D-Link Forums
The Graveyard - Products No Longer Supported => D-Link Storage => DNS-321 => Topic started by: ruans11 on December 19, 2009, 09:58:38 AM
-
I just got a DNS321 and setup with 6k jumbo frame, ext2 formated 1.5TB hard drive. I use a 2GB file to test the speed. the write to DNS321 speed is about 18MB/s, but read from DNS 321 is only about 10MB/s. anyone know it read speed is so slow compare to others? I install the funplug and telnet into DNS321 to see the CPU usage. when write to, it is ~80%. when read from, it only use ~30%.
Thanks,
-
Doubt this will fix your dilema but did you also set your computer ethernet connection to 6K jumbo frames?
I'm running 4K jumbos myself. You could try to drop those frame size to see the impact. The only other think I can think of is maybe your Antivirus program is scanning the data before writting it. Try disabling the AV if you can.
-Joe
-
I have two computer all set to 6K jumbo frames one is vista other is windows 7. between the PC, the copy speed can reach to 70~80MB/s. other interesting behaver of the DNS321 is if both PC copy the same file at same time from the DNS321, both copy can reach about 9MB/s that is total 18MB/s but single copy only 10MB/s :(. inside DNS321 two smbd run at ~30% load each when make two PC copy the same file.
Thanks,
-
Have you tried a direct connection of the NAS to a single PC?
I'm trying to rule out hardware because I just don't see why it would be so slow to read.
And you were getting that speed before installing funplug, right.
I guess the last thing two things to try is, reset NAS to defaults, and try a different modle hard drive. It doesn't need to be a large drive, just a different model.
I hope you are able to figure it out because I'm out of ideas at this time.
-Joe
-
tried with direct connection of the NAS to a PC, got same result. tried a different brand hard drive with no difference. firmware 1.02, 1.03 get same result. funplug not affect anything, nothing works :(
Thanks,
-
I ran a few tests on my NAS and computer and my Write is ~18MB/sec, Read is ~13MB/sec with 4K Jumbo set on both my NAS and my computers network adapter. At 9K the write stayed the same but the read dropped slightly. I won't speculate to the cause as I'm not a network guru and don't know much about packet collisions, etc... My network adapter only supports 1.5K, 4K, and 9K. But I tried setting the NAS to 5K, Network adapter at 4K and guess what, my read speed jumped up to 14MB/sec (some peeks at 16MB/sec) while write dropped down to ~17MB/sec. Odd for sure since I didn't expect a speed up on the read but the write did take a slight performance hit. I was transferring 1GB VOB files and the speed was the one Windoze 7 reports during the file transfer.
One thing I did note, when there was file fragmentation (the hard drive was moving the heads around allot), the read speed dropped. I'm running a RAID1 setup, EXT2 formatted drives.
Why don't you make the changes I suggest which are, MTU on NAS 5K, MTU on your network cards 4K, try transferring a 1GB file and see what your speeds are. I will leave mine like this for a while and test it again later to see if the speed holds and then switch back to 4K on the NAS.
One thing I don't know is that the NAS displays 4000 and 5000, it all ends with an even thousand number. I'm curious if they are the actual numbers the MTU is using or just represent 4096 and 5120. I would like to see those values changed to the standard used.
-Joe
-
One thing I don't know is that the NAS displays 4000 and 5000, it all ends with an even thousand number. I'm curious if they are the actual numbers the MTU is using or just represent 4096 and 5120. I would like to see those values changed to the standard used.
-Joe
I've always wondered about this as well. :)
-
finally use firmware 1.01 I can get about 19MB/s read and 16MB/s write for big file. 1.01 is much fast then 1.03 for reading.
I have test all the combination for the jumbo frame size. the best is 6k-6k setting. I think the network setting is good and the problem is NAS firmware. I used to have a Maxitor MSSII NAS at same network can get 26MB/s read and 16MB/s write. the MSSII have same 400MHz CPU and 64MB memory as DNS321.
-
Hey, glad you were able to make things work nice a fast. Too bad you had to revert to 1.01.
Bet if we could get a stripped down version of the software, maybe only with a few features, that would increase throughput. I do think some of the features are nice but I'd like to see those set up as maybe a fun plug extra.
Well I've got the 1.01 version of the code, maybe I'll take a gander at it over the next few months. One big fix I know I'd have to incorporate is the hard drive issue fixed with 1.03.
-
JoeSchmuck,
Any updates on the performance problems with firmware version 1.03? I'm seeing the same problem with my DNS-321. I was running 1.02 and only getting about 10 MBytes/Sec.
After hours on the phone with D-Link, I finally got to a product specialist and learned that 10-12 MBytes/Sec. is the expected performance level. It sounds like the subsystem that manages network traffic is the bottleneck. I was told that the engineers are currently testing a new version of the firmware that should improve performance.
