Overclock.net › Forums › Components › Hard Drives & Storage › RAID Controllers and Software › Decent RAID5 controller?
New Posts  All Forums:Forum Nav:

Decent RAID5 controller? - Page 6

post #51 of 75
Thread Starter 
Quote:
Originally Posted by the_beast View Post
But then I expect you to still run RAID5 anyway...
Quit it with the insults. Seriously. If you want to be helpful, then be helpful. But don't make sarcastic remarks based on assumptions that simply aren't true.

Quote:
Originally Posted by the_beast View Post
Now you've actually said what you're doing with it, the advice changes. You shouldn't be running RAID5 or RAID6 - you should be on RAID10, since your small write performance will be much better for the same drive investment.

We aren't gonna infer anything from your sig rig title, so if you don't mention the details you won't get the advice you need.
What do you mean by "drive investment"? The hard drives themselves, or the controller? Wouldn't the controller be the bottleneck for RAID5 anyway? How much of a speed difference would I be looking at using the PERC 5/i controller with the F3 drives and RAID10 vs RAID5?

Sorry, I thought it was obvious what the use was for given my signature and sig rig title - my bad. I shouldn't have made that assumption.
post #52 of 75
Thread Starter 
I found this comparison for RAID5 vs RAID10 performance. Definitely a vast difference.
http://weblogs.sqlteam.com/billg/arc...rformance.aspx

Any reason to not go with RAID10 then? Why do companies/businesses generally choose RAID5 over RAID10? Is it less reliable?
post #53 of 75
Quote:
Originally Posted by SgtSpike View Post
What do you mean by "drive investment"? The hard drives themselves, or the controller? Wouldn't the controller be the bottleneck for RAID5 anyway? How much of a speed difference would I be looking at using the PERC 5/i controller with the F3 drives and RAID10 vs RAID5?

Sorry, I thought it was obvious what the use was for given my signature and sig rig title - my bad. I shouldn't have made that assumption.
I meant you still need 4 drives for both. But you also don't need the PERC at all for RAID10 providing you have the requisite 4 ports on your mobo, which give you cash for extra drives or just means your $95 better off.

With small writes on a RAID5 array you have to first read the data off the drives, recalculate the parity then write back to the array. For the RAID10 array you just write the new data to the array. Which means small writes end up taking roughly half the time on a RAID10 than a RAID5 for 4 drive arrays.

Reads and large (larger than a full stripe) writes take about the same time on each (a 3-drive RAID5 and a 4-drive RAID10) as there are the same number of 'active' drives and there is no I/O overhead involved.
post #54 of 75
The main reason that companies go with RAID 5 over 10 is due to cost. IE a RAID 5 will net you around 2/3 of your total HD space as usable. So in a RAID 5 with 3 drives you will lose the capacity of 1 full drive. So if you have 3 1TB drives you will get 2TB instead of 3 in Windows. With RAID 10 the space cost goes up to 50%. RAID 10 also requires a minimum of 4 drives and must always contain a even number of drives so if you have 4 1TB drives, you will only get 2TB of usable space in Windows (before formatting of course). The cost does not seem that bad when you are working with consumer grade SATA drives as a 1TB drive can be had for around $50 these days. But in the enterprise class a 600GB fiber channel or SAS drive will run you close to $700. Also the space does not scale fully with each drive, IE if have a RAID 5 with 3 1TB drives and I need to expand it with another 1TB drive I will not gain the full 1TB, I only gain roughly 70% of the space.
Money Sink
(16 items)
 
Greased Lightning
(16 items)
 
