Resolved Optimising Backups

[Macman45]

Scenario as it stands now:

I have 3 Macs, 1 MBP one MBA one iMac. All are backed up to a 3TB TC on my network.

In addition to this, I have a four bay USB Dock containing swappable HDD's of various size to which I archive recent material.

I want to ditch the slow Startech USB dock, so I have purchased a 4X1TB Promise Pegasus Raid array which will be attached to the iMac via T/bolt.

I want to retain the TC backup system, it works well, and the TC is used soley for this purpose.

Question is....Since I have not used Thunderbolt for anything except target mode once, just how fast is that Pegasus going to be? I have as I see it two choices here:

1) Use the Promise as a dedicated RAID array ( Not keen on this) OR

2) If it really is fast enough, use the 4X1TB Pegasus to store things I'm working on. Question is, can I really use it reliably (And Quickly) to edit video and music in real time?


I'm up to my neck in too much hardware so all ideas much appreciated!

 

[simsaladimbamba]

Since TB allows you to access those 1 TB HDD in the Pegasus enclosure at full speed, you will get up to 120 MB/ sequential read and write speeds, which is more than enough to edit four tracks of HD video played back using an editing codec, but then again, if the four video tracks all access data on one HDD, it might not be fluent enough due to the read/write head of that HDD constantly searching and reading four individual files at the same time.
With RAID it would be quicker, but then again, you have to plan for redundancies.

I recently edited 1080p HD footage via Avid Media Composer on my 2009 MBP with the video stored on the internal HDD (Seagate Momentus 7200.4) and the OS on an internal SSD and it was quite okay to do so.

As music has a much lower data rate, even uncompressed, you shouldn't run into problems, but then again, if you use dozens of tracks all accessing the same HDD, the slow random access times of any HDD will work against you.

 

[throAU]

a 4 disk raid array attached via thunderbolt should be faster than your internal disk, if it is formatted in RAID10 or RAID0 - and faster for READS if it is in RAID5.

Writes to RAID5 won't be much faster (maybe even slower).

If you can configure the raid level, the benefits go like this:

RAID0 - 4x throughput of single disk, 4x capacity of single disk (e.g., up to 480 megabytes/sec if data is accessed sequentially). 1 drive failure = all data lost.
RAID10 (stripe across 2 mirror sets) - 2x write throughput of single disk, 2x read throughput (or 4x, depending on how smart the controller is), 2x capacity of single disk, can tolerate up to 2 disk failures (one in each mirror). Performance with one failed disk = same as normal, but when disk is replaced will require a short rebuild time of reduced performance.
RAID5 (stripe across 3 disks with distributed parity) - 3x read throughput, poor write throughput, can tolerate 1 drive failure (and then run in degraded mode, which will be quite slow)


If you have good (regular!) backups elsewhere, run the box in RAID0 for max performance.... if you need speed AND resiliency, run in RAID10. If it is for archiving and write performance is not important, run in RAID5 (or JBOD - individual non-raided disks).

 

[Macman45]

Since TB allows you to access those 1 TB HDD in the Pegasus enclosure at full speed, you will get up to 120 MB/ sequential read and write speeds, which is more than enough to edit four tracks of HD video played back using an editing codec, but then again, if the four video tracks all access data on one HDD, it might not be fluent enough due to the read/write head of that HDD constantly searching and reading four individual files at the same time.
With RAID it would be quicker, but then again, you have to plan for redundancies.

I recently edited 1080p HD footage via Avid Media Composer on my 2009 MBP with the video stored on the internal HDD (Seagate Momentus 7200.4) and the OS on an internal SSD and it was quite okay to do so.

As music has a much lower data rate, even uncompressed, you shouldn't run into problems, but then again, if you use dozens of tracks all accessing the same HDD, the slow random access times of any HDD will work against you.

Thanks...I was thinking about organising the 4X1TB's so that projects are only accessing one drive at a time. It's reassuring to know that the speeds they claim are there. The Startech USB was only suitable for storage.

