Considering the multiple queries about performance issues, I thought it would be worthwhile for all Drobo users to collect enough real-life statistics on Drobo actual performance.
Since Data Robotics recommends the free AJA System Test on Mac, it is what I used.
Note that in my case, the Read/Write graph shows very erratic performance, and the overall MB/s figures are about 1/6th (! ) of the ones for the ST9500420AS internal 500GB/7200RPM driveā¦
My figures also seem to be well below ābest caseā Data Robotics figuresā¦
--------Test Environment----------
Mac Mini (Late 2009) 2.66GHz Core 2 Duo 4GB
Mac OS X 10.6.4
Time Machine disabled
Drobo-V2 S/N : TDC08492****
Drobo<->Mac attachment : FW800 [no chaining]
Drobo FW : 1.3.6
Drobo DashBoard : 1.7.2
Drobo disks : 4 x 2TB WD WD20EADS
Drobo formatting : HFS+ / 8TB
Drobo usage : 62% (=3.37/5.44TB)
--------Test Utility----------
AJA KONA System Test 6.01 :
Test: Disk Read/Write
File Size: 1.0 GB
Video Frame Size: 1920x1080 10-bit
Disable file system cache
Round frame sizes to 4KB
File I/O API: Macintosh
--------Test Result [updated]----------
AJA Write [1.0 GB] : 15.8 MB/sec
AJA Read [1.0 GB] : 15.1 MB/sec
Interesting measurements. However, if we want to accumulate a large enough base of more or less comparable measurements, I believe we need a common tool (at most one on Mac and one on Windows), even if it is not a perfect one.
Unfortunately I do not expect many people to be prepared to slow down their Drobo for 10 hours, just for the sake of getting a few informative numbersā¦
Yes, DashBoard 1.7.2 is an advanced version (this one finally recognizes the Drobo after the Mac sleeps/wakes up).
Your numbers are impressive, could you give your test environment, and especially your measurement utility.
Hi geeji, Aja has numerous testing options which one are you using to perform your tests?
Also what is DRIās stance on using an unreleased version of the dashboard? Iām guessing they wonāt support anu issuers you have while running it since it hasnāt been publicly released?
[quote=āDolsJames, post:6, topic:1597ā]
Hi geeji, Aja has numerous testing options which one are you using to perform your tests?[/quote]
AJA KONA default options :
Test: Disk Read/Write
File Size: 128MB
Video Frame Size: 720x486 8-bit
Disable file system cache
Round frame sizes to 4KB
File I/O API: Macintosh
You may be right, but I have suffered the sleep/wake issue for years, and that release was the first one to correct it, soā¦[hr]
I have redone my own tests with exactly your same parameters, and also averaged 3 of them. The results are higher than with the AJA KONA default options, but still about 1/3 of yours :
Write: 10.1
Read: 13.2
Looking at the results graph, it appears that my peak results are close to yours (at the beginning of the test), but they are very unstable and cannot be sustained the whole test. At that point, I have no idea of why this discrepancy.
What Mac OS X version do you use ?
How much RAM do you have ?
How is your Drobo formatted ?
What is your Drobo storage usage ?
iMac i7, 8GB RAM, 10.6.4
FW800 (only device)
Drobo 1.6.8, 1.3.6
2x 2TB WD Caviar Black
1x 1TB WD GP
When I first ran Kona (within 24 hours of my initial setup) my performance was like yours. However, after letting my unit āsitā for a couple of days, I have consistently seen the speeds I posted above.
I thought of that too, but unfortunately, my Drobo has been set-up and filled months ago, so it is something else.
[quote=āmgriffin34, post:10, topic:1597ā]
iMac i7, 8GB RAM, 10.6.4
FW800 (only device)
Drobo 1.6.8, 1.3.6
2x 2TB WD Caviar Black
1x 1TB WD GP[/quote]
The only differences I can imagine as significant are the 8GB RAM (vs 4GB) and the disks. Hard to believe however that it generates such a large instabilityā¦
The Drobo firmware is the same, and the Drobo Dashboard is not supposed to have any real-time effect (except for alerts).
BTW, Time Machine was disabled, so it is not that either.
Following bhiga suggestion, I tried 3 different FW800 cables (including a 1" short one) as well as 1 USB2 connection : no improvement, although USB2 gave a slightly higher average throughput than FW800, which is counterintuitive.
