SSD boot drive: 2x60 in RAID 0 or single 120?

[johnnymg]

Greets

Please advise what you would do for a new MP boot drive setup.

I'll be going with the OWC SSD(s) ExtremePro for the boot drive................ either a 2x60 OSX SW RAID 0 or single 120.

I'd rather err on the side of safety but always looking for more umph~~~ 😛 If going RAID 0 would it be advised to go with a dedicated controller?

My data drive will be 2x1T RAID 0.

Thanks
JohnG

 

[Honumaui]

raid 0 on boot is kinda waste ? some will say its way faster ? but in reality its a mental thing 🙂 numbers have shown you really gain very little by doing so ?

I would go with a single SSD myself ? for boot that is

 

[WardC]

A dual hard drive RAID 0 would be faster, but I use a Single 120GB SSD for my boot drive on my Mac Pro, and it's plenty fast. I use my dual 1TB HD RAID 2TB Array for my media files, iTunes, movies, and Downloads folder.

 

[nanofrog]

Assuming you won't have any issues with the number of drives in the system, another thing to consider is cost.

If it's cheaper to use 2x units to get sufficient capacity vs. a single unit, then go for it. Unfortunately, I just checked OWC's pricing, and that's not the case right now. 🙁

2x 60GB units = $320
1x 120GB unit = $300

So for the same capacity, you pay an extra $20 (not including any shipping, or mounting solution). The positive side, is it would be faster. Now whether the OP would actually be able to benefit, is another issue.

Personally, I doubt it, as you're waiting mainly on the system's firmware during boot with SSD's (just take a look at posted boot times from SSD users in MR). Random access does improve slightly with a stripe set, it's not as much as most might thing (theoretically, a 2x set cuts the random access time in half, but is rarely ever seen).

Mechanical disks that are rated for say 12ms (single disk), will drop to 10ms or so for the set on average. Not that great an improvement IMO, but typically shows up in benchmarks. And SSD's are insanely fast compared to mechanical anyway, so a single should be impressive enough. 😱 😛

 

[WardC]

My question, is it possible to install two SSDs in one of your optical drive bays, and 4 SSDs (one in each of the main bays) and have a six-SSD RAID 0 array?? That would be hyper-fast. Those PCIe cards are 8-channel and support over a gigabyte a second, you would be approaching that with a six-SSD array.

 

[johnnymg]

OK, I'm leaning towards a single SSD solution for the boot drive. The thought of RAID 0 and boot drives just gives me the hebe-gebies. 😛

I've had a very good experience with single SSD's as boot/app drives in my MBP's so I'm inclined (baring someone convincing me otherwise) to stick with that approach.

FWIW, OWC appears to have lowered their Extreme Pro prices today. 😎

thanks for the suggestions
JohnG

 

[nanofrog]

My question, is it possible to install two SSDs in one of your optical drive bays, and 4 SSDs (one in each of the main bays) and have a six-SSD RAID 0 array?? That would be hyper-fast. Those PCIe cards are 8-channel and support over a gigabyte a second, you would be approaching that with a six-SSD array.

A couple of problems with the MP (or any other system that only contains the ICH for SATA ports).

1. Physically, the ICH contains 6x SATA ports. They're distributed as:

4x = HDD bays (one per bay)
2x = ODD bays (one per bay)

So you'd have to remove the optical drive and use that bay as well in order to connect it to a port.​

You can install more than one in an optica bay, but to get it connected to a port, you'd need to use an extension cable (still have to disconnect the optical drive), or use a separate controller.

2. Most importantly, the ICH is only good for ~660MB/s. Since most of the recent SSD's users seem to be interested in can hit 250MB/s or so sustained, you can throttle on just 3x SSD's on the ICH.

In order to get around this problem, you need more bandwidth. Which means you'll have to use a separate controller of some sort on the PCIe lanes to obtain it (8x lane Gen 2.0 card = 500MB/s per lane, so 4GB/s max for the newer cards, which are 6.0Gb/s compliant; older cards are Gen 1.0 & 3.0Gb/s, so halve it <2GB/s for an 8x lane model>).

