Is the PROMISE Pegasus2 for me.

[puckhead193]

With FCP X converting my XDCAM files into Pro-res my 4TB scratch drive is filling up fast with only a month of use.

I'm thinking it's finally about time to break down and buy a disk array. The natural choice is the Pegasus2. I don't need a RAID. Once my project is done, I can delete it (if I wanted too) as it gets uploaded the my station servers and I'm done with it. I just save it if I wanted it for my reference. So if it fails, it's not the end of the world, (after i'm done with the project of course)
So my thinking is that once I fill up the disk in the Pegasus I can replace the HDDs and it will be more cost affective then going out and buying another G-Tech Thunderbolt drive.

What drives should I replace the original one with once they are fill?

Am I thinking straight or am I missing something.

Thanks!

 

[FrankHahn]

It is definitely for you! You can use its high speed and vast storage space in Raid 5. Depending on your need, you may get an R4, an R6, or an R8.

You can replace the installed HDs with nearly all kinds of HDs, including 2.5" SSDs and HDs with adapters.

Depending on your need for data safety and for data-transfer speed, you can use consumer-level HDs (cheaper) or server HDs or professional HDs (more expensive).

 

[puckhead193]

It is definitely for you! You can use its high speed and vast storage space in Raid 5. Depending on your need, you may get an R4, an R6, or an R8.

You can replace the installed HDs with nearly all kinds of HDs, including 2.5" SSDs and HDs with adapters.

Depending on your need for data safety and for data-transfer speed, you can use consumer-level HDs (cheaper) or server HDs or professional HDs (more expensive).

Besides the bays, the only difference I see are the RAID levels. are they any differences between the 3 units? Will the R4 be as fast as the R6/R8?

 

[Yahooligan]

Also consider the Areca Thunderbolt arrays, excellent pieces of hardware. I'm using Seagate 3TB Barracuda drives in mine and they've been great, but you do need to watch out for drives that don't support TLER when installing them in a hardware RAID enclosure. Some drives that don't support it work fine in hardware RAIDs while others will fall out of the RAID randomly.

 

[snouter]

I like mine.

I bought an R4 8TB. One day I'll toss 4TB drives in there and I guess it would become an R4 16TB.

It can be a little fussy about drives from what I've read elsewhere. Stick to the approved list for best results.

http://www.promise.com/media_bank/D...egasus2_Compatibility List v1.1-201301216.pdf

Oddly, mine came with 2TB 7200GB Toshiba drives. Don't really think of Toshiba when I think of drives, but that's what's in there and they seem to work fine.

This also looks interesting:
http://www.blackmagicdesign.com/products/blackmagicmultidock

 

[DoFoT9]

Besides the bays, the only difference I see are the RAID levels. are they any differences between the 3 units? Will the R4 be as fast as the R6/R8?

Firstly, will you benefit from the Pegasus 2 (that has 20gb/s channels), or is your Mac only 10gb/s compatible?

Irregardless, you should use RAID. Even if it's RAID0 (I absolutely do not recommend this). I recommend RAID10 if you are more heavy on the reads or RAID5 if you're heavy on the writes (you're heavy on the writes, obviously).

There is a significant difference in speed between the R4 and the R8, the R8 has more drives, so obviously more capacity and ability to read/write. I can't recall the exact difference but I seem the recall the R8 being 15-25% faster overall.. Hope this helps.

Edit: Ah, I see you're on a nMP.

 

[elvisizer]

Once my project is done, I can delete it (if I wanted too) as it gets uploaded the my station servers and I'm done with it. I just save it if I wanted it for my reference. So if it fails, it's not the end of the world, (after i'm done with the project of course)
So my thinking is that once I fill up the disk in the Pegasus I can replace the HDDs and it will be more cost affective then going out and buying another G-Tech Thunderbolt drive.

sure, you can take the disks out when they're full, but I'm not sure why you would . . . you said you don't need to save the projects- just delete them. or archive them to another device or service (crashplan's good) and free up the space so you don't need to buy more drives. most cost effective of all!

 

[puckhead193]

I like mine.

How are your speeds? How do you have it configured?

Anyone using these with FCP X?

 

[Yahooligan]

I recommend RAID10 if you are more heavy on the reads or RAID5 if you're heavy on the writes (you're heavy on the writes, obviously).

You've got it backwards, RAID10 is more beneficial for write-heavy workloads because it doesn't suffer from the parity write penalty that RAID5 does. For read-heavy workloads you'll see RAID5 provide an advantage over RAID10. RAID5 will also provide more usable space than 10.

 

[snouter]

How are your speeds? How do you have it configured?

Anyone using these with FCP X?

Out of the box RAID5 with the OEM Toshiba 2TB drives.

500-600 MB/s.

I use it with Premiere Pro and After Effects with Canon HDSLR footage.

 

[DoFoT9]

You've got it backwards, RAID10 is more beneficial for write-heavy workloads because it doesn't suffer from the parity write penalty that RAID5 does. For read-heavy workloads you'll see RAID5 provide an advantage over RAID10. RAID5 will also provide more usable space than 10.

