Drobo

From Finninday
Revision as of 21:54, 31 December 2013 by Rday (Talk | contribs)

Jump to: navigation, search

The drobo showed up yesterday. The unboxing was as cool as with an Apple product. And it is really easy to setup and start using... as long as you are plugging it into a Mac or PC. The instructions don't say anything about how to get it to work on Linux. Luckily I found this via google:

[root@192.168.1.1]# lshw
[root@192.168.1.1]# /sbin/mke2fs -j -i 262144 -L Drobo -m 0 -O sparse_super,^resize_inode -q /dev/sdc
[root@192.168.1.1]# mkdir /drobo
[root@192.168.1.1]# mount -t ext3 /dev/sdc /drobo
update# ls -l /dev/disk/by-uuid
(copy the uuid for insertion into fstab)
[root@192.168.1.1]# vi /etc/fstab
 UUID=7fcc72fe-0884-4d66-b4f3-962901875650   /drobo  ext3    defaults     0 0


That was all I needed to get me going. I thought I had a bunch of old hard drives lying around that I would be able to wedge into drobo until I could afford to buy real drives, but none of them are SATA. Darn. Another bummer is that my server has a firewire plug that is heart-shaped and resembles a mac firewire, but the firewire plug on drobo is square. So I think I'm stuck with USB. My first speed test writing to drobo shows 14.8MB/sec. That feels pretty lame.

Another weird thing is that the drobo shows the full capacity of the array, not the actual capacity. That is, I put two 1TB drives in it and it looks like this with df -h:

/dev/sde         2.0T   155G   1.9T   8%   /drobo

I expected drobo to report the capacity of the array to be about 1TB instead of 2TB.

I added a few more drives to the drobo to give it the capacity to hold our vacation videos.

Interestingly, after adding two 1TB drives, drobo still reports the same capacity to df. Even more reason not to trust df numbers for drobo.

drobo performance on usb 2.0

But still the drobo is just too slow.

Drobo claims to have these throughput stats:

Max Sustained Transfer Rate:
FireWire 800: Up to 52MB/s reads and 34MB/s writes
USB 2.0: Up to 30MB/s reads and 24MB/s writes

How slow is it for me? Bonnie says this:

rday@weasel:/drobo/bonnie$ bonnie
Writing a byte at a time...done
Writing intelligently...done
Rewriting...done
Reading a byte at a time...done
Reading intelligently...done
start 'em...done...done...done...done...done...
Create files in sequential order...done.
Stat files in sequential order...done.
Delete files in sequential order...done.
Create files in random order...done.
Stat files in random order...done.
Delete files in random order...done.
Version  1.96       ------Sequential Output------ --Sequential Input- --Random-
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
weasel           6G   249  74  1019   0   500   0   876  63  1111   0  58.3   2
Latency              1516ms   21811ms    1552ms   76164us     261ms     727ms
Version  1.96       ------Sequential Create------ --------Random Create--------
weasel              -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16  1387   6 +++++ +++  1815   4  1615   4 +++++ +++  1811   4
Latency             63932us    1261us     725us    2761us    2696us     121us

I care most about block reads and writes. Those come out at 1,111 and 1,019 K/sec.

Hmm, what units is bonnie reporting? I'll guess bytes since bits would be silly. So I'm getting 1MB/sec on drobo. Bleck. That's a lot worse than the 24MB/sec that drobo advertises. I wonder what I'm doing wrong? I'm getting slightly faster reads than writes, which seems reasonable.

native internal disk performance on the same machine

Compare that to bonnie running on an internal drive on the same machine:

Version  1.96       ------Sequential Output------ --Sequential Input- --Random-
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
weasel           6G   347  98 33362  24 17006  10  1311  95 46774  13 115.1   7
Latency             83296us    2552ms    1958ms   97333us     412ms    1399ms
Version  1.96       ------Sequential Create------ --------Random Create--------
weasel              -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16  2238  86 +++++ +++ +++++ +++  2691  97 +++++ +++  6854  97
Latency             35917us     410us     319us   53088us     119us     874us

That means block reads and writes at 46 and 33 MB/sec. About 30 times better. And curiously close to what drobo claims to be able to do. So I have to presume that the "block sequential read" that bonnie is measuring is a different metric than the "max sustained transfer rate" that drobo is measuring.