 

[Honumaui]

these are the 40 gig SSD from OWC
mainly just to show how they scale on the ICH and what 3 should be able to do 🙂



3 on the Areca raid

3 discs on the ICH

two on the ICH

single on ICH

 

[EdbBob]

Raided boot drive

Hi all geeks 😉

I have an 8-core MP 2008 with 4 x 1 TB Raid0 as my boot drive. I was a bit sloppy, choosing disks, but with a current 75% disk usage, it'll still do 300 MB/s, and is way faster than working on a single drive. I can feel a difference in booting and in disk heavy applications. I work with very heavy movie files (4K Red raw), 3D apps loading huge libraries from disk, etc. I use a four drive raid0 as Time machine disk, connected via a Sonnet 4 port eSata controller (very good). I'm able to connect the Time Machine raid via FW800 if I need restoring. Btw. that's important! 😉 That way I have better safety than raid01 and even raid5, because I have the history from Time Machine. Works like a charm. I've had to replace one drive in the past year, and my Mac is running 24/7...

Some of you might have read my experiments with raiding USB keys (poor mans ssd), and I'm a raid/speed fanatic. I've been thinking about mounting 4 x OCZ Vertex 2's as my boot drive (probably mounting them internally with gaffa tape 😉 ). In theory that should give something like 1 GB/s, with close to zero access time. Sound like sweet music to my ears, and then I read Nanofrog's point about something called ICH. I tried Googling it, but can't seem to find it.

I was a PC guy until 3 years ago, so what's that ICH thingy doing?

And does it bottleneck at 660 MB/s?

Thanks in advance

EDBBOB

 

[EdbBob]

@ Nanofrog:

Ah! You mean the South Bridge 😀

Does that max out at 660 MB/s?!?

 

[lbeck]

A dual hard drive RAID 0 would be faster, but I use a Single 120GB SSD for my boot drive on my Mac Pro, and it's plenty fast. I use my dual 1TB HD RAID 2TB Array for my media files, iTunes, movies, and Downloads folder.

Did you specifically move your main users folder from the boot drive to your 1TB drive? If so, how did you do it? I've heard conflicting methods for moving your main user folder.

 

[Honumaui]

Hi all geeks 😉

I have an 8-core MP 2008 with 4 x 1 TB Raid0 as my boot drive. I was a bit sloppy, choosing disks, but with a current 75% disk usage, it'll still do 300 MB/s, and is way faster than working on a single drive. I can feel a difference in booting and in disk heavy applications. I work with very heavy movie files (4K Red raw), 3D apps loading huge libraries from disk, etc. I use a four drive raid0 as Time machine disk, connected via a Sonnet 4 port eSata controller (very good). I'm able to connect the Time Machine raid via FW800 if I need restoring. Btw. that's important! 😉 That way I have better safety than raid01 and even raid5, because I have the history from Time Machine. Works like a charm. I've had to replace one drive in the past year, and my Mac is running 24/7...

Some of you might have read my experiments with raiding USB keys (poor mans ssd), and I'm a raid/speed fanatic. I've been thinking about mounting 4 x OCZ Vertex 2's as my boot drive (probably mounting them internally with gaffa tape 😉 ). In theory that should give something like 1 GB/s, with close to zero access time. Sound like sweet music to my ears, and then I read Nanofrog's point about something called ICH. I tried Googling it, but can't seem to find it.

I was a PC guy until 3 years ago, so what's that ICH thingy doing?

And does it bottleneck at 660 MB/s?

