Post Reply 
 
Thread Rating:
  • 0 Votes - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Questions about Drobo 5D performance
08-13-2013, 01:49 PM
Post: #1
Questions about Drobo 5D performance
I am a video systems integrator and one of the things I do is recommend and build storage systems optimized for video. Drobo was often talked about but rarely on the radar as the performance of its various models wasn't considered adequate. Great for backup, archiving, replication. As a main media volume? Not likely.

Recently though I heard of 5D posting really good numbers (300MB/s reads and writes) - in Windows via TB of all things, and got really curious. 5D's support for dual parity and zero-bay SSD caching makes it quite unique amongst its competitors. I think it's a great little box, even if not exactly cheap.

[Image: 2013-08-13+Drobo5D_DiskSpeedTest+-+Janusz+Szewczak.png]

Would love to hear if any editors out there use it, and what your impressions are. If you could answer one or more of these questions, that'd be awesome.

- Anyone else achieving similar or better numbers? If so, what's your system configuration?
- What are the speeds with using USB 3.0?
- What are the speeds in dual parity mode?
- Does 5D support UAS? If not - any plans?
- Does adding an mSATA drive hurt sequential I/O speeds at all?
- Has anyone tested performance (sequential and random I/Os) when the 5D is 90% and 99% full?
- What's the reason Enterprise SSDs are not supported? (I realize SAS models aren't - there are SATA models like Intel S3700 that should work.)
- What are rebuild times per TB in single and dual parity modes? (With drives like WD SE)
- Does 5D support "RAID bitmap" (as in Linux RAID) or something similar allowing near-instantaneous rebuild when a drive was temporarily offlined or re-inserted?

Thanks!
Visit this user's website Find all posts by this user
Quote this message in a reply
08-13-2013, 01:55 PM
Post: #2
RE: Questions about Drobo 5D performance
(08-13-2013 01:49 PM)beans Wrote:  - Has anyone tested performance (sequential and random I/Os) when the 5D is 90% and 99% full?

the 5D is about the only one i havent owned,

i have owned the :

Drobo v1
Drobo v2
DroboPro
Drobo FS
Drobo 5N

For all of the above - it not recommended to fill them past 80% - performance drops off a cliff, like down to single digit MB/sec past about 85%

once you get past 90% full drobo will start to artificially limit the already extremely poor performance, to avoid you ever getting to the 100% mark

i see no reason why the 5D should be any different to the others i have owned, but i would be happy to be corrected?

My Avatar is my Panther Chameleon named 'Leon.

I have a owned a Drobo v1, v2, DroboPro, Drobo 5D, DroboFS, Drobo 5N and DroboPro FS

I currently have a Windows 2016 Server with 8 x 8TB Seagate Archive Drives, 14 x 8TB WD Red and 3 x 8TB WD Gold
Find all posts by this user
Quote this message in a reply
08-13-2013, 03:30 PM
Post: #3
RE: Questions about Drobo 5D performance
The benchmarks are very impressive. However, I don't think they take into account the modern concurrent nature of video use and editing.

A typical edit pulls multiple clips simultaneously, so it's not just the raw throughput, but rather how the system can perform for simultaneous file access.

LOL, just looked at your profile - we know each other from way back in my Canopus days. Smile

--Brandon | WHS2011+Drive Bender/2x Drobo v2/Drobo S G1/ Drobo S G2/Transporter
Drobo provides fault-tolerance, it's NOT a substitute for regular backups.
Drobo Best Practices - Official and Community-sourced.
Find all posts by this user
Quote this message in a reply
08-13-2013, 04:11 PM
Post: #4
RE: Questions about Drobo 5D performance
(08-13-2013 03:30 PM)bhiga Wrote:  A typical edit pulls multiple clips simultaneously, so it's not just the raw throughput, but rather how the system can perform for simultaneous file access.
Don't know if I agree - doc editing is mostly about stitching clips together with only one track playing at any time. Music videos (lots of fx) are often done in AE and then dynamic-linked to Pr, which means Premiere Pro only deals with one cached track at a time. Regardless, even with 4K R3D files with 60-120MB/s bitrates (well, byte-rates), these numbers are more than enough for 2-3 layered tracks. At least on paper. The issues I am seeing on the forums make me concerned about 5D's suitability for editing though.

(08-13-2013 03:30 PM)bhiga Wrote:  LOL, just looked at your profile - we know each other from way back in my Canopus days. Smile
Small world! Smile
Visit this user's website Find all posts by this user
Quote this message in a reply
08-13-2013, 04:16 PM
Post: #5
RE: Questions about Drobo 5D performance
" these numbers are more than enough for 2-3 layered tracks."

drobo's weakness has always been that simultaneous reads make the numbers tank

My Avatar is my Panther Chameleon named 'Leon.

