Post Reply 
 
Thread Rating:
  • 1 Votes - 5 Average
  • 1
  • 2
  • 3
  • 4
  • 5
read/write performance disparity
05-08-2013, 02:33 PM
Post: #1
read/write performance disparity
I'm forking a previous thread so as not to hijack the original poster's question. This is mostly an FYI in case anyone else is experiencing similar issues. I'm opening a ticket with Drobo support simultaneously and will report back with the results of that process.

My Drobo 5D shows slow read performance relative to write performance while connected via Thunderbolt. Blackmagic's Disk Speed Test shows roughly 165MB/s write speed and roughly 40MB/s read speed.

This is similar to my own "real world" testing, which averages to 159MB/s write and 58MB/s read via Thunderbolt. This test consisted of copying 4GB of jpegs directly to/from a RAM disk using Finder. I repeated the test twice, each time using a different Thunderbolt cable.

Firmware is 3.0.6. Dashboard and OS X (10.7) are up to date. The Drobo was initially configured with two 3TB WD Red drives. Three more drives (two 640GB and one 750GB) were added later.
Find all posts by this user
Quote this message in a reply
05-08-2013, 04:49 PM
Post: #2
RE: read/write performance disparity
"4GB of jpegs"

Sorry to say, but in terms of raw throughput that's not going to be a reliable test. It may be real world, but it's not going to be very consistent or repeatable. You have innumerable tiny reads/writes, which is something of a worst-case scenario for performance and it's highly variable based on a huge number of factors, even down to how the directory trees were built up as you added the files. If you want a reliable performance number, you really have to use large (multi-GB) single files. Otherwise you have who knows how many different disk accesses between files and directory data (and on networked models, the network abstraction layer like netatalkd and samba).

Drobo 5N | 250GB Samsung 850 EVO mSATA | 2 x 4TB Seagate, 3 x 4TB Hitachi | FS/EXT3 diskpack
Peak performance >100MBps read/write (based on FS disk pack, no jumbo frames, no SSD)
DroboPorts: Plex, Transmission, OpenSSH, NFS, nano, screen, bash
Find all posts by this user
Quote this message in a reply
05-08-2013, 09:57 PM (This post was last modified: 05-08-2013 10:00 PM by mgriffin34.)
Post: #3
RE: read/write performance disparity
He did say he used Blackmagic in addition to his 'real world' test.

BTW- on my 65% full 5D, my read and write speeds are within 10% of each other. About 140MB/s. (the 5D was busy with playing a movie to my laptop and receiving a MakeMKV file from my iMac)

This is using Blackmagic with 1GB and 4GB sizes.

I have it attached to a Mini (10.8.3).

i7 Mac Mini Server, 10.10
Drobo (Gen2), USB, 4x WD 4TB RED

Drobo (Gen3), USB3, 4X WD 4TB RED

Drobo 5D, TB, 5x WD 4TB RED, 120GB Intel 525 mSATA SSD
Find all posts by this user
Quote this message in a reply
05-09-2013, 07:24 AM (This post was last modified: 05-09-2013 07:30 AM by mestanczak.)
Post: #4
RE: read/write performance disparity
(05-08-2013 04:49 PM)diamondsw Wrote:  If you want a reliable performance number, you really have to use large (multi-GB) single files.

Hi diamondsw. Thanks for the reply. The Blackmagic test should be a good indicator of "raw" performance. The point of the JPEG test was to simulate my workflow as a photographer. In any case, both scenarios result in similar throughput.

I don't often deal with large chunks of data, like disc images or video files, but my experience reading a couple of ~1GB zip files points to a similar ~40MB/s throughput. This is slower than my 5400rpm bus powered LaCie rugged. It seems there is a bottleneck somewhere.
(05-08-2013 09:57 PM)mgriffin34 Wrote:  BTW- on my 65% full 5D, my read and write speeds are within 10% of each other. About 140MB/s.

Hmmm... that's closer to what I expected in terms of performance. Good to know.
Find all posts by this user
Quote this message in a reply
05-12-2013, 01:53 PM
Post: #5
RE: read/write performance disparity
hi mestanczak,
if the source data is "safe" and if youre not connected to the web, then one thing ive often noticed is that some antivirus settings can actually cause any file read, to be scanned too, and when antivirus is disabled (or the feature on every read action disabled), it really speeds things up.

(eg if i run a backup, or crc checks on safe data and offline, it goes quite quicker without the virus scan) - maybe you can check your settings for virus scans/indexing etc?

for the 4gig of images, i think what youve done is a good real life test (at least for the moment in question), because its often the "real world" use where we experience things most.
- loose files and folders sure are a perfect test, (as long as your tests are repeatable with similar types of tests) then loose files (and single large files) are good to compare.

normally though, i would have expected "writes" and "moves" to the files to be the slowest, especially on external drives, and would have thought "reads" to be the quickest, so im not quite sure why your "reads" would be slower..

i would suggest you repeat all the same tests again, but try with usb cable to compare (relatively) the percentages of difference between writing and reading and post back some results for us Smile

