Qnap TS-509 New Firmware performance & bug report

[Dennis Wood]

Kurt, I've tried the WOL procedure using the two modes that work, failover (2 ports connected) and with a single port connected in standard mode. In both cases I'm using the MAC address found by the Qfinder utility. In load balancing mode, the TS509 does not work with statically configured link aggregation. With the LAG tunnel disabled on the switch, the unit does stay connected in load balancing mode, but does not actually load balance based on analysis of port stats on the switch. I've got another 3Com switch coming that fully supports 803.2ad, so I'll have a go with that.

WOL works fine on this network with other workstations.

The fact that with the NAS off, the LAN indicator on the TS509 flashes 3 times immediately following issuing a WOL request, would suggest that the unit is receiving the request but the BIOS call to power up is failing. The unit has never successfully powered up using the admin schedule, so perhaps the problems are related.

 

[AndyJapan]

Hi, Andy:
We will try to duplicate the issue you encountered.
---
q_support@qnap.com.tw

Thanks Ivan. I already shipped the TS-509 back to the shop from whom I bought it and requested a replacement unit. If the replacement unit shows the same problems, I will ask for a refund and buy another NAS.
You should have received an inquiry (submitted through your website) from me last week. If you have a solution for the described problem, please contact me through the e-mail address I submitted or send me a PM on this forum. Thanks in advance.

 

[Dennis Wood]

So the unit does shut down, but only 24 hours after it's scheduled. In other words, if I schedule a shutdown 5 minutes from now, it ignores this, but shuts down at the set time, 24 hours later.

It has never started up on the schedule in repeated tests, but always remains off.

 

[Dennis Wood]

Beta firmware testing...

All of the issues we were seeing appear to be resolved with a beta code update.

1. Backup to esata works fine now.
2. Schedule shutdown and start works now.
3. Larger files are copying over much better at around 30 to 40 MB/s via SMB2 and the NAS is not bogging under multiple loads.
4. WOL is working. I'm using FUSION WOL, a free windows app. Haven't tested Qfinder yet.
5. Transfer rates to eSATA drive over the LAN are much faster..about 20MB/s

More testing to come using our standard test suite.

 

[Dennis Wood]

...and a few performance numbers on the 2.0.2 Beta Code which has the unit now providing respectable numbers, particularly to the Vista SP1 RAID 0 workstation. These numbers are all RAID 5 tests using 5 of the 7200rpm enterprise Seagate 1 TB drives

I. XP SP3 Core2Duo workstation with single SATA drive and PCI connected gigabit NIC:

1. In our "real world" measured tests using a 5.3 GB file copy (tiny to 1.8GB file sizes):
40.2 MB/s write
35.3MB read

2. In our FFMPEG encoding tests which split/encode an 445MB MP4 HD video file to two streams simultaneously
25.2 MB/s write
26.8 MB/s read

II. Vista SP1, Core2Duo RAID 0 workstation (3 drives), with PCIe connected gigabit LAN:

1. In our "real world" measured tests using a 5.3 GB file copy (tiny to 1.8GB file sizes):
45.3 MB/s write
60.2 MB/s read

2. In our FFMPEG encoding tests which split/encode an 445MB MP4 HD video file to two streams simultaneously
32 MB/s write
50.58 MB/s read

A 450GB file copy ticked along at 51 MB/s using the Vista workstation. Looking at the iozone plot for writes, there's an upward curve which likely keeps going up to the 50MB/s ceiling.

Iozone tested to a 4GB file size on the aforementioned workstation:

 

[donsom]

Yikes ... $1400 for 5 Seagate 1TB drives...?? Are they the Barracuda or Cheetah series (10,000 RPM)?

Maybe it's no wonder you were getting faster speeds (if they are 10K drives) than I was on my TS-409. I'm using four Seagate 7200.11 32MB cache drives. I've been picking them up at www.ncix.com (Vancouver, B.C., Canada) for as low as $128 (during sales that they have).

 

[Dennis Wood]

Nope, they're the Barracuda ES.2 7200rpm (ST31000340NS). They are the enterprise version of their 1TB drives. When you stack five of them together, there are considerations of vibration and heat tolerance that are addressed with this type of drive. My math was off..they were just over $1200.

I was looking over our exhaustive test values for all of the workstation, OS, laptop and NAS combinations we've tried over the last month. The 2.0.2 firmware 45MB/s write and 60MB/s values were the highest results of any RAID5 combination tested. The only thing faster was using the same 5 drives in the TS509 in RAID 0 which gave us write values around 80MB/s and read at 90MB/s when tested using the Vista SP1 RAID 0 workstation. If we could tolerate that kind of risk, there's no question I'd be using RAID 0.

I also tested WOL tonight over a VPN connection which worked like a charm (using the free FUSION WOL). It's a nice way to do things as there's no extra ports or forwarding to open/configure.

 

[thiggins]

Dennis,
How are you doing the RAID 0 configuration in Vista?

 

[Dennis Wood]

We have two different workstations using Vista SP1 and RAID 0. Both are running full services, antivirus, diskkeeper etc. to keep the tests "real" and ensure the disks remain defragmented and optimized. They are both running C2D processors at 2.4 Ghz with 2GB RAM. The RAID setup is done in the respective controller BIOS at boot time, and the OS installed with a slipstreamed Vista SP1 build. This way you don't need a floppy, or any input to add drivers. The slipstream is done with all of our potential disk drivers added for both RAID, SATA, and ESATA controllers on the entire hardware collection being tested. Storagecraft is being used to backup XP and/or VISTA SP1 builds so we can quickly restore from the Storagecraft boot CD using a few 1TB ESATA enclosures to store the images.

1. Workstation 1 is Intel ICH7 based, 3 drives (320GB WD) using an ASUS P5W-DH motherboard.

2. Workstation 2 is Nvidia 690i SLI based, 3 drives (300GB WD), using an ASUS P5N32 SLI motherboard.

Here's what the Intel box looks like using iozone and this command:

iozone.exe -Rab c:\test.wks -y 64K -q 64k -i 0 -i 1 -+u -f C:\001.tst -q 64k -n 32M -g 4G -z

The workstations settle in at about 155 MB/s read and 139 MB/s write once they run out of the 2GB of RAM.

 

[thiggins]

Thanks, Dennis. So you are not using the Intel Matrix Storage manager app?

 

[Dennis Wood]

I only use that to set the volume write-back cache to "on" It also runs as a service to notify of any issues. Call it a throwback to the SCSI raid controllers I"m used to working with where everything is done in BIOS. My take on these things is that if you're doing a bare metal restore (much faster) then there must be an ability to set up the array with no OS installed. Storagecraft's boot CD boots to a Vista pre-install enviroment, and allows loading drivers from USB, Flash, etc. so you can simply load the appropriate RAID/eSATA driver and then write the disk image. A 200GB image only takes about 1.5 hours to restore this way.

 

[thiggins]

Thanks, Dennis. A little birdie also told me that Matrix Storage manager also must go through a "consistency check" upon each system reboot. The birdie said the process took around 9 hours for an array of 3 x 250 GB drives!

 

[thiggins]

The workstations settle in at about 155 MB/s read and 139 MB/s write once they run out of the 2GB of RAM.

And what are you getting between the two over a gigabit link?

 

[Dennis Wood]

Between the two Vista SP1 workstations using RAID0 on both, and our 5.3GB test file set:

Write: 72MB/s
Read: 88.7MB/s

If either is using RAID 5, the numbers drop to:

Write: 22MB/s (to the RAID5)
Read: 50MB/s

Our best performance to date is between either of these RAID0 workstations and the QNAP TS509 using the 5 drives in a RAID 0 array:

Write: 81MB/s
Read: 95MB/s

I know right about now you're having some fun with the Ubuntu/Atom build

 

[thiggins]

Between the two Vista SP1 workstations using RAID0 on both, and our 5.3GB test file set: (snip)

Thanks, Dennis. That was the summary that I was looking for. I know that you probably already had it in other threads. But damned if I could find it!

 

[Dennis Wood]

Keep in mind that the results are average measured results based on a file set of various sizes. I've included the dir info here so you can see how big/small they are. In other words, if you copy and paste using Vista, the reported average transfer rates match up nicely with our measured values. Btw, the FTP read rate from the Qnap on large files is hovering around 100MB/s . Be curious to see what you see there using FTP....

Also, with regard to Intel's Matrix manager, it only does a consistency check on a dirty shutdown. As long as you've got a UPS, and a stable machine, it should never happen. With file write-back cache enabled, you really do need a UPS to avoid these situations.

 

[AndyJapan]

Thanks to QNAP Support, especially Ivan.

Hi, Andy:
We will try to duplicate the issue you encountered.

---
q_support@qnap.com.tw

I received the TS-509 back from the dealer and updated the firmware according to your instructions. Then installed the HDDs, configured the TS-509 and let it run overnight to build the RAID 1 volume.
The problem that I described earlier is solved and the TS-509 seems to be lightning-fast. Thanks again for your prompt support. I appreciated that you and your team responded so quickly and solved the problem within a few days. Kudos to QNAP Customer Support.

 

[Dennis Wood]

There are still a few issues that need tweaking and being that this is the "bug" report thread, here they are:

1. In load balancing mode with Link aggregation configured on an 803.2ad switch, performance drops severely with two workstations loading the unit. I'll have numbers for this, but I also saw this issue with 2.0.1 with the unit in failover network mode. In other words, two workstations doing something like backing themselves up to the unit at the same time will bring it to a near standstill. Two workstations hitting it the same time reduced write speeds to 2.3MB/s...not good.

2. A 10GB backup to eSATA worked fine, but a 500GB backup failed. The backup update screen showed .1% complete, and it stayed like that overnight. EDIT: Sorted this issue with a reset and reconnect of the eSATA drive and NAS.

 

[QNAPIvan]

Dear Dennis:
We are trying to set up the similar environment here.
The pipe of the writing (transferring file from PC to TS-509 Pro), the path includes network->memory->RAID 5 operation & disk I/O.

We think the possible reason of the reduced performance is the congestion occurs when the write-in speed (samba network & -> memory) is much faster than the disk I/O & RAID 5 handling. The environment you set seems give much faster traffic to the NAS, it might be over the maximum of the RAID5 operation. It becomes a pipe with unbalanced bandwidth. We will try to set up the environmet and tweak it to make it balance.
We also want to check if the same situation there when the TS-509 Pro running as RAID0 mode. (I guess RAID0 on TS-509 Pro side should gain more writing speed to balance the pipe)

Regading to back up 500GB files via eSATA, we are looking at it. For the speed at about 20MB/s of eSATA backup, in average around 1GB files/min will be transferred. 500GB files might take more than 8 hours in theory. We will check it.

Cheers,
Ivan
Product manager,
QNAP Systems

 

[js.opdebeeck]

Woooo ....

Really better performance

Cool .