Thanks in advance

EDBBOB

go up a post and those benchmarks I posted using SSD
3 drives are not that much quicker ? notice the Areca though they scaled nicely 🙂
ICH is your I/O Controller Hub 🙂

also the bold thing 🙂 raid 01 is bad !!!! 10 is good 🙂 in relation to each other 🙂

to say your stuff is safer than raid 5 ? is wrong 🙂 the reason is 1 disc fails on your 0 you go down !
if for freak reasons you loose a disc on each set you are really hosed !

in my raid 6 i can loose two discs and keep working my time machine is also a raid 5 ? again much safer

if yours works for you ? thats cool but dont kid yourself raid 0 for working files is not safe at all ? your Time machine on raid 0 on top of that ?

not trying to be jerky just point out some things 🙂


curious if your files are on your boot that you work off ? it sounds like it ? you might be better off having files on a dedicated setup 🙂
you would be even better with one SSD boot and your 4 disc on its own ?

check into a good raid controller maybe 16 discs 🙂 ? you can still run your raid 0 if you want ? and get way better speed or better yet do 1+0 not 0+1 !!!
or even better raid 6 🙂

 

[EdbBob]

@Honumaui

Thanks for your comments.

You are right to a certain degree. If a system disk fails, I have to restore from Time Machine. If a Time Machine disk fails I replace it and make a new backup, thereby loosing history. I think people are to scared of raid0. You have to backup, but performance is great. And it seems pretty stable to me.

However, raid5 is also slow (but usable) when you replace a disk. I think the main problem for me, was that raid5 was expensive hardware, and therefore I ended up with a lot of external disks software raided.

Raid01/raid10? I think I meant raid10. I've been able to make a mirrored raid1 out of two raid0's using DiskUtil, but I've not been able to do it the other way around. (Striping mirrors)

I think we agree on using SSDs as a system drive, but after reading about the ICH, I think I can do with 3 OCZ's in a raid0. That should max out the ICH, and, hey, 660 MB/s isn't bad 😉 One thing I noticed when raiding USB keys was that SSDs fail mostly at writes, whereas HDD's fail at reads. That doesn't remove the need for backup, but I like it in real world situations.

Once I get that up and running, I have a truckload of sata and eSata ports available (I have two Sonnets, and the two ports on the mobo), and then I'll try and build an 8 drive raid0 for all my work.

 

[nanofrog]

I have an 8-core MP 2008 with 4 x 1 TB Raid0 as my boot drive. I was a bit sloppy, choosing disks, but with a current 75% disk usage, it'll still do 300 MB/s, and is way faster than working on a single drive. I can feel a difference in booting and in disk heavy applications. I work with very heavy movie files (4K Red raw), 3D apps loading huge libraries from disk, etc. I use a four drive raid0 as Time machine disk, connected via a Sonnet 4 port eSata controller (very good).

Apparently you've been fine with a stripe set for a backup source, but I don't trust it for that and don't recommend it (i.e. primary array dies, gets fixed, and the backup array dies during data restoration 😱).

Not pretty. 🙁

I'm able to connect the Time Machine raid via FW800 if I need restoring. Btw. that's important! 😉 That way I have better safety than raid01 and even raid5, because I have the history from Time Machine. Works like a charm. I've had to replace one drive in the past year, and my Mac is running 24/7...

You may have lost me here...

Is the box you're using an eSATA + FW800 unit (would make sense to me)? Or do you have more than one backup of the same data?

As per the FW800 restoration, if it's a stripe set, then it's not safer than 10 or 5 (not even equivalent to RAID 5, which is only has a redundancy = 1 member can be lost without losing data). A 10 array (safer than 0+1) has a redundancy = 2.

...I read Nanofrog's point about something called ICH. I tried Googling it, but can't seem to find it.

It's only been with the increasing use of SSD's that it's even gained the attention of the public, as Intel never published this information. Even now, it's not all that commonly known from what I see/hear.

Many are surprised to discover this bit of information. 🙁 What you need to keep in mind however, is when the ICH was originally designed, there were no SSD's, so it was designed for mechanical disks. Once SSD's came on the scene, they couldn't fix it, as the design uses DMI to connect it to the chipset (1GB/s up and down), and had they increased it's alloted bandwidth, it could choke the other data from reaching the system. So they were stuck with it.

The next revision of the ICH should contain 6.0Gb/s SATA ports, so this issue will be addressed (it will have to connect via a faster bus than DMI, so it will be a substantial re-design from what's currently used, though still evolutionary).

And does it bottleneck at 660 MB/s?

ICH = I/O Controller Hub, which contains the SATA, USB, and Ethernet controllers.