 

[Macman45]

a 4 disk raid array attached via thunderbolt should be faster than your internal disk, if it is formatted in RAID10 or RAID0 - and faster for READS if it is in RAID5.

Writes to RAID5 won't be much faster (maybe even slower).

If you can configure the raid level, the benefits go like this:

RAID0 - 4x throughput of single disk, 4x capacity of single disk (e.g., up to 480 megabytes/sec if data is accessed sequentially). 1 drive failure = all data lost.
RAID10 (stripe across 2 mirror sets) - 2x write throughput of single disk, 2x read throughput (or 4x, depending on how smart the controller is), 2x capacity of single disk, can tolerate up to 2 disk failures (one in each mirror). Performance with one failed disk = same as normal, but when disk is replaced will require a short rebuild time of reduced performance.
RAID5 (stripe across 3 disks with distributed parity) - 3x read throughput, poor write throughput, can tolerate 1 drive failure (and then run in degraded mode, which will be quite slow)


If you have good (regular!) backups elsewhere, run the box in RAID0 for max performance.... if you need speed AND resiliency, run in RAID10. If it is for archiving and write performance is not important, run in RAID5 (or JBOD - individual non-raided disks).

Thanks...Good info. Shipping date on the Promise has slipped for some reason to the 29th May, so I have a little time to work out the best plan of action. It's the speed I want to maximise.

 

[flynz4]

a 4 disk raid array attached via thunderbolt should be faster than your internal disk, if it is formatted in RAID10 or RAID0 - and faster for READS if it is in RAID5.

Writes to RAID5 won't be much faster (maybe even slower).

If you can configure the raid level, the benefits go like this:

RAID0 - 4x throughput of single disk, 4x capacity of single disk (e.g., up to 480 megabytes/sec if data is accessed sequentially). 1 drive failure = all data lost.
RAID10 (stripe across 2 mirror sets) - 2x write throughput of single disk, 2x read throughput (or 4x, depending on how smart the controller is), 2x capacity of single disk, can tolerate up to 2 disk failures (one in each mirror). Performance with one failed disk = same as normal, but when disk is replaced will require a short rebuild time of reduced performance.
RAID5 (stripe across 3 disks with distributed parity) - 3x read throughput, poor write throughput, can tolerate 1 drive failure (and then run in degraded mode, which will be quite slow)


If you have good (regular!) backups elsewhere, run the box in RAID0 for max performance.... if you need speed AND resiliency, run in RAID10. If it is for archiving and write performance is not important, run in RAID5 (or JBOD - individual non-raided disks).

I have been doing some tests with my 8TB Pegasus R4. I have found very little difference in read or write speed between RAID0 and RAID5. Furthermore... even if there was a speed difference, client computers primarily do reads... to the extent that for most people and most applications, you can safely ignore write performance. This is NOT necessary the case for server/enterprise applications.

I have come to the conclusion that RAID 5 is the obvious best solution for my usage (and that of most consumers using a client machine). RAID 0 gives almost no extra speed at a significant loss in protection. RAID 10 is too wasteful of capacity.

OP: Unless you have sufficient SSD capacity to hold your entire project... your Pegasus will probably be the fastest storage in your machine. Thunderbolt flies.

/Jim

 

[throAU]

Q: with your RAID5 write testing, are you writing to clean storage, or over-writing?


An partial stripe overwrite with RAID5 will require a READ from all 4 disks, parity calc and then a write to 2 or more disks.

Writing to RAID10 does not require the initial read, as there is no parity to calculate.

If you haven't tried over-writing, try that - without some very clever write coalescing, there is no way that RAID5 can keep up with RAID10 on write (at least not to previously written sectors).

It all depends on your requirements (streaming video was mentioned, presumably write intensive).

Disk capacity is cheap - however getting more speed is not. I.e., throwing away half your capacity to use RAID10 is not a major cost vs RAID5 (cost of 1/4 your capacity). But it depends on your speed requirements. RAID5 performance also SUCKS hard when you have a failed disk and takes a long time to rebuild (i doubt a RAID5 array of 4 sata disks will keep up with HD video - so you're effectively "down" for that task anyway, even if your data is "available", its not fast enough), during which you may have a second failure, then you're screwed (well, back to your backup).... with RAID10, performance is maintained and up to half your disks can fail before you go back to backups...

