Yup, you need hot swapping for that. No doubt. I've been meaning to get me a hot swappable solution for the Mac too like on some of my other systems. I've just never seen anything I thought looked good, was good, and that was reasonably priced.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
RAID Throughput Question - Card vs. Internal
- Thread starter jethrodesign
- Start date
- Sort by reaction score
Yup, you need hot swapping for that. No doubt. I've been meaning to get me a hot swappable solution for the Mac too like on some of my other systems. I've just never seen anything I thought looked good, was good, and that was reasonably priced.
What makes it worse, is that there are hot swappable capable cards for the MP, but the PSU does not support hot plugging. So you still can't just yank out dead drives and replace them without shutting down.
Unless you like risking the system over trying it.Hmmm, so on single external drives with their own internal power supply (like a typical LaCie Quad-interface), you still need a hot-swappable card to be able to unmount & shutdown without turning the computer off?
Are there inexpensive external SATA cards (not necessarily RAID) that do this?
What about an upgrade to our 1820a that might allow this? I've seen some of the Highpoint upgrade cards that have 4 internal and 4 external ports.
I guess for some reason I just assumed since these drives came with USB/Firewire/eSATA, and no big disclaimer that they work completely differently with eSATA, that you could use them in a similar manner. Guess I was wrong?!? I suppose we can at least use FW800, which should be better than our current FW400.
Are there inexpensive external SATA cards (not necessarily RAID) that do this?
What about an upgrade to our 1820a that might allow this? I've seen some of the Highpoint upgrade cards that have 4 internal and 4 external ports.
I guess for some reason I just assumed since these drives came with USB/Firewire/eSATA, and no big disclaimer that they work completely differently with eSATA, that you could use them in a similar manner. Guess I was wrong?!? I suppose we can at least use FW800, which should be better than our current FW400.
I dunno for sure about the rest of your questions but for this one: If it's SATA (or eSATA) then the controller needs to be hot swap capable in order to physically remove the HDD from the bus. If it's not then I dunno exactly what will happen. Maybe nothing. Maybe if you eject it in the OS first the only problem will be that the mount information remains in RAM. It could also be that subsequent drives added (while still hot) will try to use that mount information incorrectly. I dunno, but anyway, you're supposed to have a hot swappable controller in order to hot swap. If you have a housing that has it's own power supply that would be the other variable in the equation. Certainly ejecting the drive and then powering it off before detaching it from the bus would eliminate most or all of the potential for damage or corruption.
Yes.
Yes, they do exist. You can take a look at Sonnet Technologies or NitroAV. I stumbled across this, and there's some information that may help. But it will take some research to make sure the cards have all the features you want.
If you mean a way to get the 1820A to do it, you might be able to find another companies drivers that support it, but it requires the hardware be capable of as well. I've no idea of the controller chip used on that card. If it's a Silicon Image part, then it may be possible.
Each interface is different, and though may have a provision for hot swapping possible in the specification, it may be optional, as is the case with eSATA. So you have to check the details everytime you buy something.
Well, thanks again for all your help on this! Didn't think it would get so involved.
I noticed there is an upgrade to our RAID card called the RocketRaid 2224 that has 4 internal SATA & 4 external eSATA ports. It does support 'hot-swap' in a sense (you have to apparently go into their software each time to disconnect or reconnect the drive). It says this is to prevent the RAID from thinking it is broken and trying to rebuild. But the eSATA ports would not be used for the RAID, just individual drives.
We can look into this, as it is a bit newer and more supported by Highpoint (ours is discontinued). But I was just wondering if we were going to upgrade our card if it might be better to get more than just the ability to hot-swap. Everything else looks very similar to our card (still 'fake' RAID).
We might be open to spending a small amount on this, but as it will have to be PCI/PCI-X to work in our G5, we don't want to invest too much as it wouldn't be transferable to a new computer in the future (which would most likely be PCI-e). So it's not worth spending a ton on an amazing RAID card that would have a fairly limited lifespan.
- Any suggestions for something not too expensive that would get us the hot-swapable feature and possibly a processor and/or cache? Needs to be PCI/PCI-X.
THANKS AGAIN!!!
I noticed there is an upgrade to our RAID card called the RocketRaid 2224 that has 4 internal SATA & 4 external eSATA ports. It does support 'hot-swap' in a sense (you have to apparently go into their software each time to disconnect or reconnect the drive). It says this is to prevent the RAID from thinking it is broken and trying to rebuild. But the eSATA ports would not be used for the RAID, just individual drives.
We can look into this, as it is a bit newer and more supported by Highpoint (ours is discontinued). But I was just wondering if we were going to upgrade our card if it might be better to get more than just the ability to hot-swap. Everything else looks very similar to our card (still 'fake' RAID).
We might be open to spending a small amount on this, but as it will have to be PCI/PCI-X to work in our G5, we don't want to invest too much as it wouldn't be transferable to a new computer in the future (which would most likely be PCI-e). So it's not worth spending a ton on an amazing RAID card that would have a fairly limited lifespan.
- Any suggestions for something not too expensive that would get us the hot-swapable feature and possibly a processor and/or cache? Needs to be PCI/PCI-X.
THANKS AGAIN!!!
Well I did some quick research on an upgrade to our RAID card. There's not many options that are still PCI-X, most decent cards are PCI-e.
But I did come across an Areca ARC-1120. It is an 8 port internal card like ours, BUT is a true hardware card (has Intel processor and 256MB cache). It supports RAID 0,1,5,6,10, etc. It appears to support hot-swap, but their specs are a bit vague. So I could hopefully convert a couple of those extra ports to external (we got our adapter today).
- Anyone familiar with this card (or at least family or manufacturer)?
- This being true hardware RAID, would we be any safer running a RAID 5 with 3 drives on this (less chance of data loss on repair, and higher throughput on write speed)?
- Any other brands that might have something (I checked Highpoint, sonnet, firmtek, caldigit) and work well with Mac?
THANKS!
But I did come across an Areca ARC-1120. It is an 8 port internal card like ours, BUT is a true hardware card (has Intel processor and 256MB cache). It supports RAID 0,1,5,6,10, etc. It appears to support hot-swap, but their specs are a bit vague. So I could hopefully convert a couple of those extra ports to external (we got our adapter today).
- Anyone familiar with this card (or at least family or manufacturer)?
- This being true hardware RAID, would we be any safer running a RAID 5 with 3 drives on this (less chance of data loss on repair, and higher throughput on write speed)?
- Any other brands that might have something (I checked Highpoint, sonnet, firmtek, caldigit) and work well with Mac?
THANKS!
I've not used that model, but I'm rather familiar with Areca (I have 2 of them).
They're one of the best companies out there, and are well known. That particular card can even boot OS X & Windows. As it's hardware, it contains an NVRAM solution to RAID5 & 6's write hole issue. It will outperform software RAID, particularly in parity based arrays (not so much for 0/1/10), but you have additional ports that can be used on the array (8 total), and if you use all of them, you will get faster throughput due to the increased parallelism alone. You can use eSATA cables to go from internal ports to external drives if needed, as it would be a tough fit to get all of them internally, unless you moved the optical drive external, and used the bay for the additional drives.
I'm not a big fan of Highpoint actually (save the 43xx series), and despise CalDigit (really bad experience with their gear).
OK, so this could possibly be OPTION 1 for us:
- Areca ARC-1120 RAID controller
- 4x Western Digital RE3 500GB drives (possibly 3 in RAID 5 + system drive)
Total cost: $800 or so plus tax.
Now I'm thinking for a second option that would have a lot more future upgrade potential - some sort of external RAID enclosure with decent hardware RAID card built-in. Probably 4-5 drive bays. Then a simple eSATA card that might offer hot-swap, but not necessarily RAID (since enclosure would handle this).