I'm sorry, Yahooligan, you are incorrect. RAID10 is more beneficial for read requests, as the RAID0 part of the array can read 2 alternate segments of data - for writing it must simultaneously do this to the drives, unlike reading which can alternate. RAID10 will cut your writes in half. Correct, RAID10 will mostly give less storage capacity, but higher redundancy. RAID5 taxes drives a little more also.

 

[Yahooligan]

I'm sorry, Yahooligan, you are incorrect. RAID10 is more beneficial for read requests, as the RAID0 part of the array can read 2 alternate segments of data - for writing it must simultaneously do this to the drives, unlike reading which can alternate. RAID10 will cut your writes in half. Correct, RAID10 will mostly give less storage capacity, but higher redundancy. RAID5 taxes drives a little more also.

Repeating what you said before won't make it true.

RAID5's write penalty is essentially double what it is for RAID10, RAID5 not only has to calculate the parity, it has to write that parity to disk...along with writing the data itself to disk, but only after reading the old data and parity. Under heavy write operations that write penalty with a factor of 4 and all the overhead will become your bottleneck.

RAID10 has a write penalty of 2; One write to each mirror.
RAID5 has a write penalty of 4: Read old data, read old parity, write new data, write new parity.

Also, don't mistake RAID10's additional mirror(s) as being more safe. On paper you can argue RAID10 can survive double disk failure, but since we cannot predict when or where a drive is going to fail it's entirely possible to lose both disks in a mirror and boom, data loss.

Still don't believe me? Plenty of documentation on the subject from Dell, EMC, Oracle...I thought for sure the RAID5 vs RAID10 write penalty was common knowledge. Guess some people missed that memo.

As for usable capacity, RAID10 is N/2, RAID5 is N-1. Do the math, that difference could be huge.

12-disk RAID10 using 12x2TB = 12TB usable. (12*2TB)/2
12-disk RAID5 using 12x2TB = 22TB usable (12*2TB)-2TB

10TB difference. No biggie.

 

[FrankHahn]

Besides the bays, the only difference I see are the RAID levels. are they any differences between the 3 units? Will the R4 be as fast as the R6/R8?

If you use a raid setting other than the mirror, the more disks you have, the faster the data-transfer speed is.

Let us take Raid5 as an example. Assume that all the disks are installed. Suppose that the data-transfer speed of a single disk is v.
R4: The data-transfer speed can be up to (4-1)*v = 3v.
R6: The data-transfer speed can be up to (6-1)*v = 5v.
R8: The data-transfer speed can be up to (8-1)*v = 7v.

In summary, the data-transfer speed is (n-1)v for n disks in Raid5 with v the data-transfer speed of a single disk.

 

[DoFoT9]

RAID10 has a write penalty of 2; One write to each mirror.
RAID5 has a write penalty of 4: Read old data, read old parity, write new data, write new parity.

You base your premise on the fact that the write parity is the point, which I don't think is really what's best to compare. You'll find in most cases that RAID10 will give better performance for both random and sequential reads, and writes it really depends on the scenario. Overall performance is probably the point, and write latency in RAID5 is higher than RAID10, however RAID5 tends to write faster (obviously has a higher write penalty of course, due to the parity).

Also, don't mistake RAID10's additional mirror(s) as being more safe. On paper you can argue RAID10 can survive double disk failure, but since we cannot predict when or where a drive is going to fail it's entirely possible to lose both disks in a mirror and boom, data loss.

This is true - and fair point. You can't predict which of the drives will fail, however keep in mind that during a 1 drive RAID5 failure that your disks are something like 170x more likely to have a write hole error. This is something I'd rather stay away from. I believe it's something like 1 in 11.3TB will give this error (URE). RAID1 will have a 25% probability on a full drive (3-4TB disks), RAID5 will have a 50% proability (3-4TB disks). That's not really the topic, though it is a concern.

Still don't believe me? Plenty of documentation on the subject from Dell, EMC, Oracle...I thought for sure the RAID5 vs RAID10 write penalty was common knowledge. Guess some people missed that memo.

Write penalty, obviously - but like I said.. Not always the concern nor the real world performance mirrors the theoretical stuff. Also, probably best to keep your remarks to yourself.

 

[Yahooligan]

You base your premise on the fact that the write parity is the point, which I don't think is really what's best to compare. You'll find in most cases that RAID10 will give better performance for both random and sequential reads, and writes it really depends on the scenario. Overall performance is probably the point, and write latency in RAID5 is higher than RAID10, however RAID5 tends to write faster (obviously has a higher write penalty of course, due to the parity).

Now you're just contradicting yourself. You start off by saying, correctly, that RAID5 has a higher write latency (Time to complete the write is higher due to the penalty of "4") but that RAID5 tends to write faster. You can't have it both ways. As you increase write IOPS you will see RAID5 performance start to suffer and start to fall off while RAID10 marches along with higher peak IOPS and write performance.

My "premise" is not based on parity, it's based on penalty and it's based on simple technological fact due to how RAID works.

RAID10 and RAID5 will provide nearly identical read performance as well.