The Southbridge term was done away with when Intel introduced the Nehalem architecture, but the actual ICH used was released earlier in Q3 2008 (ICH 10; ICH10R = Q4 2008, and isn't used in the MP's).

Unfortunately, it's the source of the throughput cap, and it exists in all Intel MP's (this has been around awhile).

 

[Honumaui]

@Honumaui

Thanks for your comments.

You are right to a certain degree. If a system disk fails, I have to restore from Time Machine. If a Time Machine disk fails I replace it and make a new backup, thereby loosing history. I think people are to scared of raid0. You have to backup, but performance is great. And it seems pretty stable to me.

I have nothing against raid 0 🙂 I have used it in many cases the one thing to know is if a drive dies you are hosed ? but as I say you loose a single drive and you are hosed 🙂 so why I prefer raid 10 or raid 6 or even 1 for boot is good 😉

However, raid5 is also slow (but usable) when you replace a disk. I think the main problem for me, was that raid5 was expensive hardware, and therefore I ended up with a lot of external disks software raided.

raid 6 is the way to go areca controllers are not slow in raid 6 I can guarantee you that 🙂
who cares if its slow on rebuild ? its still quicker than trying to start from scratch 🙂

and yes agree its expensive to do properly ? but in the long run its not so bad to be safe and fast

Raid01/raid10? I think I meant raid10. I've been able to make a mirrored raid1 out of two raid0's using DiskUtil, but I've not been able to do it the other way around. (Striping mirrors)

raid 1- would be making two raid 1 sets(mirrored) and then putting them together with a Raid 0(striped)

with the two raid 0s as base ? that is 01 the downside is if your base is two raid 0 you can not loose a HDD on each side ? only one ! with 10 you can loose one on both sides and still be OK 🙂

I think we agree on using SSDs as a system drive, but after reading about the ICH, I think I can do with 3 OCZ's in a raid0. That should max out the ICH, and, hey, 660 MB/s isn't bad 😉 One thing I noticed when raiding USB keys was that SSDs fail mostly at writes, whereas HDD's fail at reads. That doesn't remove the need for backup, but I like it in real world situations.

look at my Images above 🙂 or are they not working ???? 😱
I posted some aja screen grabs ? I hope people can see em 🙂 hehehehe
unless the attachment thing did not work for me ?

you can not get 3 SSD on the ICH it throttles big time notice the areca I put 3 SSD in 0 to show what it should do !
I have 5 SSD all OWC ones to play with 🙂 and they are really the same things Sandforce based as the OCZ
so yes you can do it but why waste the 3rd if you keep them on the ICH ?


again when you look at my aja output you only gain a touch with that 3rd so it would be a total waste ? the only way to get around that would be to run 2 off ICH and then run the other off a quality card in the PCI slot ? like your sonnet ?

Once I get that up and running, I have a truckload of sata and eSata ports available (I have two Sonnets, and the two ports on the mobo), and then I'll try and build an 8 drive raid0 for all my work.

use aja and post some results 🙂

 

[EdbBob]

@nanofrog

(Cool name btw). I agree, that raid0 is risky, but then you should have seen my face when I lost all data some years ago, after making a jbod, with no backup. Now that's failure waiting to happen.

One disk failing, is not something you experience everyday. Two disks failing at the same time is very rare, and would mean total disaster for me. I remember many years ago, when Micropolis made a batch of disks, that had fine metal particles in the oil for the main bearing. I had 8 of those in raid0 on an old SGI machine. Doh! 🙁 Compared to that I feel pretty safe with my setup nowadays.

Loosing all data, due to the jbod, was frightening, saddening and also refreshing... 😉 It took me a few days, then I was up and running again, and in a week I forgot about the old holiday images, I never watched anyway. It's a really bad experience, but to me it wasn't the end of the world.

My Time Machine consists of two WD Mybook 4TB in raid0 connected to another Sonnet (yes, I have two Sonnets). Restoring the system disk is slow using FW800, but OSX only recognize the TM-raid connected via eSata using a driver. So in everyday use, I backup at around 200 MB/s, and if I need restoring, I drink a swimming pool of coffee, while I wait... I've only tried that once, but it worked perfectly, even though all the coffee is hard on the stomach 😉

I've been considering selling my MP2008 to get the new 12 core version with hyperthreading, since that would run almost 3x faster in some applications, but the ICH9 issue, makes me wanna stick with my old trusty MP, a little longer. Thanks for the info! 🙂

@Honumaui:

Your images is not working on my machine (Firefox, Opera or Safari). Could you try posting them again.

From what I learned fiddling around with my USB key raids, it seems that OSX is not disk heavy, once it's running. It's loading apps, and apps loading stuff that's time consuming. Running OSX of a single fast USB key is very slow booting, but once running it's usable, and very interesting to watch. Therefore I'll put my SSDs straight onto the ICH. As I mentioned above, booting from anything else seems impossible, due to the Sonnets using a driver. The OCZs are rated at close to 300 MB/s, so you might have a point about using only two SSDs, but on the other hand, 3 SSDs should give me some overhead, and assure that the ICH is maxed out. And it'll give me a little more drive space, and that's gonna be an issue for me.

I have some WD's and some Samsung F3s. The latter are far better performance wise, so I'll stick with those for the eSata raid. And I'll post some Aja's 😉 In the long run I'll try and go for raid6, but that's scifi for me right now.

OT: Now that I'm surrounded by like minded geeks, I have a question, that's been puzzling me for a while. Many years ago, Tom from TomsHardware, made a test raiding two and three disks in raid0. The big performance jump came from adding the second disk, but what he saw was, that adding the third disk didn't triple one disks performance. However the phenomena of disks performing slower when filled with data disappeared. Like the performance of a three disk HDD raid0 was like a fresh empty two disk raid, no matter how full it was. Can anyone confirm/explain this?

Thanx again for your info 🙂

 

[mjsmke]

May be due to a limit in how much you bus can handle. If the speed of 2 HDD's in RAID 0 is near the optimum the bus can handle, then a faster drive (3 HDD's in RAID 0) wont make much difference.

However, as any drive fills up the speed drops a little. so if the 3 HDD's in RAID 0 can operate much faster than your machine can take advantage of, when they fill and slow down, the speed may still be above the optimum of 2 HDD's in RAID 0.

Im only guessing but hope that makes sense.

 

[EdbBob]

May be due to a limit in how much you bus can handle. If the speed of 2 HDD's in RAID 0 is near the optimum the bus can handle, then a faster drive (3 HDD's in RAID 0) wont make much difference.

However, as any drive fills up the speed drops a little. so if the 3 HDD's in RAID 0 can operate much faster than your machine can take advantage of, when they fill and slow down, the speed may still be above the optimum of 2 HDD's in RAID 0.

Im only guessing but hope that makes sense.

It makes sence, since this was back when you could email Tom, and he would answer, read: Many years ago 😉 Back then the PCI bus seemed like the bottleneck when raiding.

 

[Honumaui]

EdbBob
OK for some reasons my attachments did not work the first round ???

anyway redid them this morning so here it is again ? double post with images 🙂 but figure its good to see incase its on a previous page 🙂

these are the 40 gig SSD from OWC
mainly just to show how they scale on the ICH and what 3 should be able to do 🙂



3 on the Areca raid

3 discs on the ICH

two on the ICH

single on ICH

 

[nanofrog]

@nanofrog

(Cool name btw). I agree, that raid0 is risky, but then you should have seen my face when I lost all data some years ago, after making a jbod, with no backup. Now that's failure waiting to happen.

Running without a backup is begging to be burnt, as it's a matter of when, not if (applies to both RAID and single disk implementations).

One disk failing, is not something you experience everyday. Two disks failing at the same time is very rare, and would mean total disaster for me. I remember many years ago, when Micropolis made a batch of disks, that had fine metal particles in the oil for the main bearing. I had 8 of those in raid0 on an old SGI machine. Doh! 🙁 Compared to that I feel pretty safe with my setup nowadays.

It does happen, and possibly more than you realize (can be better seen in large implementations, such as what's found in data centers).

One way to mitigate it, is to get the drives from multiple sources to avoid sequential drives (and hopefully batches, though the latter is less certain). But this will cost more.

In data centers, they can't do this to the extent a smaller user can, given the number of drives used and suppliers available (i.e. they tend to buy in cases). So when there's a bad batch, there's usually multiple disks from it in the configuration/s (i.e. multiple arrays, and they may each have one or more drives from the bad batch in them). So it's really critical here that they mitigate sequential S/N's and batch #'s to avoid issues as best as possible.

Loosing all data, due to the jbod, was frightening, saddening and also refreshing... 😉 It took me a few days, then I was up and running again, and in a week I forgot about the old holiday images, I never watched anyway. It's a really bad experience, but to me it wasn't the end of the world.

People tend to remember when they learn the hard way. 😱 😛 Glad it wasn't catastrophic for you when it happened.

My Time Machine consists of two WD Mybook 4TB in raid0 connected to another Sonnet (yes, I have two Sonnets). Restoring the system disk is slow using FW800, but OSX only recognize the TM-raid connected via eSata using a driver. So in everyday use, I backup at around 200 MB/s, and if I need restoring, I drink a swimming pool of coffee, while I wait... I've only tried that once, but it worked perfectly, even though all the coffee is hard on the stomach 😉

You can use other levels to reduce the time input required to recover from a failure. But it costs money to obtain it (i.e. the ability to run parity based arrays via a proper hardware RAID controller).

Even a backup solution. The Qx2 from OWC for example, is a hardware controller in a box (simple RoC = RAID on a Chip, but it works). You can run the thing to the system via eSATA, get it started, and walk away. If the backup system has a failure, replace the disk and it automatically rebuilds.

Same goes for RAID 5 on the primary array (you will need a RAID card), and RAID 6 (what Honumaui uses), allows 2 disks to fail without data loss.

Much less time involved.

I've been considering selling my MP2008 to get the new 12 core version with hyperthreading, since that would run almost 3x faster in some applications, but the ICH9 issue, makes me wanna stick with my old trusty MP, a little longer. Thanks for the info! 🙂

You'd benefit from a proper RAID card from what you've posted, and it can offer options that will reduce your time input due to a failure. And you can transfer the thing from one system to another.

Fast computers don't fare as well as you'd hope if there's an unaddressed bottleneck, and storage is a big one (single HDD solution is chosen to save costs). System vendors expect the user to upgrade it if needed to suit their needs.

 

[EdbBob]

A minor update...

@Honumaui: I decided to look more carefully at my setup, and found that I was running raid01. As you said raid10 is preferable, so I did some research and found a way to do that using diskutil, and WD RaidManager. First I took the two 4TB WD I have and used WD RaidManager to make them into two 2TB mirrors each. Then I took two 2TB Lacies and mirrored those with diskutil. Finally I created a stripe of all three mirrors. That's the tricky part. You have to go to the raid tab, click "+", setup a stripe, add the WDs and then comes the important part: You have to drag the mirrored Lacies from the "Raid listing", not from the sidebar to the left. If you try to add a mirror from the sidebar diskutil just ignores it. Thanks for the info! 🙂

Just for fun, while I wait for the SSD's I've tried to raid 5 Samsung 1TB F3's using ICH. I reach about the same speed as Honumaui does with three SSD's on ICH. I'll try and post some AJA's when I finish restoring the system.

Starting from scratch I would have gone for raid5. However, I started out with much less disk space than today, and build and build on top of that. I think others do something similar, so I just wanted to share what I found out about using diskutil.

More later...

EDBBOB 🙂

 

[EdbBob]

OCZ Vertex 2

These are hard to find, where I live, but today I received 2x 120 GB. They rock! In raid0 they deliver 440/460 MB/s and the lack of access time is amazing. Applications load within a second. Truly one of the best performance increases I've experienced, since HDD replacing Floppy! I've no idea, what problems lurks in the shadows, but so far, this is very cool 🙂

 

[Honumaui]

cool 🙂 whats funny is how light and small and silent they are 🙂

cant wait for them to get larger and cheaper !! 🙂