 
CPUMotherboardGraphicsGraphics
Core i7 960 D0 @ Stock Asus P6X58D-Premium MSI GTX465 Twin Frozr II GE UL to GTX470 PNY Refurbished GTX465 UL to GTX470 with H60 Mod 
RAMHard DriveOptical DriveCooling
12GB G.Skill Ripjaws 1600Mhz @ 9-9-9-24-2T 2 X 64GB M4 SSD in RAID0 3 X 500GB in RAID5 LG BD-R Corsair H100 
OSMonitorKeyboardPower
Window 7 64-bit, Ubuntu Linux 10, MAC OS X Dell 21" and LG 23" Razer BlackWidow Ultimate Corsair HX750 
CaseMouseAudio
CM 690 II Advanced Mad Catz Cyborg R.A.T. 7 Soundblaster recon3D Fatal1ty Champion 
CPUMotherboardGraphicsRAM
Core i7 5930K MSI X99a Gaming Godlike PowerColor Red Dragon RX 580 8GB 32 GB GSKILL 2400 DDR4 
Hard DriveHard DriveOptical DriveCooling
Samsung 256GB 950 Pro M.2 4 x Seagate 2TB SSHD in RAID 10 LG Blue Ray Burner EKWB Predator 360 AIO 
OSMonitorMonitorKeyboard
Windows 10 Pro 27" XStar 1440P IPS Korean Monitor ASUS 24" 1080P Logtech G710+ 
PowerCaseMouseAudio
Corsair AX850 Corsair 780t Logitech G700S Onboard 
  hide details  
Reply
Money Sink
(16 items)
 
Greased Lightning
(16 items)
 
 
CPUMotherboardGraphicsGraphics
Core i7 960 D0 @ Stock Asus P6X58D-Premium MSI GTX465 Twin Frozr II GE UL to GTX470 PNY Refurbished GTX465 UL to GTX470 with H60 Mod 
RAMHard DriveOptical DriveCooling
12GB G.Skill Ripjaws 1600Mhz @ 9-9-9-24-2T 2 X 64GB M4 SSD in RAID0 3 X 500GB in RAID5 LG BD-R Corsair H100 
OSMonitorKeyboardPower
Window 7 64-bit, Ubuntu Linux 10, MAC OS X Dell 21" and LG 23" Razer BlackWidow Ultimate Corsair HX750 
CaseMouseAudio
CM 690 II Advanced Mad Catz Cyborg R.A.T. 7 Soundblaster recon3D Fatal1ty Champion 
CPUMotherboardGraphicsRAM
Core i7 5930K MSI X99a Gaming Godlike PowerColor Red Dragon RX 580 8GB 32 GB GSKILL 2400 DDR4 
Hard DriveHard DriveOptical DriveCooling
Samsung 256GB 950 Pro M.2 4 x Seagate 2TB SSHD in RAID 10 LG Blue Ray Burner EKWB Predator 360 AIO 
OSMonitorMonitorKeyboard
Windows 10 Pro 27" XStar 1440P IPS Korean Monitor ASUS 24" 1080P Logtech G710+ 
PowerCaseMouseAudio
Corsair AX850 Corsair 780t Logitech G700S Onboard 
  hide details  
Reply
post #55 of 75
Quote:
Originally Posted by SgtSpike View Post
I found this comparison for RAID5 vs RAID10 performance. Definitely a vast difference.
http://weblogs.sqlteam.com/billg/arc...rformance.aspx

Any reason to not go with RAID10 then? Why do companies/businesses generally choose RAID5 over RAID10? Is it less reliable?
Companies typically use RAID5 or RAID10. All of the setups I've done for clients have been one or the other.

Typically they go with RAID5 because it is cheaper, and that's really the only reason why they use it. You don't need as many disks to run RAID5 and when read/write/access speeds aren't an issue then that's another reason.

Personally if I had about 6-10 drives for a raid array I would do RAID 0+1 for best performance and redundancy. Now there are others that may be a better solution, but most RAID controllers (with the exception of enterprise controllers) don't even support the more exotic RAID arrays.
Money Pit
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel Q6600 3.9ghz (433x9) eVGA 780i FTW PNY GTX 480 + eVGA 9800gtx+ (physx) Corsair XMS2 4x2gb 833mhz 
Hard DriveOSMonitorKeyboard
Seagate 2x750gb RAID0 | 5x1.5tb RAID5 Win 7 Ult. x64 22" Samsung 2233rz | 22" Acer 223wx Logitech G15 
PowerCaseMouseMouse Pad
Corsair HX750w + FSP GPU PSU 450w MX-Alien Razer Copperhead Thermaltake LCD Pad 
  hide details  
Reply
Money Pit
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel Q6600 3.9ghz (433x9) eVGA 780i FTW PNY GTX 480 + eVGA 9800gtx+ (physx) Corsair XMS2 4x2gb 833mhz 
Hard DriveOSMonitorKeyboard
Seagate 2x750gb RAID0 | 5x1.5tb RAID5 Win 7 Ult. x64 22" Samsung 2233rz | 22" Acer 223wx Logitech G15 
PowerCaseMouseMouse Pad
Corsair HX750w + FSP GPU PSU 450w MX-Alien Razer Copperhead Thermaltake LCD Pad 
  hide details  
Reply
post #56 of 75
Thread Starter 
Quote:
Originally Posted by the_beast View Post
I meant you still need 4 drives for both. But you also don't need the PERC at all for RAID10 providing you have the requisite 4 ports on your mobo, which give you cash for extra drives or just means your $95 better off.