I'm imagining it might be a hair more expensive, but then we'd be able to port it over to a new system in the future (wouldn't be tied to the PCI card solution).
- Any suggestions here???
- Areca ARC-1120 RAID controller
- 4x Western Digital RE3 500GB drives (possibly 3 in RAID 5 + system drive)
Total cost: $800 or so plus tax.
Now I'm thinking for a second option that would have a lot more future upgrade potential - some sort of external RAID enclosure with decent hardware RAID card built-in. Probably 4-5 drive bays. Then a simple eSATA card that might offer hot-swap, but not necessarily RAID (since enclosure would handle this).
I'm imagining it might be a hair more expensive, but then we'd be able to port it over to a new system in the future (wouldn't be tied to the PCI card solution).
- Any suggestions here???
The ARC-1120 can handle 8 drives, so you won't need another card. You can use external enclosures. Cabling could be an issue if adapters are used, but just run cables from the internal ports on the card, and out through a PCI bracket to an external enclosure that accepts one eSATA cable per drive. No Port Multiplier gear is needed.
The other option is to make room internally, such as pulling the optical drive out of the system, and moving it externally. Then use the newly empty bay for additional drives (2x 3.5" are certainly possible, possibly a 3rd if you're creative).
This should be a less expensive way to go, as at least one card is eliminated.
As far as the ability to transfer, probably not due to the systems now switching to PCIe (unless you meant a PC), as you'd need to get a new card that works in the next system. The drives could be transferred however, and the external enclosure as well, if you pay attention to the cabling details.
Hey, quick question about your Areca cards. I got a message back from Areca concerning my inquiry, and they too recommended the ARC-1120. But they specifically stated that they're not sure about compatibility with some G5s. It sounded like they'd had some problems. This obviously makes me pretty nervous. What system and what cards were you using? Are there drivers up-to-date for Mac? Do they seem to support Mac decently?
Yeah, I was actually thinking more about finding an external enclosure that would have the RAID card built-in to it. Then I would just find a simple eSATA adapter that allowed hot-swapping so I could use it with this external enclosure and our individual backup drives. This way, I could move the RAID to any future computer that just had eSATA. No internal RAID cards necessary.
- Any recommendations for a good external enclosure with decent built-in RAID card?
THANKS!
It's PPC, and was harder to support I would think.
I'm using PCIe models, and they do keep up with things. Drivers and firmware. They're rather good at that. Sometimes it can seem like a long time between updates, but it takes a lot of testing to develop them properly, and thier market demands it if they want to stay in business.
In the case of MP's, the updates are what you have to worry about, and haven't had major issues with it, though no issue free.
You don't actually need a unit with the RAID built in. Just get a PM (Port Multiplier) enclosure, with a decent eSATA card that has the ability to perform software RAID via drivers (RAID10 support is common enough).
As the array is based off the drivers, it can be transferred. Best to use a separate drive for the OS with these, to avoid any issues with booting.
I don't usually deal with these often enough to know which units are best, so just check around at what's out there, and read up on professional reviews and customer reviews on places like newegg.
Yeah, I was actually thinking that if the external enclosure had a decent 'true hardware' RAID card built-in, it could function kind of independent of the system it's connected to. You would still get the good performance of a true hardware RAID system, but it wouldn't matter really what system you were plugging it into - PCI G5, PCIe MP, Tiger, Leopard, Snow Leopard, even possibly a PC (I didn't say that). Just seems safer and more flexible.
But I looked around briefly and they're not easy to find. Most boxes are using port multiplying like you're stating. But all that really does is cut down on cable clutter, right?!? If I was going to do software RAID, I might as well stick with internal (we already bought the G5 Jive).
The few that said 'hardware RAID' were using an Oxford or SI chipset to handle RAID functions. But is that just 'fake RAID' like my Highpoint card?