(btw i have XP home SP2, a Drobo v1 with 2x 1TB/2x 1.5TB WD greens, & a bkp Drobo v2 with the same + a DroboShare: unused)
& a DroboS v2 with 3xWD15EADS &2x1TB in DDR mode on win7, & a drobo5D (all usb)
  • btw i did a sustained (write) operation for about 6 hours, and got 13.2MB / sec ...objection? "sustained" :)
    (16.7MB/s on a v2 & 47-96MB/s drobo-s)
Find all posts by this user
Quote this message in a reply
05-13-2013, 11:46 AM (This post was last modified: 05-13-2013 11:49 AM by mestanczak.)
Post: #6
RE: read/write performance disparity
(05-12-2013 01:53 PM)Paul Wrote:  hi mestanczak,
maybe you can check your settings for virus scans/indexing etc?

i would suggest you repeat all the same tests again, but try with usb cable to compare (relatively) the percentages of difference between writing and reading and post back some results for us Smile

Hi Paul,

Thanks for your helpful suggestions. I'm working with support now and have sent them some diagnostic tests after following a specific speed test procedure. Just waiting to hear back from them.

The antivirus issue is relevant, as my workplace requires us to run Symantec on our Macs. Luckily my permissions are such that I'm able to disable auto scanning during these test. Perhaps there's another element of the AV suite that's still lingering in the background. I'll look into that.

Regarding USB, unfortunately I don't have access to a USB3 system. Over USB2 I'm seeing reads & writes top out around 30MB/s, with reads around 10% faster.

Today after running the test procedure for Drobo support I noticed Thunderbolt read speed occasionally spiking higher than before. Crossing my fingers support narrows these issues to a single drive that can be pulled to resolve the problem.
Find all posts by this user
Quote this message in a reply
05-13-2013, 12:11 PM
Post: #7
RE: read/write performance disparity
ok thanks for the update so far mestan

(btw i have XP home SP2, a Drobo v1 with 2x 1TB/2x 1.5TB WD greens, & a bkp Drobo v2 with the same + a DroboShare: unused)
& a DroboS v2 with 3xWD15EADS &2x1TB in DDR mode on win7, & a drobo5D (all usb)
  • btw i did a sustained (write) operation for about 6 hours, and got 13.2MB / sec ...objection? "sustained" :)
    (16.7MB/s on a v2 & 47-96MB/s drobo-s)
Find all posts by this user
Quote this message in a reply
05-20-2013, 11:10 AM
Post: #8
RE: read/write performance disparity
The saga continues. I gained access to a retina Macbook Pro with USB3 and Thunderbolt and spoke with a support representative for some time on Friday. We discussed the typical performance that should be expected from the 5D and what factors affect performance. The representative noted that spreading data across more drives will increase maximum throughput. Given that this 5D contains 5 7200rpm drives, we should be seeing high sustained throughput (150+ MB/s). We do see spikes in that range, but read performance even for large 4GB zip files is not consistent.

Curiously, while reading back identical duplicate 4GB zip files I notice a specific performance pattern for each file. This pattern is the same regardless of transfer method (usb3, usb2, thunderbolt). For example, zip file #1 is read back quite quickly for the first 25% of the file. Thunderbolt ~220MB/s, USB3 ~150MB/s, USB2 ~35MB/s. The remaining 75% of the file is transferred quite slowly. Thunderbolt ~60MB/s, USB3 ~60MB/s, and USB2 ~3MB/s.

Yes, 3MB/s. It seems to me this points to some sort of basic problem with my setup. File fragmentation, drive errors, Drobo's performance algorithms, or I don't know what. Another identical 4GB zip file reads back with a different, yet consistent, performance pattern. It transfers a bit more quickly than the first zip file, yet sustained read speeds are still not as good as a fast internal 7200RPM HDD.

Support's suggestion was to install an mSATA drive in this Drobo. I am skeptical that this will improve read performance in my situation. Regardless of the above performance issues, the way I work I need to leave files sitting untouched for 3 months and then pull all of them at once in one operation. This situation seemingly won't give the 5D an opportunity to promote that data to SSD storage before it is needed. But maybe I'm wrong in that assumption.

My next steps are as follows:

1) Remove the oldest three drives individually, allowing Drobo to rebuild each time. If sustained read speeds become more consistent this would point to an issue with the older/smaller drives.
2) If step one doesn't work, add back all 3 older drives and install an mSATA card.
3) If step two doesn't work, return the Drobo 5D and replace it with something like a couple of LaCie 2big drives.
Find all posts by this user
Quote this message in a reply
05-20-2013, 11:29 AM
Post: #9
RE: read/write performance disparity
Given your file usage pattern, I doubt the mSATA will help you.

I recommend adding two steps to your proposal:
0) Submit diagnostic logs to Drobo, see if they notice anything out of the ordinary
and
1a) Try to get Drobo to swap out your chassis - this will eliminate some kind hardware problem with the Drobo itself

--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
05-21-2013, 07:24 AM
Post: #10
RE: read/write performance disparity
(05-20-2013 11:29 AM)bhiga Wrote:  Given your file usage pattern, I doubt the mSATA will help you.

I recommend adding two steps to your proposal:
0) Submit diagnostic logs to Drobo, see if they notice anything out of the ordinary
and
1a) Try to get Drobo to swap out your chassis - this will eliminate some kind hardware problem with the Drobo itself

Thanks, bhiga. I agree about the mSATA. Drobo support has an open ticket for this issue and several diagnostic files they've supposedly already looked through. I'll press them again before ordering any new hardware.
Find all posts by this user
Quote this message in a reply
Post Reply 


Forum Jump: