Because I never actually get to use the massive storage in the DroboFS (because it's so slow it is unusable), I got the biggest internal SSD drive available for the Air. Now that space is running low, so I am ready to fight with the Drobo again to try to get some value out of it.
A fairly nice mechanical (as opposed to SSD) SATA hard drive may be able to read/write at 100MBs (megabytes per second). To put that into network throughput terms, which are measured in megabits, multply by 8 because there are 8 bits in a byte. So you need at least 800mbps of network bandwidth to work a decent SATA hard drive to its limit. I'm parenthesizing megabits per second (mbps) and megabytes per second (MBs) throughout this post to relate network speed to hard drive speed (measured in MBs), since the point is answering why Network Attached Storage (NAS) like a DroboFS is slow on a slow wireless network.
Recent computers with a "SATA II" bus are capable of moving data at 3 gigabits per second (3000mbps, about 6 times faster than USB speed). Machines made after 2011 probably are "SATA III," and can drive 6 gigabits per second (6000mbps). A "SATA III" machine like my 2011 Macbook Air can work a hard drive up to 750MBs. The fastest SSD drives available right now go only 500MBs, 5 times faster than a decent mechanical drive. Is it starting to be clear that if your network has only 54mbps (54/8 = 6MBs) of bandwidth, there is no way you can take advantage of what even a crappy hard drive can do? That is my problem. Disk access to the DroboFS is only like 5MBs = unusable.
In the DroboFS are a bunch of 2TB Western Digital WD20EARS energy saving "SATA II" hard drives. The lowest benchmarks on these individual drives are around 80MBs. When added to the Drobo, though, they become part of its disk array, and how fast is that? I need to measure from the DroboFS itself, not from my Mac, because I want an answer that does not include any slowdown caused by accessing the device over a network. I just want to know what the Drobo is capable of, and then try to see how close I can get to that when accessing the device as a NAS.
So I get a root shell on the Drobo (DroboApps Dropbear) and ask how long does it take to write a 2,048 megabyte file ("8" * 1024 byte blocks, written "256" * 1024 times)
# time sh -c "dd if=/dev/zero of=/mnt/DroboFS/output.img bs=8k count=256k && sync"
real 0m 54.97s
Now I flush out any filesystem cache that might exist in the Drobo's memory by writing another file that is at least as large as the amount of Drobo's memory, so that when we read the first file back in, it will be a full read, with no "cheating" from the use of any cache. This Drobo hunk of junk has only 128MB of RAM, so 16000 * 8000 should wipe anything.
# dd if=/dev/zero of=/mnt/DroboFS/flush.img bs=8k count=16k
After any caches are flushed, how fast can it read the first file we wrote?
# time dd if=/mnt/DroboFS/output.img of=/dev/null bs=8k
real 0m 33.70s
Thanks to this blog for the commands, info about sync time and filesystem caches.
RESULTS:
write: 37MBs (2048/54.97)
read: 60MBs (2048/33.7)
Now I see that the maximum potential disk speed on the DroboFS is only 40-60% of what the bare drives have benchmarked when connected directly to a computer with a SATA cable. Even though I know from internet chatter that the processor and memory components of a DroboFS are cheap and under-powered, this performance loss is still a surprise. RAIDed drives add spindles, which should increase performance. But the Drobo is not RAID, it is a proprietary "Beyond RAID." Whatever.
The true performance that I will see in real world use will actually be even less than the result from the test above because those tests are done with "dd" which doesn't really consider the overhead of filesystem format. So, real world maximum potential will be worse than those figures, let's say 30MBs, which is what most people who are giving the DroboFS favorable ratings say that it can do.
Botom line: You take a 80MBs drive, put it in a DroboFS, and now it is a 30MBs drive. Boo.
30MB/s should still be usable though for Time Machine and other backups, iTunes and iPhoto Libraries, which are the things that are taking up all the space that pushed me to external storage in the first place. Except that when I access the NAS over wifi, I'm getting only like 5MBs.
In my last post, I said I thought the slowness was due to the network, related to my Airport Extreme Base Station being from 2008 when the 802.11n spec (the fastest wifi) was still in draft. Looking at my wifi connection settings on my Air, I was always seeing slow transmit rates like "30" or "54" (megabits per second), which, divided by 8, works out to those slow MBs disk speeds on the NAS.
So I bought a new Time Capsule (instead of a cheaper diskless Airport Extreme, hedging against getting rid of the stupid Drobo at some future time) to see if that would boost my wifi speed from 50mpbs to something closer to the maximum potential of 802.11n wifi, which is 300mbps (37MBs).
I also bought an Apple USB-to-Ethernet adapter made for the Macbook Air without realizing that it has a maximum throughput of only 100mbps. I thought it was GigE (1000mbps) and just expected it to be capped at USB's maximum of 480mbps. It's not, and I've only been able to get about 80mbps out of it. That's still not fast enough (10MBs) to do anything useful with the Drobo, so it will just be an extra part I have lying around.
With the new Time Capsule, I saw my wifi Transmit Rate jump up to 130, almost triple what I was getting with the old Extreme.
But I'm sitting right next to the Time Capsule, and the maximum "n" speed is 300, so it should be faster. Then I notice that it is connecting to the 2.4Ghz band. A nice thing about the new Airport Extremes and Time Capsules is that they broadcast on both the 5Ghz and 2.4Ghz frequencies (so 2.4Ghz-only "n" devices like iPhones and iPads can still join). I wanted to connect at 5Ghz, expecting it to be faster, but the laptop always ended up on 2.4.
I fixed this with Airport Utility by setting the 5Ghz frequency to get a different SSID under the Wireless tab, then the Wireless Options button.
Now my wifi is 6 times faster than when I started. It's still nowhere near the 1000mbps of a wired GigE network, where a NAS would work fine because there's more than enough bandwidth to get full performance from even a 100MBs (800mbps) hard drive. But accessing the same drive at maximum wireless speed, 300mbps, still cuts off about 60% of the drive's performance (300/8 = 37.5MBs). Even so, 30MBs should at least be usable, not like 5MBs. Or 2.5MBs, as experienced by this person whose DroboFS review I enjoyed.
I'll tell you whether any of this saves my DroboFS from the garbage can as soon as my Time Machine backup to the Time Capsule finishes.
* * * *
After my new Synology DS412+ arrived, I took 4 WD20EARS 2TB drives out of the DroboFS and put them in the new NAS configured in a RAID10 set.
RESULTS from "dd" tests at DS412 shell:
write: 273MBs (2048/7.49)
read: 218MBs (2048/9.36) <-- wow, is that crazy fast for "green" disks?
RESULTS from "dd" tests at Mac shell writing to DS412's AFP share over Wifi
read: 24MBs (2048/85.76)
write: 27MBs (2048/74.68)
RESULTS from "dd" tests at Mac shell writing to Time Capsule (4th gen) share over Wifi
read: 19MBs (2048/108.82)
write: 12MBs (2048/175.14)
Summary: Time Capsule passes (it's just backups and doesn't need to be very fast). DS412+ fileshare over 802.11n wifi is in the usable range. Numbers directly on the DS412 made me recheck my math 3 times.
read: 19MBs (2048/108.82)
write: 12MBs (2048/175.14)
Summary: Time Capsule passes (it's just backups and doesn't need to be very fast). DS412+ fileshare over 802.11n wifi is in the usable range. Numbers directly on the DS412 made me recheck my math 3 times.
This tells me that, yes, the DroboFS was about the worst network attached storage device available on the market when I bought it in 2010. They still sell them today. Garbage.
Posts
No comments:
Post a Comment