It does cut down on clutter, but it's also inexpensive. There is a cost however, and that's predominately throughput and features.
You can also use enclosures that have an eSATA port per drive, also attached to an eSATA card. You have the potential of greater throughput for similar money, if not a little less (not much, but every cent you can save, assuming the gear is what you need and works well, is a good thing).
Those are also FakeRAID solutions, and PM units tied to eSATA cards are as well. In fact, they usually are built out of Silicon Image or Oxford chips. If you go this route, go with a Silicon Image, as they have fewer issues than Oxfords do. Just make sure the card does support PM enclosures.
(speaking about RAID 5 being better than RAID 10 for sequential read/write.)
Are you sure? I mean, what makes you so sure your actual workload is significantly more sequential than random? Or may be you don't care about what happens on your drives and just want to be sure they deliver the best sequential IO, even if it kills your random IO (that's just a choice).
Before you read the following, I want you to understand that I don't intent to deliver the absolute truth below. It's just a bunch of remarks, questions, and few attempts to answer these questions. Every constructive criticism is very welcome.
Focussing on sequential access (and if you are right, most people here are), is like focussing on megapixel when choosing a digital camera. That's important, but that's not enough. And may be it's not even the most important thing.
I would like to make a quick comparison (data taken from storagereview.com).
Let's take 1 SATA HDD and 1 SSD :
Ok, it's not the best SSD around, but recent SSDs don't outperform this one by a factor 5. So it's ok.
Compared to HDDs, SSDs don't offer a very good "Maximum Transfer Rate". They are good, but not really better. And sequential access is bound to this tranfer rate.
Compared to HDDs, SSDs do offer a *very* good "Average Random Access Time", and a very good number of I/O per second: about 120 times faster for random access time, and about 10 ou 20 times the number of I/O. And random access is bound to random access time and number of I/O per sec.
You'll have to admit, most people here want SSDs, because they think SSDs deliver really good performances (they are right).
I don't think people want to pay ~3$ per GB for a SSD that deliver between 110 and 200 MB/s Maximum Transfer Rate when 1 or 2 HDD will deliver the same at only ~0.1$ per GB.
So why do people love their SSDs, if it's not about random access performance boost they deliver?
If they really feel the difference (and they do), that's because random access is everywhere on the file system, and it happens during every workloads.
Sometimes, under particular workloads, random access "disappears" under a huge sequential transfer (big files duplication/transfer ), but it's still here, under the hood. That's why Apple has introduced an auto-defragmentation in Mac OS X, that's why ZFS optimize writes so that it can flush them sequentially. Everybody tries to reduce random access because it's what kills the performance of your drives.
In the real world, it's a matter of balance between sequential and random access (and it's quite complicated to separate one from the other), but random access is definitively of paramount importance.
I'm in the process of collecting data about my own I/O using DTrace tools, during my regular daily workload (client side, not server side). So far, it shows that unless you play with very big files (media postprod, big content duplication...) most of daily usage is more random than sequential (email, browsing, chat, coding, usenet). You can take a look at a first test here : http://patpro.net/~patpro/iopattern2.eps
regards
Are you sure? I mean, what makes you so sure your actual workload is significantly more sequential than random? Or may be you don't care about what happens on your drives and just want to be sure they deliver the best sequential IO, even if it kills your random IO (that's just a choice).
Before you read the following, I want you to understand that I don't intent to deliver the absolute truth below. It's just a bunch of remarks, questions, and few attempts to answer these questions. Every constructive criticism is very welcome.
Focussing on sequential access (and if you are right, most people here are), is like focussing on megapixel when choosing a digital camera. That's important, but that's not enough. And may be it's not even the most important thing.
I would like to make a quick comparison (data taken from storagereview.com).
Let's take 1 SATA HDD and 1 SSD :
Code:
- Western Digital Caviar Black WD1001FALS (1000 GB SATA)
Average Random Access Time (Read) 12.2 ms
Average Random Access Time (Write) 13.2 ms
Maximum Transfer Rate 111.0 MB/sec
IOMeter File Server - 1 I/O 89 IO/sec
Code:
- MTRON MSP-SATA7035-64 (64 GB SATA)
Average Random Access Time (Read) 0.1 ms
Average Random Access Time (Write) 6.8 ms
Maximum Transfer Rate 108.0 MB/sec
IOMeter File Server - 1 I/O 688 IO/secOk, it's not the best SSD around, but recent SSDs don't outperform this one by a factor 5. So it's ok.
Compared to HDDs, SSDs don't offer a very good "Maximum Transfer Rate". They are good, but not really better. And sequential access is bound to this tranfer rate.
Compared to HDDs, SSDs do offer a *very* good "Average Random Access Time", and a very good number of I/O per second: about 120 times faster for random access time, and about 10 ou 20 times the number of I/O. And random access is bound to random access time and number of I/O per sec.
You'll have to admit, most people here want SSDs, because they think SSDs deliver really good performances (they are right).
I don't think people want to pay ~3$ per GB for a SSD that deliver between 110 and 200 MB/s Maximum Transfer Rate when 1 or 2 HDD will deliver the same at only ~0.1$ per GB.
So why do people love their SSDs, if it's not about random access performance boost they deliver?
If they really feel the difference (and they do), that's because random access is everywhere on the file system, and it happens during every workloads.
Sometimes, under particular workloads, random access "disappears" under a huge sequential transfer (big files duplication/transfer ), but it's still here, under the hood. That's why Apple has introduced an auto-defragmentation in Mac OS X, that's why ZFS optimize writes so that it can flush them sequentially. Everybody tries to reduce random access because it's what kills the performance of your drives.
In the real world, it's a matter of balance between sequential and random access (and it's quite complicated to separate one from the other), but random access is definitively of paramount importance.
I'm in the process of collecting data about my own I/O using DTrace tools, during my regular daily workload (client side, not server side). So far, it shows that unless you play with very big files (media postprod, big content duplication...) most of daily usage is more random than sequential (email, browsing, chat, coding, usenet). You can take a look at a first test here : http://patpro.net/~patpro/iopattern2.eps
regards
Interesting topic. It totally makes sense to me. And I had noted that in our ancient RAID 5 setup here, the only numbers that looked decently respectful were the sequential read with larger blocks (256k).
So I would imagine you would be saying that a RAID 10 might have almost similar sequential read times, but much better random read/write, correct?
But for us, who are not working on the machine at all, and only opening mostly larger files from it over the network (usually at least 2MB+, sometimes with photos 100MB+), would we fall into the category where the larger sequential read times are most important? And what about write times, I've heard that a RAID 10 is much faster for write operations than a RAID 5?!?
I know this was posted a week ago, but these numbers really blew me away. How many disks is this? Thats outright pathetic.
I just got some new HDs and have a 2 disk software stripe for my system and user. I just ran the Xbench test and it's way way faster than your raid. Of course I'm on an 09 Mac Pro though.
Yeah, those numbers do pretty much blow ours away
And I was hoping more people would help us examine these numbers and see what we could expect from an upgrade.
But we obviously have to keep in mind our specs: Ancient G4, PCI bus RAID interface ('fake' RAID at that), and older Seagate HDs that are pretty full. So this is a fairly worst-case scenario for a typical RAID 5 I would imagine.
That's why I was trying to see what we could do to gain some performance. But our upgraded server will still be fairly old and stuck with a PCI bus.
So that's why I was wondering if a software RAID 1 on the built-in SATA ports might be better, or a RAID 10 with our existing card, or finally upgrading to a 'true' hardware RAID card to do a RAID 5 or 10.
And I was hoping more people would help us examine these numbers and see what we could expect from an upgrade.
But we obviously have to keep in mind our specs: Ancient G4, PCI bus RAID interface ('fake' RAID at that), and older Seagate HDs that are pretty full. So this is a fairly worst-case scenario for a typical RAID 5 I would imagine.