But if I just need to find the right firewire cable to switch off of USB, then I'm a dummy for using USB all this time.

Wikipedia says that USB 2.0 generally performs at 240Mbit/sec and 1394 does 800Mbit/sec with less CPU load. I've got to find that cable.

Hopefully, I'll be able to post bonnie stats for drobo over firewire soon.

drobo over firewire 400 (1394a)

Version  1.96       ------Sequential Output------ --Sequential Input- --Random-
Concurrency   1     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
weasel           6G   265  98 16951  14 10546   6   961  97 28571   8 123.6   7
Latency               110ms    6345ms     537ms   67886us     230ms     796ms
Version  1.96       ------Sequential Create------ --------Random Create--------
weasel              -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16  8777  28 +++++ +++ 16891  46 14137  43 +++++ +++ 15867  47
Latency             87774us   12285us   13658us   13962us   11168us   13427us

So let's see, that means 17MB/sec block writes and 28MB/sec block reads. Wow, that's much better. And if I wanted to invest in a $25 1394b card, I could probably double that. Well, I'd have to buy another $37 cable too. So maybe I'll save up for awhile.

Watching movies now seems much better. I only saw one stutter on a dvd served by the drobo and that was while the cache was still filling right after starting VLC. Now I just need to watch more movies and gather much more data.

drobo and the 2TB limit

I only just noticed that my second generation regular drobo has a 2TB limit on the LUNSIZE and that when I filled all four bays with 1TB drives, it didn't just give me (4TB - raid overhead) usable storage space. It gave me 2 x (2TB - raid overhead) of usable storage space. That is, I now have two devices, not one big pot to put all my files in. Bummer. And according to this, it is not advisable to use LVM to wrap the devices up into a single logical unit. I find that hard to believe, but I'm not immediately prepared to test my theory.

Maybe I can live with two medium-sized buckets for my data.

iostat while copying from one side of drobo to the other

Device:            tps   Blk_read/s   Blk_wrtn/s   Blk_read   Blk_wrtn
sda               2.00        20.80        22.40        104        112
dm-0              5.40        20.80        22.40        104        112
dm-1              0.00         0.00         0.00          0          0
sdb               0.00         0.00         0.00          0          0
sdc               0.00         0.00         0.00          0          0
dm-2              0.00         0.00         0.00          0          0
dm-3              0.00         0.00         0.00          0          0
dm-4              0.00         0.00         0.00          0          0
sdd              56.80     13633.60         0.00      68168          0
sde              18.80         0.00      9593.60          0      47968

Device:         rrqm/s   wrqm/s     r/s     w/s   rsec/s   wsec/s avgrq-sz avgqu-sz   await  svctm  %util
sda               1.80     6.00    2.40    1.00    33.60    56.00    26.35     0.05   14.71   4.71   1.60
dm-0              0.00     0.00    4.20    7.00    33.60    56.00     8.00     0.09    7.68   1.43   1.60
dm-1              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
sdb               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
sdc               0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
dm-2              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
dm-3              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
dm-4              0.00     0.00    0.00    0.00     0.00     0.00     0.00     0.00    0.00   0.00   0.00
sdd               2.40     0.60   79.80    1.40 19126.40    16.00   235.74     2.00   24.70  12.29  99.80
sde               0.00  1633.20    0.00   13.80     0.00 13176.00   954.78    13.92 1008.41  30.00  41.40

The near 100% utilization of sdd tells me that reads are the bottleneck. And the number of blocks that have been read in a 5-second interval was 13633.60 with a standard block size of 4KB means that drobo is reading at a rate of 55.84MB in 5 seconds or 11.17MB/sec. That's a bummer when you are trying to move several hundred gigabytes.

