, or desktops with applications that are really heavy on the disks)
And if the file is so large that it is already spread on 4 disks, it also won't help seek time of course.

If you want to see proof that this works you can have a look at an (old) test on anandtech. If you at the graphs you will see that the best performance is always with a large stripe size.
http://www.anandtech.com/storage/showdoc.aspx?i=1491&p=18

A short but clear explanation here: http://www.pcguide.com/ref/hdd/perf/raid/concepts/perf_Stripe.htm
But in a usual desktop enviroment, it doesnt work... no gains, mostly losses. Servers are not desktops. What desktop app is heavy on a disk? No amount of raid will improve your descktop enviroment where more ram will
 
karlotta said:
But in a usual desktop enviroment, it doesnt work... no gains, mostly losses. Servers are not desktops. What desktop app is heavy on a disk? No amount of raid will improve your descktop enviroment where more ram will

Are you talking about raid in desktops in general, or more specifically about the seek time issue?

In general I would spend my money on thing like ram too, instead of on raid arrays. The only reason I have a raid5 array in my own desktop is for reliability, not for performance.

I agree on the "no gains". But I disagree about the losses. This thread wasn't about Raid5 arrays which can indeed perform a little lower than a single disk, but about Raid0. And I have never seen losses with a Raid0 array. (or Raid1, Raid0+1 for that matter)

Maybe you are thinking about NCQ, which is also only usefull in heavy disk loads, and WILL actually give losses in usual desktop environments?
 
He hasn't said how much RAM he has, so I find it odd that you all assume he needs more. It is possible he already has 2 GB, in which case adding more would give practically no benefit at all, while doubling hard drive capacity is always a welcome improvement (especially if he only has 144GB total right now, that fills up quick).

In terms of the performance benefit of striping on a workstation, you can't ignore the fact that all those gigabytes of data in games has to be loaded in memory at some point. Is disk access going to make a difference while playing a map in a game? If you have plenty of RAM, no--but it will help reduce the time spent staring at the progress bar between maps. Or, if you're playing MMORPGs... all addicts want to reduce the time between launching the game and entering the world.

I think having adequate memory and a high end video card is more important (and a good CPU of course, but that's pretty much a given these days), but as long as you cover those areas there's no reason not to run a faster disk system. It's well justified IMO.
 
karlotta said:
But in a usual desktop enviroment, it doesnt work... no gains, mostly losses. Servers are not desktops. What desktop app is heavy on a disk? No amount of raid will improve your descktop enviroment where more ram will
See the key word here?
Desktop...
mjtdevries, do you think that 4 10K drives equal a single 40K drive?
Seek time isn't reduces, it's the service time.
http://forums.storagereview.net/index.php?showtopic=16484&view=findpost&p=175824
Here's a nice long thread to wrap your head around
http://forums.storagereview.net/index.php?showtopic=15912
 
sorry about the delayed reply

for ram I currently have 4gb (although only using 2 while on xp32 - moving to s2003 eventually, will use it then)

first I'll say thank you for all your replies :) it's greatly appreciated

The thing is, I've decided to take the attitude that your only as productive as your tools let you be. And for me the obvious tools are VS 2005 and my PC. I can't really do much to VS, and in a way I don't want to, however I can make my pc as good as I can (within reason) by targetting where I upgrade it.
I effectivly want a pc that in 3-5 years (with the exception on video card) it will still be a capable machine.

Further more, I'm not too much of a gamer, although now I've moved up to my new video card I just cannot believe how much more immersive they are (even doom3 is actually good at 1600x1220 with aa :p) - but ignoring that...
the big thing is, I'm not entirly sure where development will take me over the next year, but it's looking likly I'll be starting research work on various geographic visualisation systems.. doing things like aiding search and rescue teams, etc. So needless to say, this calls for a fast machine perticuarly with good data storage, because the data sets could potentially be huge, say possibly gigabytes (especially when some of the ideas I have get involved) - so fast data storage has been a priority for me equal to a fast video card / cpu. Volume of data storage has never really been that important to me however.

currently the machine is:

athlon X2 4200+ (I didn't feel the speed increases were justified above the 4200, considering the price)
x1800xl - I had one day left in the US to buy stuff, so I got it while I could :) - it's nearly twice the price here. Would have prefered to wait for R580, but meh, I'll just upgrade later ;)
4gig ram, ram is obviously going to be imprtant for geovis, if I go that path
2x raptors of course
20" lcd (yay!)
all in a swanky lian li case :p damn it's a nice peice of work.

So yeah. So I don't really see any potential down sides to this machine,. overall I think it doesn't have any perticular weak spots. I'm just contemplating if I should go the extra mile, so to speak.
 
radeonic2 said:
See the key word here?
Desktop...

Yes I have seen that.
But it doesn't matter whether we are talking about desktops or desktops with demanding applications, or servers: your claim that Raid arrays just affect sequential read/writes and notthing else is still wrong

radeonic2 said:
mjtdevries, do you think that 4 10K drives equal a single 40K drive?

Have I given you any reason to believe that I would think that?

But in a synthetic best case scenario it will undoubtedly beat the single40k drive. Even though it will never happen in real-life scenario's.

radeonic2 said:
Seek time isn't reduces, it's the service time.
What is your definition of service time?

I have never claimed that seek time is reduced. I have claimed that "effective seektime" is reduced. I have also explained what I mean by that: The perceived seek time when your raid array opens 4 files at the same time.
I thought that I had explained it well, but if not, then please tell me which part you didn't understand and I will try to explain it better.

If you definition of service time is what I called "effective seektime" then I don't see your point. Are you now agreeing with me that Raid arrays do not only effect sequential read/writes?

And depending on the I/O usage on your desktop you will have some benefit from this.

radeonic2 said:
[/QUOTE]

While it is interesting to read about toilet paper raid arrays, I didn't find any new information in those threads. Please tell me what point you were trying to make.
 
Let me be blunt.
Desktop raid for 90%+ users makes zero sense.
There are pleanty of real world benchmarks to back that statement up.
What's to argue?
 
A general remark about RAM. The reason karlotta and I mention RAM is because in general, spending money on RAM is more effective then spending the same amount on Raid arrays.

To answer Graham's latest post.
Your machine doesn't seem to have ovious weak spots. But it all depends on the applications.

You'll have to do some monitoring on your machine to see if you can spot any bottlenecks. Especially since we don't know that geographic visualisation system. (though I can remember some HUGE photo's that we used for geologic work)

But take videoediting for instance. Lots of people claim that that needs high sequential read/writes. But when I edit DV video, on a prosumer editing board, it only needs 10MB/s maximum and something like 128MB.
Video can easily be edited in streaming mode, so that a smart editing tool will never need to load large amounts of data.

A lot depends on how geovis works. If it loads the entire dataset in memory then you high sequential speed to load the data. (assuming it's just one file) and of course lots of memory once it is loaded and at that point the disk is probably not used at all anymore.
But the programmer of such a program might have created smart ways to reduce the memory footage be loading only parts of the data and then you need less of both.
That is something that you can only determine yourself by monitoring the system while you are using the application.

The question is also if the loading of the dataset is a problem for you. Loading a single 1GB file will take something like 30 sec on a modern disk, and you can reduce that to 15 seconds with a raid0 array. But is that worth the trouble if you then spend an hour working with that file while during that time the harddisk is not used at all?


Then something not related to harddisks.
I would adivce you to look at XP x64 instead of Server2003.

Server2003 will probably not gain you much. It is also limited to the same 2GB because it is limited by the same 32bit cpu.
There is a 3GB switch in Windows Server, (which XP might also have now, but info from MS on that is contradicting) but that will only help you with programs which support that feature. (Very few programs support it) Also the /3GB switch can easily cause problems because you also have a modern videocard with lots of ram. (and you wouldn't want to ditch that one would you? ;)

Or perhaps your application can make use of PAE and AWE options? That is a method to use more than 4GB ram on 32bit cpu's. In that case it would certainly help. But again very few programs support it, so you would have to check that first too.

When you use a x64 version of windows, the OS won't have that memory limitation anymore. But it could well be that your application is still limited to 2GB, if it was developed with that limit in mind.
But at least the OS would then be more efficient in handling the large amount of memory and it is easier to make good use of the rest of it.
 
hmm odd. I thought xp64 was effectivly just a cut down version of server 2003 standard edition (64bit with 32gb mem limit).? just with all the good servery bits removed...
 
Last edited by a moderator:
Graham said:
athlon X2 4200+ (I didn't feel the speed increases were justified above the 4200, considering the price)

Buying a X2 4800+ would have been the obvious thing to do if you want to burn money on a system. A better CPU benefits all things.

However, it seems you are very low on HDD space (Raptor * 2 does not much storage space make) so you should probably pick up some larger drives to use for storage. No RAID or anything like that. Just good, fast, large drives.

Maybe a UPS system. Maybe some nice decorations or an awesome chair for the computer room to make those long hours more tolerble. Computers come and go, but your ass stays with you forever*. :p

* Hopefully. No promises are made and I cannot be held liable should it turn out otherwise.
 
Back
Top