That's why I was trying to see what we could do to gain some performance. But our upgraded server will still be fairly old and stuck with a PCI bus.
So that's why I was wondering if a software RAID 1 on the built-in SATA ports might be better, or a RAID 10 with our existing card, or finally upgrading to a 'true' hardware RAID card to do a RAID 5 or 10.
I think it's kind of a "ya get what ya pay for" kinda thing. The G4 is looooog dead and fairly useless. As you're seeing in RAID performance it's almost not worth the electricity it takes to run the thing. I get about the same drive I/O specs as Abidubi with a 3-drives RAID0 on a 2006 mac pro embedded RAID. I think you can find a 2006 now for about $1,000. Right?
eSATA hardware RAID5 tower five drive
Well for a file server - it's prefer a Hardware RAID - Sonnet or High point, both are FakeRAID (Linux term).
I would suggest look at this Five drive hardware raid:
http://eshop.macsales.com/item/DAT Optic/T5R5ESUF/
It's fast and OS independent - you can move to MACPro if needed
After digging i also found this.
Chris Wilson @ Cactus Field comment on this
http://www.datoptic.com/eBOX-R5_comment.html
I wish I could have that instead ARC-1210 with four HDD
Well for a file server - it's prefer a Hardware RAID - Sonnet or High point, both are FakeRAID (Linux term).
I would suggest look at this Five drive hardware raid:
http://eshop.macsales.com/item/DAT Optic/T5R5ESUF/
It's fast and OS independent - you can move to MACPro if needed
After digging i also found this.
Chris Wilson @ Cactus Field comment on this
http://www.datoptic.com/eBOX-R5_comment.html
I wish I could have that instead ARC-1210 with four HDD
AHA! That looks like what I was talking about for an external hardware RAID enclosure! Not sure how I missed that at OWC's website. I had been looking everywhere for something like this.
So it appears that the 'hardware RAID' part of this is provided by JMicron (http://www.jmicron.com/Product_JMB394.htm). I don't know how to properly 'interpret' the specs - so is this a 'true' hardware RAID solution outside of the computer?
If this is good (anyone know about the company/reputation) it might be way more flexible than the ARC-1120. We would be able to use it with any server we ended up with, and I imagine it would be immune from any OS update issues, as it's all self-contained.
- So any suggestions for a basic eSATA PCI interface that would allow hot-swapping so we could also use it with our backup drives?
- And any other solutions similar to the DAT Optic here?
THANKS!
"Supports RAID 0, 1, 10, 5, 50, CLONE and JBOD" <-- no kidding ? RAID 50 with only 5 HDDs. Harry Potter's RAID !
Unfortunately, it's not a proper hardware solution at all. It's still FakeRAID. No cache or processor on the board. It's still driver based.
You'd be better with the ARC-1120 if you want a true hardware RAID solution. But given the overall system, you might want to wait until you get a newer system (at least something with an 8x PCIe slot).
As your needs aren't that stringent for throughput, the software/FakeRAID solutions in a non parity based array type would suit your current needs IMO (using the G5).
Hmmm, I'm a bit confused. According to their specs (below) they 'say' it has a hardware accelerated RAID engine, etc. And if it's connected by just a single eSATA port (or firewire for that matter) I don't see how it would tax your internal processor & system like a 'fake' RAID might. But maybe I'm still missing something. I do understand the RAM cache thing a bit, although it states 'RAID 3/5 Write-Back Cache'. Are they misleading a bit?
And again, not sure if this is apples-apples comparison, but the specs on their site for performance (http://www.datoptic.com/) look better than what I would think we'd get with our Highpoint RR 'fake' RAID internally in a RAID 5. They say write/read times over 200MB/s in a 5-drive RAID 5. Would we get much more than that with the internal Areca ARC-1120 on a standard PCI bus?