Essentially, unless you NEED the additional capacity, IMHO running RAID5 is to gain 1/4 of your capacity back is not worth the tradeoff vs RAID10. But as you say, horses for courses....

It would be an idea for the OP to play with RAID0, RAID10 and RAID5 for his particular application before making a choice, and evaluate whether or not the trade-offs each level make are acceptable in his situation.

OP: Unless you have sufficient SSD capacity to hold your entire project... your Pegasus will probably be the fastest storage in your machine. Thunderbolt flies.

Agreed there.. thunderbolt is essentially PCIe, and the disk subsystem is not going to saturate that in a hurry. Even enterprise storage arrays are mostly only 10 gigabit today.

 

[Macman45]

I have been doing some tests with my 8TB Pegasus R4. I have found very little difference in read or write speed between RAID0 and RAID5. Furthermore... even if there was a speed difference, client computers primarily do reads... to the extent that for most people and most applications, you can safely ignore write performance. This is NOT necessary the case for server/enterprise applications.

I have come to the conclusion that RAID 5 is the obvious best solution for my usage (and that of most consumers using a client machine). RAID 0 gives almost no extra speed at a significant loss in protection. RAID 10 is too wasteful of capacity.

OP: Unless you have sufficient SSD capacity to hold your entire project... your Pegasus will probably be the fastest storage in your machine. Thunderbolt flies.

/Jim

Noted and thanks for the information...I'm going to see which woks best in real time for me anyway...I'm normally an early adopter but T/Bolt has been a very slow starter, so no experience with it to relate.

 

[flynz4]

Q: with your RAID5 write testing, are you writing to clean storage, or over-writing?


An partial stripe overwrite with RAID5 will require a READ from all 4 disks, parity calc and then a write to 2 or more disks.

Writing to RAID10 does not require the initial read, as there is no parity to calculate.

If you haven't tried over-writing, try that - without some very clever write coalescing, there is no way that RAID5 can keep up with RAID10 on write (at least not to previously written sectors).

It all depends on your requirements (streaming video was mentioned, presumably write intensive).

I tried a number of tests, but with 8TB/6TB (RAID0/RAID5)... it is hard to tell where the data was placed.

In any case... it is moot. the vast majority of client platforms perform very limited writes. Your example of video streaming... the video is written once (as a large file nonetheless), and then later read (streamed) many times.

One of my more demanding applications is Aperture... where each master and version is written approximately once (at import), and the vast majority never change. Yet... this large database (>50K unique images) is read repeatedly as I navigate and search through the library. The read queues are getting hammered, yet the writes are almost non-existent. The read performance is what affects the usability of the application. This is the dominant client behavior.

The obvious write exception of course is the OS, paging, etc... but those writes would be on the internal SSD (or HDD for those less fortunate)... not on an external RAID array.

/Jim

 

[throAU]

Capturing/Editing raw HD video will require high write throughput (which is kinda what I meant by "Streaming" - not playback of compressed data, you can do that from a usb memory stick over USB, that doesn't need speed at all...).

But yes, I agree most usage is read-centric.

However, there is also the RAID5 degraded performance issue to take into account. The performance when degraded SUCKS. Rebuilds take a long time.

Which means:
- if you need fast storage: while degraded, it is useless (you're "down" as far as getting work done goes anyway) - there's no point having fault tolerant storage (presuming you have a backup, as you should - because RAID is not a replacement for backups) if you can't work during degraded operation? May as well go RAID0 and get more space/faster writes...
- if you need resilient storage, there is an extended window during which time your data is at risk of double-disk failure, and the remaining disks are under much higher load, which may cause said failure to happen.

combine the above and you may be better off with RAID0 as it should be faster on writes (certainly when you are updating a previously written block - and gives you more space), or if you need resiliency, RAID10 (at the cost of an extra disk worth of capacity).