-
I'm also experiencing disappointingly slow transfer rates, read and write. It's a great little device for what it costs, but the transfer rates are a joke and it's obvious the inclusion of gigabit is just for namesake. I'm getting typical 100mbps transfer rates.
I tried the jumbo frames settings on all gigabit devices on the home network and actually saw a slight drop. 1500k seems to work best for my setup despite having a switch and PCs NICs capable of up to 9k frames. I immediately updated to 1.03 firmware as my 2 x 1.5TB drives wouldn't install with 1.00
I know my internal network isn't the issue - the NAS replaces a Vista home theater ("server") gigabit machine and the transfer rates were pretty insane (30+mbps read/write). I'll miss these speeds but for the compromise I'm willing to keep the dns-321. It's a great storage device, and with the p2p and ftp capabilities it's completely replaced my home 'server'. Having raid1 is just bonus.
-
I had my best results with 4k jumbo frames, you might give them a try. You should be able to do slightly better, perhaps in the mid teens.
-
I gave 4k jumbo frames another try and they MAY have helped, pushed 10mbps last night on a large transfer. This was read speed. It's not horrible but it's not what I expected.
-
Well, with firmware 1.03, a RAID-1 array, gigabit all around, and 4k jumbo frames, I'm only getting about 11mbyte/sec from my DNS-321 reading a 690mbyte file. Maybe this is it, haven't done this benchmark lately. I think with no RAID I used to get around 14, so I don't know what changes.
-
Well, with firmware 1.03, a RAID-1 array, gigabit all around, and 4k jumbo frames, I'm only getting about 11mbyte/sec from my DNS-321 reading a 690mbyte file. Maybe this is it, haven't done this benchmark lately. I think with no RAID I used to get around 14, so I don't know what changes.
The shock is using the same network setup with the only change being the NAS and losing 20+ MB/sec.
-
Well, I've never seen more than 14-15mbytes/sec from this unit, I can't imagine how you can get 30mbyte/sec under any configuration!
-
Well, I've never seen more than 14-15mbytes/sec from this unit, I can't imagine how you can get 30mbyte/sec under any configuration!
I suspect you're correct. Point is - my upgrade path pretty much demonstrates the issue is with the NAS. Any explanation otherwise is a lie.
Do they need to fix it? Not sure. The expectation I had can't be unique to potential buyers. So unless they want to keep tarnishing their brand name they either need to get busy fixing the problem (not likely) or offer upgrade promotion pricing to existing DNS-321 owners in order to get a product whose network performance comes remotely close to what people actually want from their NAS. This is like buying a 5,000 pound Cadillac to find out it has a 1.6L 4 cylinder under the hood. Technically it will move the wheels, but you should expect to get out and push on the hills.
-
It definitely seems internal, since I had two drives in my NAS (non-RAID) and attempted to copy everything from one drive to the other (500 GB). This took over 40 hours! That's less than 4 MB/s for transfer speed. It seemed like a difficult task for my device, since it had to both read and write without using the RAID controller, but wow, that was slow!
-
There must be a difference in the firmware between the 321 and the 323 model. I have the 321 at home and noticed the poor read/write speeds while the 323 at work is very fast. I didn't need the USB ports at home, so I saved a few bucks and got the 321. Maybe I shouldn't have. I hope they fix the firmware.
-
The DNS-321 and DNS-323 have different processors, so the DNS-323 will be a bit faster, about 20% if my tests can be believed.
-
Before blaming the DNS, one can verify if the Gigabit network is indeed supporting that bandwidth. Below is the link to download iperf to a windows box, you can run it to find out what bandwidth your switch and router gives you.
download link:
http://www.noc.ucf.edu/Tools/Iperf/
command from windows 1 box:
iperf -c <ip1> -u -b400M -w2m -i1 -t60
e.g. ipeft -c 192.168.1.101 -u -b400M -w2m -i1 -t60
where:
<ip1> is the ip address of another windowx/linux box
Wait for a minute for it to transfer data and report result, then issue another command
iperf -c <ip2> -u -b400M -w2m -i1 -t60
where:
<ip2> is the ip address of the NAS box
Another way to find out about the true transfer rate is via Filezilla, which fluctuates between 6 MB/s to 20 MB/s, average about 10 MB/s. Do this exercise between 2 boxes and then repeat with DNS as server (enabling FTP server option would help here).
Cheers
-
For me, since the slow read speed occurred within the NAS system as a disk-to-disk copy, I'm going to go ahead and blame the DNS.
-
Well, I think the DNS does have some contribution to the slow speed as well, but not all of it. I also noticed that when I forced the speed to 1000 Mbit/sec instead of Auto and used 9K Jumbo frames the DNS performed better than 10 MB/s most of the time. Since I switched back to Auto (& left the 9K frames), the performance went back down to about 6MB/s. So playing with the configuration could help.
-
10MBps is pretty crappy gigabit throughput. The fact this is acceptible for this device speaks volumes.
I wanted to swap out my router and switch setup anyway, so I did a couple days ago. No change (at least things didn't get worse lol). I'm resigned to live with the product but it's tarnished my belief in cheap solutions and next time I'll pass on any low-end dlink products and save a few pennies for something better with better overall reviews.
-
You should be able to do better than 10mbytes/sec with this box, but I wouldn't expect miracles. :) Figure in around 15mbytes/sec...
-
Being a newbie at network speeds, my new 321 is taking about 14 minutes to copy a 1GB movie file from my Vista 64 machine to the 321. What's the math for figuring out the Mbps - and is it bits or bytes? Both are hardwired through a dlink 8 port gigabit switch and both show a gigabit connection at the switch. I have configured the 321 for 1000 Mbps, 5000 Jumbo frame size.
14 minutes to copy a 1GB file seems absurd to me. 1GB network speed would seem to imply that this task should take just a few seconds on a network with nothing else running??
Like I said, a newbie at this - so please help me align my expectations.
Thanks
-
More info - found the program LAN Speed Test and ran it on the 321. Write was recorded at 26 Mbps, read at 9 Mbps. However, it took 30.5 seconds to write a 100MB file and 88 seconds to read, so the math still doesn't seem to add up. And nothing is anywhere near the Gigabit performance advertised.
Disappointed to say the least.
-
For typical speeds of the DNS-321 on a gigabit network with jumbo frames, I'd expect that copy to take something over a minute, surely less than two. Figure on around 15 mbytes/sec on a good day.
1024 mbytes / 15 mbytes = 68.27 seconds.
Now, things never seem to go quite as fast as you expect, hence my hedging on the speed. :)
Your speed is barely over 1 mbyte/sec, clearly way slower than it should be, even on a 100mbit network. Those are the speeds I'd expect on a 10mbit link.
-
14.5mb/s is about all your going to get with the dns-321 or dns-323. I got a great rebate on the 321 but I had no ideal the speed was so slow. My thumb drive is a tad bit faster then this. I only get 13.5mb/s :(
http://www.smallnetbuilder.com/component/option,com_nas/Itemid,190
-
Excuse my ignorance - Where do you set the jumbo setting in the 321
-
Open the configuration menu, SETUP and select LAN on the left. Jumbo frames are at the bottom of the page.
-
I am assuming "MTU"
Thanks
-
Yep, that's the frame size.
-
I bought the DLink DNS-321 based on following advertized "FEATURES & BENEFITS" on their spec sheet at ftp://ftp10.dlink.com/pdfs/products/DNS-321/DNS-321_ds.pdf :
1 10/100/1000 Gigabit Ethernet Port
High-speed Gigabit Ethernet Connectivity
Utilizing SATA hard drives and gigabit connectivity, the 2-Bay Network Storage Enclosure (DNS-321) is a high performance, cost-effective solution for your home or small office data storage needs.
However, the device does not actually provide Gigabit Ethernet Connectivity. A search on the forums shows that typical data transfer speeds are in the range of 10 Mbps.
Other devices on my gigbit network transfer in the range of 100-140 MB/sec. However, I've never seen anything faster than 10MBps transfer speeds to or from the 321. That is not "Gigabit Ethernet Connectivity". Therefore, either the product is not working correctly, or DLink is lying in their specifications for the product. I've made several calls to Dlink support, and they have been unable to provide answers to this issue. They cannot explain why the device does not actually provide Gigabit Ethernet Connectivity.
DLink should not be claiming specs that is cannot substantiate. DLink should not be claiming specs that is cannot substantiate. That's false advertizing, and is against the law.
-
What devices do you have on your network that are able to transfer at 140mbytes/sec on a gigabit link? Since the theoretical limit is 125mbytes/sec assuming no protocol framing overhead, that would be a pretty neat trick!
-
I have a screenshot showing a transfer between two computers at 105 MB/Second. That is typical but I'm sure I've seen much faster. Regardless, there's a huge difference between 100 and the 10 that the DNS provides.
-
Let me correct my previous statement - "Other devices on my gigbit network transfer in the range of 100-120 MB/sec (not 140)"
Sorry for the typo. Regardless, Dlink is advertizing "High-speed Gigabit Ethernet Connectivity" an "high performance" that it apparantly does not deliver.
I've never seen anything faster than 10MBps from the device. Therefore, either the product is not working correctly, or DLink is lying in their specifications for the product.
-
I've owned the DNS321 for a little over a year now and I get between 6 MB/s -14 MB/s. I did extensive testing to try to figure out what was causing such slow speeds. I thinking it comes down to having a slow processor in the unit. I have Jumbo Frames set to 7000 for all of my computers and my Enterprise HP Switch supports Jumbo Frames and Gigabit Connection. I can get 115 MB/s file transfers from PC to PC since all of my computers have raid 0 configured. Essentially if you plan on owning or currently own a DNS321 don't expect to get higher transfer rates then 6 - 14 MB/s. Now with that said, I like the device very much I have it set to be hot swappable in case one of the Hard Drives fail. This device has saved me at least twice since I store all my important data on it. One time a hard drive failed and all I had to do was pop it out and push in a new one...it did all the formatting and file recovery to the second drive from the first one. Neat little device....but slow as heck!
Just one more comment to the previous guy. For a NAS it's actually a decent speed for the price. It may be misleading because I wouldn't call it FAST! But if you look around, other NAS devices cost several hundread dollars more and only give speeds up to ~40 MB/s. Netgears NAS that my buddy has tops out a ~20 MB/s and cost several hundred dollars.
-
I saw same issue with slow speed, and have switched back to 1.01 firmware.
Write speed is ~14-15 MB/s in 1.01, I didn't test thoroughly in 1.03. Read speed is 17-18 MB/s for 1.01 vs 6MB/s in 1.03. I just can't live with 6MB/s. No way, no how.
-
I noticed that the speed dropped with the newer firmware, both with this unit and the DNS-323.
I finally gave up on speed, formatted the disks EXT3, and I use these two units as unattended backup servers. I picked up a Synology DS209 as my primary NAS, and it cooks along at 40-50 mbytes/sec, I can live with that. :)
-
I have done extensive testing and found that Jumbo frames of size 9k on both DNS-321 and my Win 7 x64 machine performs better than any other Jumbo frames setting (I have also tried the jumbo frame size mismatch suggested earlier).
My read speeds are around 10MB/s and writes are 7.75MB/s, notice that this is standard non RAID setup, so I would expect the read speed to double for RAID setups.
Disabling jumbo frames reduces read speeds by about 10% and did only have slight effect on the writes.
Its worthy nothing that the 4K, 5K, 6K matching frame sizes produced worse numbers than disabling jumbo frames all together.
DNS-321 (FW 1.0.3) WD EARS 1.5TB
-
I ran some tests and noticed write and read speeds of 5.27 MBytes and 2.78 MBytes. It is really confusing to see slower read than write. Still not sure what is the purpose of giving gigabit port if it can not go beyond 100mbps.
-
The CPU in this box is very slow. There is also a bug in the 1.03 firmware in the samba version they used.
Downgrade to 1.01.
Due to the slow CPU expect no more than 15MB/s over gigabit. That's all it can do. The new boxes from D-Link have better CPUs. For us that bought this one, we're stuck with a sad performing device.
-
I wanted to add that I also experience this problem & that I'm very disappointed in D-Link.
After installing a telnet daemon, I logged in and observed the CPU usages (smbd and total), and while sending a file over a gigibit link with a MTU of 7000, I get 10-14MB/sec. CPU usage is only about 20-40% at worst and I'm getting no retransmitted packets (ie. client machine and all switches involved are handling the MTU ok).
It got me interested... If samba wasn't the bottle neck, what was? So I put a large file on /mnt/HD_a4 and then cp'd it to /dev/null (this should be pretty fast). Again, the CPU usage was about the same as samba, and worse yet, the read speed was about the same as samba.
So I thought, maybe it's quota checking... So I put a 280MB file on /mnt/HD_a2 (which has no quota checking enabled), then cp'd it to /dev/null... The results (don't hold your breath here) are the same.
This really feels like D-Link as put a speed governor into the kernel, a lot like Intuit does to their consumer grade Quickbooks products.
Perhaps they compiled the kernel as a tick based timing system (see /etc/rc.sh : adjtimex -t 10000), and the kernel thusly just doesn't have enough ticks available to service the IO requests (my gigabit cards can produce over 30k interrupts when doing a 80MB/sec transfer). But, on this one I'm probably just showing my ignorance, because they could simply be setting the resolution of the RTC (real time clock).
I would assume that they cheaped out and are doing Port IO (PIO) mode access of the HD's, except that the IO time doesn't jump up, and over all CPU usage remains under 40%.
I suppose you get what you pay for... Thought I was getting a good deal, but I can see now I was really getting cripple-ware.
Unless D-Link invests a little time and effort and fixes the horrible read/write speeds of the HD's (my laptop 486/90mhz had better speeds, and certainly ultra cheap SATA devices should be reading at more than 30MB/s), I'm going to have to seriously consider never touching another D-Link product again, and since I'm a computer consultant, I can honestly tell my clients my experiences with D-Link products and that will likely discourage them from throwing away good $$ on crippled products.