looking for a good test

scp read from drobo, write to /tmp

rday@weasel:/drobo/dvds$ scp Zombieland.iso localhost:/tmp
The authenticity of host 'localhost (127.0.0.1)' can't be established.
RSA key fingerprint is be:7f:1e:a6:2a:e3:01:1e:5f:25:2d:9c:6a:0f:ef:10.
Are you sure you want to continue connecting (yes/no)? yes
Warning: Permanently added 'localhost' (RSA) to the list of known hosts.
Enter passphrase for key '/home/rday/.ssh/id_dsa': 
Zombieland.iso                                    66% 5331MB  22.6MB/s   01:59 ETA

timed cp from drobo to /tmp

rday@weasel:/drobo/dvds$ /usr/bin/time -v cp Zombieland.iso /tmp
	Command being timed: "cp Zombieland.iso /tmp"
	User time (seconds): 0.11
	System time (seconds): 81.54
	Percent of CPU this job got: 23%
	Elapsed (wall clock) time (h:mm:ss or m:ss): 5:51.06
	Average shared text size (kbytes): 0
	Average unshared data size (kbytes): 0
	Average stack size (kbytes): 0
	Average total size (kbytes): 0
	Maximum resident set size (kbytes): 992
	Average resident set size (kbytes): 0
	Major (requiring I/O) page faults: 1
	Minor (reclaiming a frame) page faults: 313
	Voluntary context switches: 36004
	Involuntary context switches: 2110
	Swaps: 0
	File system inputs: 16434904
	File system outputs: 16439552
	Socket messages sent: 0
	Socket messages received: 0
	Signals delivered: 0
	Page size (bytes): 4096
	Exit status: 0
Total bytes of this file: 8417050624
Total seconds: 351.06
Bytes/sec: 23,976,102

timed cp from drobo to /dev/null

I guess the write to a spinning disk still has some cost and is not completely parallelized with the read.

rday@weasel:/drobo/dvds$ /usr/bin/time -v cp Zombieland.iso /dev/null
	Command being timed: "cp Zombieland.iso /dev/null"
	User time (seconds): 0.10
	System time (seconds): 17.98
	Percent of CPU this job got: 5%
	Elapsed (wall clock) time (h:mm:ss or m:ss): 5:02.58
	Average shared text size (kbytes): 0
	Average unshared data size (kbytes): 0
	Average stack size (kbytes): 0
	Average total size (kbytes): 0
	Maximum resident set size (kbytes): 992
	Average resident set size (kbytes): 0
	Major (requiring I/O) page faults: 1
	Minor (reclaiming a frame) page faults: 313
	Voluntary context switches: 34180
	Involuntary context switches: 292
	Swaps: 0
	File system inputs: 16434432
	File system outputs: 0
	Socket messages sent: 0
	Socket messages received: 0
	Signals delivered: 0
	Page size (bytes): 4096
	Exit status: 0
Total bytes of this file: 8417050624
Total seconds: 302.58
Bytes/sec: 27,817,604

dd from /dev/zero to drobo

rday@weasel:/drobo/dvds$ /usr/bin/time -v dd if=/dev/zero of=./test bs=16k count=163840
163840+0 records in
163840+0 records out
2684354560 bytes (2.7 GB) copied, 213.715 s, 12.6 MB/s
	Command being timed: "dd if=/dev/zero of=./test bs=16k count=163840"
	User time (seconds): 0.08
	System time (seconds): 17.57
	Percent of CPU this job got: 8%
	Elapsed (wall clock) time (h:mm:ss or m:ss): 3:33.88
	Average shared text size (kbytes): 0
	Average unshared data size (kbytes): 0
	Average stack size (kbytes): 0
	Average total size (kbytes): 0
	Maximum resident set size (kbytes): 924
	Average resident set size (kbytes): 0
	Major (requiring I/O) page faults: 2
	Minor (reclaiming a frame) page faults: 289
	Voluntary context switches: 12677
	Involuntary context switches: 243
	Swaps: 0
	File system inputs: 192
	File system outputs: 5242880
	Socket messages sent: 0
	Socket messages received: 0
	Signals delivered: 0
	Page size (bytes): 4096
	Exit status: 0

drobo drives

Drobo contains four bays, numbered from the top.

drive barcode model
drive 1 5VX00PHA ST31000520AS
drive 2 5VX015VB ST31000520AS
drive 3 9VP8QLAS ST31000528AS
drive 4 9VP8Q9KL ST31000528AS

drobom info