RAID5 was a lot more attractive when disks were $1/megabyte - these days with capacity going for $0.10/gigabyte or less, most people have way more storage than they need. You may as well use the spindles (i.e., "waste" disk capacity, as it is cheap) for resiliency or speed.

 

[Macman45]

Capturing/Editing raw HD video will require high write throughput (which is kinda what I meant by "Streaming" - not playback of compressed data, you can do that from a usb memory stick over USB, that doesn't need speed at all...).

But yes, I agree most usage is read-centric.

However, there is also the RAID5 degraded performance issue to take into account. The performance when degraded SUCKS. Rebuilds take a long time.

Which means:
- if you need fast storage: while degraded, it is useless (you're "down" as far as getting work done goes anyway) - there's no point having fault tolerant storage (presuming you have a backup, as you should - because RAID is not a replacement for backups) if you can't work during degraded operation? May as well go RAID0 and get more space/faster writes...
- if you need resilient storage, there is an extended window during which time your data is at risk of double-disk failure, and the remaining disks are under much higher load, which may cause said failure to happen.

combine the above and you may be better off with RAID0 as it should be faster on writes (certainly when you are updating a previously written block - and gives you more space), or if you need resiliency, RAID10 (at the cost of an extra disk worth of capacity).

RAID5 was a lot more attractive when disks were $1/megabyte - these days with capacity going for $0.10/gigabyte or less, most people have way more storage than they need. You may as well use the spindles (i.e., "waste" disk capacity, as it is cheap) for resiliency or speed.

This is my way of thinking too...The TC takes care of backups, I have a single bay USB caddy which I can use to archive large files...It's the performance of the Pegasus I'm after here, not it's RAID capabilities...A plan begins to form..

 

[flynz4]

Essentially, unless you NEED the additional capacity, IMHO running RAID5 is to gain 1/4 of your capacity back is not worth the tradeoff vs RAID10. But as you say, horses for courses....

I wanted to follow up with you and let you know that I will be switching from RAID 5 to RAID 10.

I attended an executive meeting today... and talked with the key architect of the PMC RAID chip that is used in the Pegasus R4. After a lengthly discussion, I became convinced that RAID 10 would be best for my configuration. I felt that I owe you a follow-up since this thread is what prompted my discussion. Thanks.

/Jim

 

[Macman45]

I'm going to close this one resolved now, even though it isn't, but I have another thread on specific issue's I'm having with my new R4. Thanks for the help here though..

 

[driftless]

I asked this question in another thread and it wasn't answered but it is topical here. My setup is very similar to the OP's and I am also thinking about adding a 4 x 1TB Raid array. My question is will Time Capsule back up there array if it is set at Raid 5?

 

[Macman45]

I asked this question in another thread and it wasn't answered but it is topical here. My setup is very similar to the OP's and I am also thinking about adding a 4 x 1TB Raid array. My question is will Time Capsule back up there array if it is set at Raid 5?

Yes...I have the new R4...TC continues to backup as normal, which is why I need to free up one of the 4 drives on the Pegasus...If you choose to include the Array (The default is exclude) the Time machine will back it up...It's overkill though...At present I have the new R4 set to RAID 5 which I'm attempting to change, without much success so far) and my Time machine backups go to a TC....I ensured that the R4 was Excluded from the backup set on the TC....I have 5 backups+ the physical drive on my iMac...Unlucky I would have to be for them all to fail!

 

[driftless]

Does it make a single copy of what is on the Raid5 drive? Is it readable so it can be used as backup in case of failure and the need to backup?

 

[Macman45]

Does it make a single copy of what is on the Raid5 drive? Is it readable so it can be used as backup in case of failure and the need to backup?

Unsure, But I assume you would be able to restore from it...GTFS notwithstanding....I'm rebuilding the array, but I'm pretty sure I'm going to be back at RAID5. It would seem that with an iMac, and not a MAc Pro, it's all you get.

I can live with it, but I might pull a drive out...Sort of defeats the object I know, but the Promise software is Garbage.