does it ever rain in Seattle?
The grunt sculpin is my favorite fish at the Seattle Aquarium. The tank at the aquarium has about 4 to 5 grunt sculpin. It’s really interesting to see how they get into the empty giant barnacle casings. I observed one of them headed toward an empty casing and as it got close to the opening, it rotated 180 degrees and went into the empty casing tail first.
Check out the slideshow i created for the aquarium trip.
Isn’t it the cutest thing you have ever seen?
My Thecus N2100 had always been running hot ever since i bought it in March this year. During a hot weekend in mid summer when the daily highs were in the mid to high 90s, i noticed the internal disk drive (HDD) temperature alarmingly hit 58°C. I took the drastic measure of shutting it down for the whole weekend. Consistently running disk drives at more than 50°C can’t be good for their reliability and the longevity of the disk drives in the long run. For comparison, the 2.5″ 60-GB Fujitsu disk drive in my first generation 17″ G4 Powerbook normally stays in the range from low to mid 40°C.
According to the specifications for my Seagate Barracuda 400 GB SATA disk drives, the drive case temperature should not exceed 69°C (156 °F) and the ambient temperature should fall between 0 and 60°C. The FAQ at Seagate defines ambient temperature as the temperature of the environment immediately surrounding the drive.
I have been researching the past week trying to come up with a good solution to cool my disk drives down. Initially, i took the lid off the case and had a floor fan blow toward case. This managed to bring the temperature down to about 39°C when the NAS was idle. Even when operating under load, the disk drive temperature would never exceed 42°C. This tells me that Thecus did not design the airflow within the case properly.
My guess is that the tiny vent slits in front of the case are simply not providing openings for cool air to enter the case. I decided to implement Leyton’s two-fan mod at the Thecus User Group forum.
I bought two Link Depot 40×40x20mm case fans and two three-pin Y fan cables at the local Fry’s Electronics on Saturday. To make a cut on the only side of the case roomy enough to fit the 40×40x20mm fans, i bought a Dremel rotary tool kit. I also bought two Irwin Quick-Grip clamps to hold the case in place when i am making a cut on the case. The Irwin Quick-Grip clamps are such a delight to use and they are so much more efficient than the old style C clamps.
The first thing i did was to remove the HDD cage from case. After that, i removed the four screws at the back of the base that were holding the stock fan in place. Next, i removed the motherboard from the case.
After keeping the case in place with the two Irwin Quick-Grip clamps, i used the drill bit attachment on the Drexel and made about 20 holes on the side of the case closer to the 128-meg memory module.
On the opposite side of the case, the only area in the case roomy enough to install additional fans, i cut a 80×40mm rectangular hole with the cutting attachment on the Dremel. The edges of the rectangular hole is not as straight as i like it to be, but for its intended purpose of fitting the two fans, it works fine.
After placing the motherboard back into the case and reinstalling the stock fan, i connected the fans to the only power source with the Y fan cables (sharing with the stock fan) and placed the fans in the cut area.
The first thing that i noticed after powering up the N2100 was the fans were running at very low RPM. I could barely feel the air drawn out by the fans. The result of this two-fan mod was very disappointing. The idle HDD temperature remained high, at around 47-48°C. When placed under load, the HDD temperature went up to 50-51°C. This was when the room temperature was probably around 75°C. I immediately realized that the problem of sharing one power source among three fans is reduced airflow. When my stock fan was the only operating fan, it would normally run between 5000 - 7000 RPM. With the three fans installed, it was difficult to get an accurate reading of the fan speed but i estimated that they were only running at between 500 - 1500 RPM. The Link Depot fan is rated to push 7.5 CFM of air and that’s probably when it is running at full speed. So having three fans installed was actually no better than having just the lone stock fan.
Late last night, i decided to try using just two fans - the stock fan and one of the two Link Depot fans. For the Link Depot fan, instead of having it move hot air out of the case, i flipped it around and have it move cool air into the case. To my delight, the idle HDD temperature went down to 41°C. When operating under load, the HDD temperature went up a few degrees to 44°C.
As shown in the graph above, the HDD temperature during idle system between 0310 - 1007 hovered around 41°C. To make sure that the fan blowing air into the case was indeed helping to keep the idle temperature low, i removed the Link Depot fan at 1007. The HDD temperature went from 41 to 46°C in 50 minutes. When i reinserted the fan at 1057, the HDD temperature started dropping and it went down to 41°C at 1214 and it was able to keep constant at that temperature until i started to write/read a 24-GB file at 1252. The load test ended at 1418 and the temperature gradually went down by one degree. The temperature near the N2100 was about 77°F (25°C) during the load test.