The USB2 showed the same kind of instability I had with the FW800 connection.
I even tried to disconnect the Drobo ground connection (in case it was a ground loop problem between the Drobo and the Mac Mini) : no improvement.
I also checked the Drobo Log : no disk I/O error.
The only benefit from all those experiments was related to the first Drobo restart, which after a few minutes of activity (Drobo internal fsck ?) improved somewhat my transfer values, which are now, using FW800 and averaged over 3 runs, up to :
Write: 15.8 MB/sec
Read: 15.1 MB/sec
But after all those experiments, my results are still highly unstable, and less than half of mgriffin34 ones .
More interestingly, the instabilities vary from one run of AJA System Test to the next (as shown by the attached graphs). The close correlation between the Read and Write graphs suggest however that those instabilities occur WITHIN the Drobo itself and are NOT FW800 or USB2 random connections instabilities.
=> @Jennifer : have you any suggestion ? have you seen anything like it ?
Although my Drobo-V2 āworksā, I seem to be loosing half of its bandwidth (assuming AJA System Test is not bugged), which is definitely NOT something I want
--------Test Utility----------
AJA KONA System Test 6.01 :
Test: Disk Read/Write
File Size: 1.0 GB
Video Frame Size: 1920x1080 10-bit
Disable file system cache
Round frame sizes to 4KB
File I/O API: Unix
--------Test Environment 1----------
Mac Mini (Late 2009) 2.53GHz C2Duo 4GB OSX 10.6.4
MTU: Off
Drobo<->Mac attachment: AEBS (2007) --------Test Environment 2----------
Macbook Pro (2,2) 2,33ghz C2Duo 4GB OSX 10.6.4
MTU: Off
Drobo<->Mac attachment : GBE Direct --------Test Environment 3----------
Macbook Pro (2,2) 2,33ghz C2Duo 4GB OSX 10.6.4
MTU: Off
Drobo<->Mac attachment : Dlink GB Switch --------Test Environment 4----------
Mac Mini (Late 2009) 2.53GHz C2Duo 4GB OSX 10.6.4
MTU: Off
Drobo<->Mac attachment : Dlink GB Switch --------Test Environment 5----------
Macbook Pro (2,2) 2,33ghz C2Duo 4GB OSX 10.6.4
MTU: 9000
Drobo<->Mac attachment : Dlink GB Switch + AEBS --------Test Environment 6----------
Mac Mini (Late 2009) 2.53GHz C2Duo 4GB OSX 10.6.4
MTU: 9000
Drobo<->Mac attachment : Dlink GB Switch + AEBS --------Test Environment 7----------
Macbook Pro (2,2) 2,33ghz C2Duo 4GB OSX 10.6.4
MTU: 9000
Drobo<->Mac attachment : Dlink GB Switch --------Test Environment 8----------
Macbook Pro (2,2) 2,33ghz C2Duo 4GB OSX 10.6.4
MTU: 9000
Drobo<->Mac attachment : GBE Direct --------Test Environment 9----------
Mac Mini (Late 2009) 2.53GHz C2Duo 4GB OSX 10.6.4
MTU: Off
Drobo<->Mac attachment : GBE Direct --------Test Environment 10----------
Mac Mini (Late 2009) 2.53GHz C2Duo 4GB OSX 10.6.4
MTU: 9000
Drobo<->Mac attachment : GBE Direct
--------Test Result 1----------
AJA Write [1.0 GB] : 22 MB/sec
AJA Read [1.0 GB] : 24 MB/sec --------Test Result 2----------[color=#6B8E23]Updated[/color]
AJA Write [1.0 GB] : 52 [color=#6B8E23]24[/color] MB/sec
AJA Read [1.0 GB] : 52 [color=#6B8E23]27[/color] MB/sec --------Test Result 3----------
AJA Write [1.0 GB] : 25 MB/sec
AJA Read [1.0 GB] : 27 MB/sec --------Test Result 4----------
AJA Write [1.0 GB] : 23 MB/sec
AJA Read [1.0 GB] : 27 MB/sec --------Test Result 5----------
AJA Write [1.0 GB] : 28 MB/sec
AJA Read [1.0 GB] : 29 MB/sec --------Test Result 6----------
AJA Write [1.0 GB] : 27 MB/sec
AJA Read [1.0 GB] : 29 MB/sec --------Test Result 7----------
AJA Write [1.0 GB] : 25 MB/sec
AJA Read [1.0 GB] : 27 MB/sec --------Test Result 8----------
AJA Write [1.0 GB] : 26 MB/sec
AJA Read [1.0 GB] : 24 MB/sec --------Test Result 9----------
AJA Write [1.0 GB] : 21 MB/sec
AJA Read [1.0 GB] : 25 MB/sec --------Test Result 10----------
AJA Write [1.0 GB] : 26 MB/sec
AJA Read [1.0 GB] : 27 MB/sec
@DroboLars : Very interesting : your Drobo-FS when Ethernet directly attached (Test Result 2) is 3 times faster than my Drobo-V2 FW800 attached.