With small writes on a RAID5 array you have to first read the data off the drives, recalculate the parity then write back to the array. For the RAID10 array you just write the new data to the array. Which means small writes end up taking roughly half the time on a RAID10 than a RAID5 for 4 drive arrays.

Reads and large (larger than a full stripe) writes take about the same time on each (a 3-drive RAID5 and a 4-drive RAID10) as there are the same number of 'active' drives and there is no I/O overhead involved.
Makes sense. I don't have the extra ports on the motherboard though, so I do need a controller regardless. I have a 2-drive array on a cheapo controller, and 4 SSD's plus a VR on the motherboard ports, so I only have one open SATA port at the moment.

So RAID5:
- More convenient to replace dead drive (can use one as hotspare, would take three drive failures to kill array)
- Slower small write speeds
- Takes longer to rebuild array after drive failure

RAID10:
- Faster small write speeds
- Could kill array with two drive failures
- Rebuild array more quickly after a drive failure

At this point, I'll probably go with RAID10 on the PERC 5/i. The PERC 5/i because I know it is a reliable controller, and a decent 4-port controller on newegg is almost as much anyway. I'll run performance tests in both modes (RAID10 and RAID5) just to make sure it matches with my expectations.
post #57 of 75
Thread Starter 
Quote:
Originally Posted by Warsteiner View Post
The main reason that companies go with RAID 5 over 10 is due to cost. IE a RAID 5 will net you around 2/3 of your total HD space as usable. So in a RAID 5 with 3 drives you will lose the capacity of 1 full drive. So if you have 3 1TB drives you will get 2TB instead of 3 in Windows. With RAID 10 the space cost goes up to 50%. RAID 10 also requires a minimum of 4 drives and must always contain a even number of drives so if you have 4 1TB drives, you will only get 2TB of usable space in Windows (before formatting of course). The cost does not seem that bad when you are working with consumer grade SATA drives as a 1TB drive can be had for around $50 these days. But in the enterprise class a 600GB fiber channel or SAS drive will run you close to $700. Also the space does not scale fully with each drive, IE if have a RAID 5 with 3 1TB drives and I need to expand it with another 1TB drive I will not gain the full 1TB, I only gain roughly 70% of the space.
Quote:
Originally Posted by ChieF-TroN View Post
Companies typically use RAID5 or RAID10. All of the setups I've done for clients have been one or the other.

Typically they go with RAID5 because it is cheaper, and that's really the only reason why they use it. You don't need as many disks to run RAID5 and when read/write/access speeds aren't an issue then that's another reason.

Personally if I had about 6-10 drives for a raid array I would do RAID 0+1 for best performance and redundancy. Now there are others that may be a better solution, but most RAID controllers (with the exception of enterprise controllers) don't even support the more exotic RAID arrays.
Makes sense. If I go RAID5, one of them will be a hotspare anyway, so it'll be the same number of drives for the same 2TB of usable space anyway.
post #58 of 75
People don't go with RAID5 because it's cheaper, they go with RAID5 because it's faster. In certain circumstances. Or at least those people who know what they're doing choose either RAID5 or RAID10 depending on the situation. Cost is a non-issue - you don't drop $20k on a server then worry about another few hundred on a drive if it will boost your performance by 50%, which is easily possible with higher spindle counts for read-heavy uses...