I am going to tape up the opening of the side of the Link Depot fan to see if that affects the temperature reading. If that does not have a detrimental effect, i am planning to aesthetically improve the look of the case by adding some metal mesh to both sides of the case.
Note: To capture the temperature data, i created a cron job on the N2100 and have it write the temperature readings to a file. This is the same temperature reading shown in the System Status page on the Thecus N2100 web admin tool.
Update: 9/18/2006 22:20
As the room temperature falls, so does the idle HDD temperature. The idle HDD temperature is now 37°C. The HDD temperature has never been that low.
Now that i have pretty much narrowed down the disk configuration choices for my N2100 between RAID0 and JBOD, i want to compare the AFP/SMB file transfer rate using my gigabit LAN instead of a crossover patch cable. Testing it on a gigabit LAN also allows me to compare the performance with the IP-sharing option enabled.
Test configuration:
Thecus N2100 firmware 2.1.04.1 RAID0 vs JBOD (PDF)
The file transfer rates for RAID0 and JBOD are almost identical.
Thecus N2100 firmware 2.1.04.1 Share IP vs non Share IP (PDF)
Having the IP-sharing option enabled actually causes the performance to be degraded quite severely. The average write transfer rate (i take the average of the degradation percentage) goes down by 18% while the average read transfer rate takes a nose dive, degrading by 45%. Another odd behavior of enabling the IP-sharing option is that the write transfer rates are better than the read transfer rates.
Because of the lack of documentation, i am not exactly sure what the IP-sharing option does. My guess is that this is some form of etherchannel, which is probably not supported by my Netgear gigabit switches. During my tests, i connected both gigabit ethernet ports to my Netgear gigabit switch and i was able to verify that they were both running at gigabit rate.
One thing i found out during the tests is to verify that the ethernet port on the N2100 is running at 1000 Mbps instead of 100 Mbps. Even though all my switches are capable of running at 1000 Mbps, during a serie of tests, i realized that the file transfer rates had dropped quite significantly. It was then i determined that the N2100 was only running at 100 Mbps. Next to each ethernet port at the back of the N2100 are two LEDs. If the right LED shows orange, then it is running at 1000 Mbps. If it’s green, it’s only doing 100 Mbps.
When i did my previous two performance tests (SMB and AFP) on the new beta firmware version 2.1.01.4, the hard drives in my N2100 were configured in RAID0. I was told that if the N2100 was “re-RAID” with the new beta firmware, the performance would even be better.
I spent the previous few days migrating the more than 400 GB of data on my N2100 to all my available storage (two external firewire drives and the internal storage on my PowerMac Cube). In addition to validating the beta firmware re-RAID performance claim, i wanted to test the new firmware in all possible configurations (JBOD, RAID0 and RAID1). I also tested the performance under Windows XP since the gigabit ethernet card in my IBM notebook is the only device i have in my possession that supports Jumbo Frame.
All the tests were done with stripe size of 64 kbyte. For some strange reason, when i configured the N2100 to use JBOD, it asked for the stripe size. It’s my understanding that the stripe size is only used when the drives are being configured in RAID0. When the drives are configured in RAID1, the “stripe size” option is greyed out. I attempted to test the JBOD configuration with stripe size of 256 kbyte but IOzone kept breaking the network connection between my Windows XP-based IBM Thinkpad (i had to disconnect and reconnect the crossover patch cable to reestablish connectivity between the N2100 and the Thinkpad).
Test configuration:
JBOD:
N2100 beta firmware 2.1.01.4 JBOD test (PDF)
Ignoring the read transfer rate of more than 30 meg/s for XP, which was most likely caused by the memory cache as i was not able to utilize the option in IOzone to unmount the share folder after each operation in Windows XP, the read and write transfer rates for SMB in Mac OS X are very similar to XP in Jumbo Frame mode. I had thought that having the ethernet card in Jumbo Frame mode would make a big difference but it does not appear to be the case in the JBOD test.
AFP (Apple Filing Protocol) has the slowest read and write transfer rates.
RAID0:
N2100 beta firmware 2.1.01.4 RAID0 test (PDF)
Again, ignoring the 30 meg/s transfer rate for XP, transfer rates for SMB in Mac OS X is only slightly slower than the transfer rates SMB in XP with Jumbo Frame enabled.
AFP has the slowest read and write transfer rates.