This is true - and fair point. You can't predict which of the drives will fail, however keep in mind that during a 1 drive RAID5 failure that your disks are something like 170x more likely to have a write hole error. This is something I'd rather stay away from. I believe it's something like 1 in 11.3TB will give this error (URE). RAID1 will have a 25% probability on a full drive (3-4TB disks), RAID5 will have a 50% proability (3-4TB disks). That's not really the topic, though it is a concern.

Agreed.

Write penalty, obviously - but like I said.. Not always the concern nor the real world performance mirrors the theoretical stuff. Also, probably best to keep your remarks to yourself.

I don't work in theory, I work in the real world. You are giving bum advice in terms of RAID5 vs RAID10. RAID10 writes faster, both in terms of lower latency and higher throughput. Under light loads are you going to see a big difference? No, but as soon as you start running up IOPS you're going to start seeing a gap and that gap gets wider the more you hammer the writes with RAID10 being the clear winner.

To sum it up, if performance is what you're after and if your workload is mostly reads then go with RAID5 since it's cheaper $/GB and has excellent read performance, if it's mostly-writes then go with RAID10 at the expense of higher $/GB, it will win out over RAID5 10 times out of 10 (No pun intended...or was there?! ) and read performance is similar to RAID5.

 

[DoFoT9]

I don't work in theory, I work in the real world. You are giving bum advice in terms of RAID5 vs RAID10. RAID10 writes faster, both in terms of lower latency and higher throughput. Under light loads are you going to see a big difference? No, but as soon as you start running up IOPS you're going to start seeing a gap and that gap gets wider the more you hammer the writes with RAID10 being the clear winner.

Quite. Not sure why I was getting confused with myself.

 

[Yahooligan]

Quite. Not sure why I was getting confused with myself.

No worries, it happens to the best of us.

At the end of the day, in consumer-land, I think most people would be perfectly happy with RAID5 and not need to swallow the expense of RAID10. My little 4-disk Areca does 475MB-500MB/s read and write with RAID5 which is more than I need. The Pegasus/Pegasus2 seem to be running about the same in terms of performance. Neither the Pegasus or Pegasus2 will fully utilize TB1, let alone TB2, if using HDDs. Even the R8 won't max out TB1, so TB2 won't give you any advantage here except for allowing TB2 daisy chaining. So, if you plan on having multiple TB2 devices then the Pegasus2 may make device management easier, otherwise you could just put the TB1 Pegasus at the end of your TB2 chain.

If the TB2 version is the same price as the TB1 version then screw it, get the TB2 version.

This all changes if you do SSD RAID.

 

[puckhead193]

Well I ordered the PROMISE Pegasus2 R4. Hopefully it will arrive Monday or Tuesday from B&H. I just hope I didn't make a mistake.

 

[AidenShaw]

So my thinking is that once I fill up the disk in the Pegasus I can replace the HDDs and it will be more cost affective then going out and buying another G-Tech Thunderbolt drive.

If you're thinking that you can "archive" the data by storing the old drives, and later swap the old drives back to grab something old - be sure to verify that it is possible to do that.

Smart RAID controllers maintain meta-data about volumes - usually both on the disk and in NVRAM in the controller.

The controller might not be happy if you put in drives that don't match the NVRAM data - even for JBODs.

I've used RAID controllers that simply reformat "unknown" drives when you insert them!

 

[puckhead193]

If you're thinking that you can "archive" the data by storing the old drives, and later swap the old drives back to grab something old - be sure to verify that it is possible to do that.

Smart RAID controllers maintain meta-data about volumes - usually both on the disk and in NVRAM in the controller.

The controller might not be happy if you put in drives that don't match the NVRAM data - even for JBODs.

I've used RAID controllers that simply reformat "unknown" drives when you insert them!

Crap that's not good...

 

[Yahooligan]

I don't know about the Promise, but Areca stores the RAID info only on the disks so that should there be a controller failure you can install the disks in a replacement Areca and be up and running without issue. I would hope the Promise does the same.

 

[Jim-H]

I found this interesting thread on the Pegasus2 vs Areca

 

[puckhead193]

I spoke with Promise and they said it works liked I hoped and that If I wanted for achieving purposes I can do what I had planned and replace the drives once they are filled put them on the shelf and get new drives. If I wanted to go back to the old drives I can.

 

[Yahooligan]

I spoke with Promise and they said it works lille I hoped and that If I wanted for achibing purposes I can do what I had planned and replace the drives once they are filled put them on the shelf and get new drives. If I wanted to go back to the old drives I can.

Perfect, so they also store the RAID config data on the drives and not in the controller NVRAM. Good to know!

 

[AidenShaw]

I spoke with Promise and they said it works liked I hoped and that If I wanted for achieving purposes I can do what I had planned and replace the drives once they are filled put them on the shelf and get new drives. If I wanted to go back to the old drives I can.

That's good to know.

Perfect, so they also store the RAID config data on the drives and not in the controller NVRAM. Good to know!

It could be stored both places, with the firmware written to prefer the on-disk metadata.