It's worth noting (regarding your pros/cons) that a RAID5 array can take hours or even days to rebuild (especially if the array remains in service during the failure) - so relying on a hotspare is not ideal, especially as a rebuild triggers heavy workload on the remaining drives, so a second failure is more likely to happen during a rebuild as the drives take a beating.

Remember when you run your tests you need to replicate the exact usage you want to use the array for - which is all but impossible. If you set up the test wrong (like doing sequential read tests for an array which is used for small writes) then your benchmarks could throw you off in completely the wrong direction - kinda like when people point to a 4-drive RAID0 array of mechanical drives being faster than a C300 for sequential reads, so therefore it must be a better OS drive...
post #59 of 75
Quote:
Originally Posted by the_beast View Post
People don't go with RAID5 because it's cheaper, they go with RAID5 because it's faster. In certain circumstances. Or at least those people who know what they're doing choose either RAID5 or RAID10 depending on the situation. Cost is a non-issue - you don't drop $20k on a server then worry about another few hundred on a drive if it will boost your performance by 50%, which is easily possible with higher spindle counts for read-heavy uses...

It's worth noting (regarding your pros/cons) that a RAID5 array can take hours or even days to rebuild (especially if the array remains in service during the failure) - so relying on a hotspare is not ideal, especially as a rebuild triggers heavy workload on the remaining drives, so a second failure is more likely to happen during a rebuild as the drives take a beating.

Remember when you run your tests you need to replicate the exact usage you want to use the array for - which is all but impossible. If you set up the test wrong (like doing sequential read tests for an array which is used for small writes) then your benchmarks could throw you off in completely the wrong direction - kinda like when people point to a 4-drive RAID0 array of mechanical drives being faster than a C300 for sequential reads, so therefore it must be a better OS drive...
Apparently you don't do business with small businesses... Cost is majority of the time an issue.

Not every company has endless funds to throw around. I have clients that can only afford a $2-3k server but at the same time have clients that can afford $200k clustered servers with fibre SAN's etc.

When you're dealing with small clients you have to make cuts some where and depending on their needs it's sometimes worth going with RAID5 with a faster cpu and more ram over slower cpu/ram with RAID10.
Money Pit
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel Q6600 3.9ghz (433x9) eVGA 780i FTW PNY GTX 480 + eVGA 9800gtx+ (physx) Corsair XMS2 4x2gb 833mhz 
Hard DriveOSMonitorKeyboard
Seagate 2x750gb RAID0 | 5x1.5tb RAID5 Win 7 Ult. x64 22" Samsung 2233rz | 22" Acer 223wx Logitech G15 
PowerCaseMouseMouse Pad
Corsair HX750w + FSP GPU PSU 450w MX-Alien Razer Copperhead Thermaltake LCD Pad 
  hide details  
Reply
Money Pit
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel Q6600 3.9ghz (433x9) eVGA 780i FTW PNY GTX 480 + eVGA 9800gtx+ (physx) Corsair XMS2 4x2gb 833mhz 
Hard DriveOSMonitorKeyboard
Seagate 2x750gb RAID0 | 5x1.5tb RAID5 Win 7 Ult. x64 22" Samsung 2233rz | 22" Acer 223wx Logitech G15 
PowerCaseMouseMouse Pad
Corsair HX750w + FSP GPU PSU 450w MX-Alien Razer Copperhead Thermaltake LCD Pad 
  hide details  
Reply
post #60 of 75
Quote:
Originally Posted by ChieF-TroN View Post
Apparently you don't do business with small businesses... Cost is majority of the time an issue.

Not every company has endless funds to throw around. I have clients that can only afford a $2-3k server but at the same time have clients that can afford $200k clustered servers with fibre SAN's etc.

When you're dealing with small clients you have to make cuts some where and depending on their needs it's sometimes worth going with RAID5 with a faster cpu and more ram over slower cpu/ram with RAID10.
Not if their HDDs will be bottlenecking them it isn't - a better disk array will help more than extra RAM or faster CPUs if what you're doing is I/O bound.

The point is that you have to gear the hardware precisely to the type of usage it will see. You can't just say one setup is always faster than another (or cheaper for a given performance level), because it just isn't true.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: RAID Controllers and Software
Overclock.net › Forums › Components › Hard Drives & Storage › RAID Controllers and Software › Decent RAID5 controller?