Write performance for Zyxel NAS540
beuford
Posts: 10 Freshman Member
I just purchased a new NAS540. I loaded it with a single HGST 4TB NAS drive (7200RPM, 128MB cache and a SATA 3 interface) for testing. It came with version 5.20 firmware, but I've upgraded it to version 5.21 firmware. I purchased this to replace a WD MY Cloud 4TB. All Internet apps were disabled for testing (media server, iTunes server, etc).
Testing was done over a direct 1Gb Ethernet connection between devices. Test computer is an Apple Macbook Pro 15" (quad-core i7,16GB RAM, 256GB SSD) using a Thunderbolt to Ethernet adapter.
When transferring from the NAS to the Mac (read) I get ~ 115MB/sec, but when transferring from the Mac to the NAS (write) I get only ~45MB/sec. For the My Cloud I get 115MB/sec read and 63MB/sec write.
I kind of expected better performance from the NAS540 relative to the My Cloud; what gives? Any ideas where I can improve performance?
Thanks,
Beuford
#NAS_March
Testing was done over a direct 1Gb Ethernet connection between devices. Test computer is an Apple Macbook Pro 15" (quad-core i7,16GB RAM, 256GB SSD) using a Thunderbolt to Ethernet adapter.
When transferring from the NAS to the Mac (read) I get ~ 115MB/sec, but when transferring from the Mac to the NAS (write) I get only ~45MB/sec. For the My Cloud I get 115MB/sec read and 63MB/sec write.
I kind of expected better performance from the NAS540 relative to the My Cloud; what gives? Any ideas where I can improve performance?
Thanks,
Beuford
#NAS_March
0
Comments
-
I kind of expected better performance from the NAS540 relative to the My Cloud; what gives?As far a I know the My Cloud has the same processor as the '540; a Mindspeed Concerto 2000. So I'd expect equal performance. (And at least of a '540 I know it can do better than 63MB/sec)Any ideas where I can improve performance?
Have a look in the webinterface which process is eating most of the CPU cycles, when writing.
0 -
It's usually the web interface....
Funny, though, how Python is always consuming 12 - 20% of the CPU. Memory is nowhere near maxed out, but processor ^ 100% frequently during writes. It's the file system and python competing for CPU resources.
I don't have the NAS540 connected to the internet (security); could this be causing problems? Is there a process trying to "phone home" that might be the cause of the high python utilization?
You'd think it would give up when it realized that it didn't have a default gateway route...
beuford0 -
OK - So here is something interesting. It seems that the processor in my NAS540 is a bit slower than anticipated. From the /proc/cpuinfo files of my NAS540, MyCloud Gen 2 and a MyCloud Gen 1:NAS540:*****Processor : ARMv7 Processor rev 1 (v7l)processor : 0BogoMIPS : 1196.85processor : 1BogoMIPS : 1196.85Features : swp half thumb fastmult vfp edsp neon vfpv3 tlsCPU implementer : 0x41CPU architecture: 7CPU variant : 0x2CPU part : 0xc09CPU revision : 1Hardware : Comcerto 2000 EVMRevision : 0001Serial : 0000000000000000*****MyCloud Gen1:*****Processor : ARMv7 Processor rev 1 (v7l)processor : 0BogoMIPS : 1299.25processor : 1BogoMIPS : 1292.69Features : swp half thumb fastmult vfp edsp neon vfpv3 tlsCPU implementer : 0x41CPU architecture: 7CPU variant : 0x2CPU part : 0xc09CPU revision : 1Hardware : Comcerto 2000 EVMRevision : 0001Serial : 0000000000000000*****MyCloud Gen2:*****processor : 0model name : ARMv7 Processor rev 1 (v7l)BogoMIPS : 1594.16Features : swp half thumb fastmult vfp edsp neon vfpv3 tlsCPU implementer : 0x41CPU architecture: 7CPU variant : 0x4CPU part : 0xc09CPU revision : 1processor : 1model name : ARMv7 Processor rev 1 (v7l)BogoMIPS : 1594.16Features : swp half thumb fastmult vfp edsp neon vfpv3 tlsCPU implementer : 0x41CPU architecture: 7CPU variant : 0x4CPU part : 0xc09CPU revision : 1Hardware : Marvell Armada 375 (Device Tree)Revision : 0000Serial : 0000000000000000*****Can someone PLEASE explain why my NAS540 has a slower processor than a Gen1 MyCloud?!?Beuford0
-
Sorry for the repeated posts - I didn't see the note at the bottom stating that the post had to be approved. I assumed something had gone wrong and tried to repost (again and again...).
Sorry!
Beuford0 -
Short version: /proc/cpuinfo shows a 2-core processor running at ~1200 BogoMIPS per core on the NAS540, but ~1595 BogoMIPS per core on a My Cloud Gen 2 and ~1300 BogoMIPS per core on a MyCloud Gen 1.
Seems a little strange that the NAS540 would report a slower speed than an old My Cloud... That explains a lot, though, since the performance is more on par with my old Gen1 My Cloud.
Beuford0 -
From a web site that records specific information from NAS systems, we have this /proc/cpuinfo output for the same model (NAS540):
**********Processor : ARMv7 Processor rev 1 (v7l)processor : 0BogoMIPS : 2387.14processor : 1BogoMIPS : 2393.70Features : swp half thumb fastmult vfp edsp neon vfpv3 tlsCPU implementer : 0x41CPU architecture: 7CPU variant : 0x2CPU part : 0xc09CPU revision : 1Hardware : Comcerto 2000 EVMRevision : 0001Serial : 0000000000000000**********
My system reports almost exactly HALF of the BogoMIPS reported here, even though everything else in the output matches and the machines are the same.
Any ideas?
Beuford0 -
That's fishy. I guess your data comes from zyxel.nas-central.org. That NAS540 is mine. In that time it was running firmware 5.00. Meanwhile it's upgraded to version 5.11. And guess what? The output of /proc/cpuinfo is changed. It now runs 1196.85 bogomips.
Don't know how I must interpret that. On a 1.2GHz Armv7 CPU I'd expect something like 1200 bogomips. I'm pretty sure the '540 has a 1.2GHz processor. Some Googling around gives no conclusion about the MyCloud. I can find both 1.2GHz and 650MHz.
The bootlog of firmware 5.00 shows[ 0.190282] xor: measuring software checksum speed [ 0.240157] arm4regs : 140.800 MB/sec [ 0.290111] 8regs : 147.200 MB/sec [ 0.339960] 32regs : 89.600 MB/sec [ 0.390023] neon : 147.200 MB/sec [ 0.390031] xor: using function: neon (147.200 MB/sec)
while the current bootlog shows[ 0.192035] xor: measuring software checksum speed<br>[ 0.240014] arm4regs : 2130.000 MB/sec<br>[ 0.289997] 8regs : 1398.000 MB/sec<br>[ 0.339981] 32regs : 1186.400 MB/sec<br>[ 0.389974] neon : 1384.400 MB/sec<br>[ 0.389982] xor: using function: arm4regs (2130.000 MB/sec)<br>
which is not even comparable. So my guess is that there were some timing issues in the 5.00 kernel. Which has impact on bogomips measurement.
0 -
Thank you for the revised info.
However, I am still at a loss with respect to the poor write performance of the NAS540.
I'm only getting 44MB/s writes (copy from computer to NAS, direct 1Gb link, 10GB file). The My Cloud gen 1 does about the same, and the gen 2 beats it by 50% (~64 to 66 MB/s. Reads are at 115MB/s.
Note that I am now running 2 1TB drives in a RAID 0 configuration.
Any ideas??
Beuford0 -
UPDATE
OK, so something I left out - I'm using a Mac.
I pulled out my wife's laptop running Windows and ran the same tests - and got 90-100MB/s for both reads and writes. What the ... ??
So what is different about AFP that causes a 50% performance hit? I tried all three of my Macs and all had similar results. I also tried to use an SMB connection from the Mac, but got ridiculously slow throughput (12MB/s).
Since the WD My Cloud gen 2 does better own writes from the Mac (64MB/s vs 45MB/s) but is otherwise a similar system to the NAS540, I have to believe it's either an AFP version or configuration issue. Any suggestions for improving AFP performance with the NAS540?
Thanks in advance!
Beuford0 -
I never looked at AFP before, and don't know much about it's quirks.
As Samba is fast, it makes no sense to look at disk&network performance. (Although that is other hardware. You could measure network throughput using iperf, which is included in the RandomTools package , provided iperf is also available on your Mac)
My first action would be to compare the configuration. For the '540 you can find it in /etc/netatalk/afp.conf. Do you have shell access to the MyCloud as well? I'd expect an afp.conf somewhere in /etc/. If not, you can runps | grep netatalk
to see which file it got specified. Reading the man page I don't see many options which I would expect to have big impact on transferring a single file. Maybe tcprcvbuf.
0
Categories
- All Categories
- 415 Beta Program
- 2.5K Nebula
- 152 Nebula Ideas
- 101 Nebula Status and Incidents
- 5.8K Security
- 296 USG FLEX H Series
- 281 Security Ideas
- 1.5K Switch
- 77 Switch Ideas
- 1.1K Wireless
- 42 Wireless Ideas
- 6.5K Consumer Product
- 254 Service & License
- 396 News and Release
- 85 Security Advisories
- 29 Education Center
- 10 [Campaign] Zyxel Network Detective
- 3.6K FAQ
- 34 Documents
- 34 Nebula Monthly Express
- 87 About Community
- 76 Security Highlight