Lucky for you but very suspect for meā¦
Yup, also Interesting that AEBS slows it down over 50%ā¦
I wonder if the AEBS (2009) is fasterā¦?
Notice that I have File I/O API: Unix. I think DataRobics recomended this when I tested my first Drobo-FS (wich was replaced due to slow performance and noise).
@geeji: Couldnāt see your image, so I canāt tell how much of a difference you saw between USB2 and Firewire. Itās possible that thereās something causing the system to run at FW400 speed instead of FW800, but your performance numbers donāt saturate either bus, so I think you are correct - the problem lies elsewhere.
Have you submitted your diagnostic log to DRI support? Itās possible that one of your drives is ālaggingā enough to cause performance degradation but not enough to flag it bad (yet).
[quote=āDroboLars, post:16, topic:1597ā]Yup, also Interesting that AEBS slows it down over 50%ā¦
I wonder if the AEBS (2009) is fasterā¦?
Notice that I have File I/O API: Unix. I think DataRobics recommended this when I tested my first Drobo-FS (which was replaced due to slow performance and noise).[/quote]
But what is the benefit of Drobo-FS connection to AEBS instead of direct Ethernet connection? WiFi access ? Multiple users sharing while your Mac is off ?
In the past there has been multiple reports of poor AEBS performance, but that was when connecting a Drobo through USB to it, and then using the Airport Disk feature (instead of DroboShare, which is not an option anymore since it has been discontinued).
Sorry about the images, for whatever reason they sometimes would not display; I changed the images hosting server, seems to work better now.
And no, I have not submitted anything to DRI support yet, I was hoping to get some feedback from Jennifer before.
At that point, my guess is that the 1GB file created by AJA System Test is never created at the same place in the Drobo file structure, and may be because of some internal fragmentation due to past activity (??), the access time varies a lot frame by frame.
If that is true, this slow down has little to do with actual disks performance, but it is also very worrying since it would mean that the Drobo file structure āwears downā.
On the other hand, till we get more results similar to mine, it could be a specific bug in my Droboā¦
@mgriffin34 + @DroboLars : if you could publish the graph for your results, it would help me to understand if my instabilities are peculiar to my configuration or not. Your graphs should be more regular than mine, since my instabilities seem to be the main cause of my poor results.
Oh sorry, was not following this thread as a thread for issues with Drobos. If you experiencing performance issues then definitely open a support case. Especially if you have tried different cables/ports/computers and can reproduce the behavior and the behavior follows the drobo.
[quote]But what is the benefit of Drobo-FS connection to AEBS instead of direct Ethernet connection? WiFi access ? Multiple users sharing while your Mac is off ?
In the past there has been multiple reports of poor AEBS performance, but that was when connecting a Drobo through USB to it, and then using the Airport Disk feature (instead of DroboShare, which is not an option anymore since it has been discontinued).[/quote]
Drobo-FS = Drobo-FileSharing and, yes that is my purpose.
Right now it is placed right beside my Mac Mini wich is the computer that will mainly use the Drobo so it could easily be connected but I do plan on putting it somewhere away from hearing. Since the Mac Mini only has one GBE port and I also wont it wired for internet and file-sharing its not an option to directly connect the Drobo. I also like to be able too access my media after Mac mini bed-time (set to sleep).
In the everyday 20-something mbs is fast enough for streaming 1080p and that is the main purpose so the concern is not that big, but when you know you can more than double the speedā¦
) of the ones for the ST9500420AS internal 500GB/7200RPM driveā¦