RAID1:
N2100 beta firmware 2.1.01.4 RAID1 test (PDF)
AFP has the slowest write transfer rate but in the RAID1 test, but its read transfer rate is comparable with SMB in both Mac OS X and Windows XP (Jumbo Frame disabled).
To be able to tell which disk drives configuration in the N2100 offers the better overall performance, i plotted the AFP and SMB test results for Mac OS X in two separate graphs:
N2100 beta firmware 2.1.01.4 SMB comparison (PDF)
N2100 beta firmware 2.1.01.4 AFP comparison (PDF)
As i have expected, when the drives are in the RAID1 configuration, the write transfer rate is the slowest. If i am correct, the processor has to wait for the write to both disk drives to complete before continuing on with the next block of data waiting to be written. What surprises me is that the RAID1 read transfer rate is also the slowest. Prior to the test, i had expected the RAID1 read transfer rate would be the higher than JBOD or RAID0. During the read transfer test for RAID1, i noticed that, based on the lights in front of the N2100, it seemed to alternate the reads between the two disk drives.
I had also expected disks configured in RAID0 would have the fastest read and write transfer rates. Surprisingly, the JBOD configuration has comparable transfer rates (JBOD has a slightly better read transfer rate in AFP).
Prior to these extensive tests on the N2100, my N2100 had always been configured in RAID0. The reasons why i went with RAID0 instead of JBOD or RAID1:
The surprising results for configurations in RAID0 or RAID1 seem to indicate that the CPU in the N2100 is underpowered to perform RAID operation. Other than hacking into the N2100, there does not appear to be an easy way to determine the CPU utilization for the N2100 during the tests.
Now that the test results are showing me that JBOD has comparable results with RAID0, when i have completed all my tests, i am seriously considering using JBOD on my N2100. I believe JBOD has a better MTBF (Mean Time Between Failures). For RAID0, the MTBF is inversely proportional to the number of disk drives. So for a 2-disk drives RAID0 array, since the data is striped across the disk drives, it has the reliability of half a disk drive. If i lose one of the two disk drives, there is very little chance of recovering the data in the remaining good disk drive.
The N2100 re-RAID using the new beta firmware seems to have better performance for SMB in Mac OS X, in comparing with the preliminary test i did earlier. The gain is about 2 - 3 meg/s in both read and write transfer rates.
During the tests, i forgot to disable the iTunes service (it was configured to rescan every minute). In my next post, i will add more Mac OS X-specific test results.
Finally figured out why i was not able to mount the share folder in SMB earlier. When i rebooted my Powerbook, i found myself staring at the spinning wheel of the OS X startup. No matter how many times i rebooted it and no matter how many devices i disconnected from it, my Powerbook would remain stuck at the startup screen. After i boot into single-user mode (press cmd-s during startup), i realized that i forgot to rename the mount command back, so basically, on startup, OS X was running the shell script i created for the IOzone performance test. After mounting the root filesystem in write mode (by default, starting OS X in single-user mode mounts the root filesystem in read-only mode), i restored the original mount command. Duh!
The results from the IOzone test is pretty amazing. The improvement in the write transfer rates range from 76% to 96%. The improved write transfer rates are now better than the theoretical transfer rate of a 100 Mbps LAN! Except for a 9% gain in read transfer rate for the 32-meg test file, the rest of the read transfer rates range from 52% to 101%!
I am extremely pleased with the performance improvement in this firmware. Prior to this firmware, i pretty much gave up on any hope that the transfer rate for the N2100 would increase beyond what it was. Kudos to the engineers at Thecus for making this possible.
Test configuration:
I was a little skeptical when being told that this version of the beta firmware for the Thecus N2100 would provide “noticeable speed difference”. So i ran through an IOzone test, with the same parameters as the previous test i did when i had the 2.1.00.2 version of the firmware.
The results blew me away. The improvement in the write transfer rates range from 116% to 164% for various file sizes! The improvement in the read transfer rates is pretty astounding as well, ranges from 55% to 68%.
I included the results from the previous test in the chart as well for comparison.
After installing the firmware, i have problem mounting share folder in SMB, getting the following error:
I have no problem at all mounting the same SMB folder in my Windows-XP-based notebook. When this issue is resolved, i will post the performance result for SMB.
I have just received a new beta firmware version 2.1.01.4 for the Thecus N2100. This firmware is supposed to have some performance improvement. Once i have backed up the data i have on my N2100, i will install the beta firmware and repeat my IOzone test.