I have a owned a Drobo v1, v2, DroboPro, Drobo 5D, DroboFS, Drobo 5N and DroboPro FS

I currently have a Windows 2016 Server with 8 x 8TB Seagate Archive Drives, 14 x 8TB WD Red and 3 x 8TB WD Gold
Find all posts by this user
Quote this message in a reply
08-13-2013, 04:32 PM
Post: #6
RE: Questions about Drobo 5D performance
(08-13-2013 04:16 PM)Docchris Wrote:  drobo's weakness has always been that simultaneous reads make the numbers tank

That's a bummer. If that's the case then most likely drobo's brain isn't fast enough to handle multiple requests, rather than it being a drive response issue. Or maybe it's a BeyondRAID issue, being too compute- or I/O-costly to provide decent performance. MS Storage Spaces technology, also a departure from standard RAID modes, has an absolutely horrific random I/O performance.

Did anyone benchmark 5D for 4K random I/Os with decent mechanical and perhaps SSD drives? That might just tell where the bottleneck is.
Visit this user's website Find all posts by this user
Quote this message in a reply
08-13-2013, 04:54 PM (This post was last modified: 08-13-2013 04:55 PM by rdo.)
Post: #7
RE: Questions about Drobo 5D performance
Performance tuning (and testing) is highly sensitive to details. It's possible to have two situations which differ by only some very minor thing (say, a streaming read vs. the same data stream with occasional background writes) that radically changes observed performance.

Accept benchmarks as representative measurements, but keep in mind that deviations from the benchmark conditions can cause performance deviations out of proportion to the change in conditions.

To invent an example, I could create a storage volume optimized for 8KB data blocks. If you're reading 10KB data chunks and writing 1KB checkpoint files, your 11KB total data requires 4 or 5 times the number of operations as reading 8KB. Most storage subsystems are limited by IOPS, so your MB/s numbers just got screwed.

In short: expect a significant difference between what the benchmark says and what you observe. If you don't have a large margin of safety, then get a product which you can return.

You earn respect by how you live, not by what you demand.
-unknown
Find all posts by this user
Quote this message in a reply
08-13-2013, 05:03 PM
Post: #8
RE: Questions about Drobo 5D performance
(08-13-2013 04:54 PM)rdo Wrote:  In short: expect a significant difference between what the benchmark says and what you observe.

Thanks rdo, that's a solid advice.
Visit this user's website Find all posts by this user
Quote this message in a reply
08-14-2013, 09:19 AM
Post: #9
RE: Questions about Drobo 5D performance
You have a point about doc editing... that's generally not very heavy on the concurrency.

Honestly, Drobo, even with the speedy 5D, is far from a magic bullet.
Vanilla RAID striping is much simpler to maintain and compute, and fragmentation latency is mitigated to a degree.

Drobo's BeyondRAID is more complex, and the "load" can be spread unevenly across different drives, especially since it is perfectly acceptable to mix drives of different makes and sizes.

Long story short, Drobo suitability for a specific use will depend largely on the specific configuration - use of specific drives and drive sizes.

As a starting point for testing, I would go with a number of WD Red drives and a couple of "typical case" projects to do real-world testing.

Important note: After setting up the Drobo, give it a day or two to "settle" - while immediately usable, it does some internal housekeeping and configuration in the background that affects the initial observed speeds.

--Brandon | WHS2011+Drive Bender/2x Drobo v2/Drobo S G1/ Drobo S G2/Transporter
Drobo provides fault-tolerance, it's NOT a substitute for regular backups.
Drobo Best Practices - Official and Community-sourced.
Find all posts by this user
Quote this message in a reply
08-14-2013, 03:44 PM (This post was last modified: 08-14-2013 03:59 PM by beans.)
Post: #10
RE: Questions about Drobo 5D performance
(08-14-2013 09:19 AM)bhiga Wrote:  Vanilla RAID striping is much simpler to maintain and compute, and fragmentation latency is mitigated to a degree.

Couldn't agree more - thanks Brandon. I got excited initially seeing great speeds, and after reading the forums and getting good responses it's easy to see that while 5D is good, it's not a replacement for vanilla RAIDs.

Thanks again.
Some questions had answers on the forums but not all of them...

- What are the speeds with USB 3.0? Seemingly up to 250MB/s although speeds of 150MB/s were more commonly reported.
- Does 5D support UAS? If not - any plans? Doesn't seem like it.
- Does adding an mSATA drive hurt sequential I/O speeds at all? Ditto, doesn't seem like it.
- Has anyone tested performance (sequential and random I/Os) when the 5D is 90% and 99% full? Yes and it drops like a hot potato.

Questions I couldn't find answers to:
- What are the speeds in dual parity mode?
- What's the reason Enterprise SSDs are not supported?
- What are rebuild times per TB in single and dual parity modes? (With drives like WD SE)
- Does 5D support "RAID bitmap" (as in Linux RAID) or something similar allowing near-instantaneous rebuild when a drive was temporarily offlined or re-inserted?

I am curious about "rebuild" times, how long does it take for Drobo to get back to fully "healthy" after replacing a failed drive. Also very curious what happens when you re-insert a drive. Take it out, wait a few minutes, re-insert. In most standard RAID systems, it's a full rebuild. In RAIDZ, MSS, Linux RAID, it's a near-instantaneous rebuild (a few seconds). How about Drobo?

Thanks!
Visit this user's website Find all posts by this user
Quote this message in a reply
Post Reply 


Forum Jump: