Unfortunately no one can be told what fun_plug is - you have to see it for yourself.
You are not logged in.
Pages: 1 2
My home network is as follows:
PC#1: Gigabit Ethernet Connected to Unmanaged Gigabit Switch.
PC#2: Fast Ethernet Connected to NAT Router (10/100).
PC#3: Fast Ethernet Connected to NAT Router (10/100).
PC#4: Fast Ethernet Connected to NAT Router (10/100).
Unmanaged Gigabit Switch: Connected to NAT Router (10/100)
D-Link DNS-323: Connected to Unmanaged Gigabit Switch
PC#1 has average data transfer rates of between about 12 MB/sec and 13.8 MB/sec.
The other PC's have data transfer rates of around 7 to 8 MB/sec.
D-Link DNS-323 configuration:
2 Seagate ST3500NS41 500 GB SATA2 Hard Drives
RAID1
Firmware 1.03
What data transfer rates have others been experiencing?
Thank you.
Offline
Offline
Does it matter that we use Samba or NFS? I don't have a DNS-323(it will arrive in a few days) but I have a Linux machine, and I notice the transfer rate is much slower if I share a directory with Samba comparing with NFS. I am wondering if DNS-323 has the same problem since it also runs Linux.
Offline
I could be wrong, but I don't think the DNS-323 exports an NFS file system out of the box.
Offline
Shouldn't we distinguish between reading from the DNS and writing to it?
Or doesn't it matter?
Cheers,
Emacs
Offline
Although I don't usually report read & write separately, direction does matter - disk subsystem throughput can be significantly different - RAID1 for example can be read quite a bit faster than it can be written, and you also need to consider the disk(s) at the other side of the transfer.
Essentially what it boils down to is the transfer will only be as fast as the slowest link in the chain, whatever that happens to be.
Offline
Well then:
DNS with two 500GB Samsung HD501LJ (SpinPoint T166, Sata-2 interface)
Configured as individual discs - no RAID.
Connection: Gigabit Switch (D-Link)
Testfile: 350 MB (367.029.198 bytes)
Writing to DNS-323: 15MB/s (transfer took 23 seconds)
Reading from DNS-323: 18MB/s (transfer took 20 seconds)
Since nearly every component in the set-up should be capable of _much_ higher transfer speeds, I take it, the IO-Performance of the DNS-323 board/chipset is the limiting factor here.
Still it is lightyears ahead of the Linksys NSLU2 (Slug) I was using before.
Cheers,
Emacs
Offline
Emacs wrote:
Since nearly every component in the set-up should be capable of _much_ higher transfer speeds, I take it, the IO-Performance of the DNS-323 board/chipset is the limiting factor here.
Cheers,
Emacs
That was my conclusion also.
Offline
fordem wrote:
Emacs wrote:
Since nearly every component in the set-up should be capable of _much_ higher transfer speeds, I take it, the IO-Performance of the DNS-323 board/chipset is the limiting factor here.
Cheers,
EmacsThat was my conclusion also.
I saw the smbd process is having 50% CPU usage while transfering data via 100Mbps LAN at 9MBytes/sec. I assume it will be 100% when transfer rate is around 18-20MBytes/sec.
Offline
Nope - it's not the CPU. Top shows mine between 50% and 59% during a 15-18MB/s transfer.
Cheers,
Emacs
Offline
my top shows a cpu load of 86.7% while uploading at 140 ~ 160mbps = 17MBps ~ 20 MBps
Offline
It may not be the CPU that's limiting it - but something in the box is - because I can achieve significantly higher transfer rates (>150%) when transferring to/from my server.
Edit - for the sake of clarity - transfers between a PC running Windows XP Pro and a server running Windows Server 2003 run faster than transfers between the same PC and the DNS-323 - the PC is on 100 mbps and the server and the DNS-323 can be switched between 100 mbps and 1000 mbps.
On 100 mbps, PC to server transfers hit 98 mbps compared to 60-70 for PC to DNS-323 - roughly 150% faster and on gigabit, PC to server hits a maximum of 290 mbps as compared to roughly 150 for the DNS-323 - almost 200%.
With no changes in the network or environment other than the DNS-323 being the source/destination, the bottleneck causing the drop in transfer rate has to be internal to the DNS-323 - whether it is in the silicon or the firmware.
Last edited by fordem (2007-07-30 16:14:33)
Offline
I would assume the limiting factor is something comparable to the chipset (northbridge/southbridge) on a fullsize computer's motherboard.
It's not the disks, neither the gigabit interface. And as we see it isn't the CPU.
Basically the amount of data that can be shuffled around by the backplane of the DNS-323 seems to be the bottleneck.
Cheers,
Emacs
Offline
Out of curiosity - how do you know it's not the gigabit interface - or for that matter, the disk interface?
Here's where I'm coming from - I've seen very disappointing throughput on a gigabit NIC on a Windows box, caused by using Windows drivers rather than the drivers that shipped with the card. With linux, and moreso imbedded linux, it can be quite difficult to pinpoint such a bottleneck.
Also - I believe there is some overhead in the tcp/ip stack - I'm seeing a move towards toe (tcp offload engine) in enterprise grade servers used with SANs to allow for faster throughput on the ethernet connections.
Offline
I was testing the transfer rates between the DNS (firmware 1.03) and my Windows+Ubuntu dual-boot laptop. In windows, the transfer rate is around 9MB/sec (via a Linksys 100Mbps router). In Ubuntu, I tried to mount the DNS share folder as both smbfs and cifs, and I found the transfer rate is about 4MB/sec for smbfs and 10MB/sec for cifs (Sometime the system monitor even reports 11MB/sec), which is faster than what I got in Windows. I don't know if this is just some random fluctuation.
Offline
In my case, NFS is much faster than samba. I tried to copy over hundred G of files (pictures each around 2M and movies each around 700M, a few file from 3G to 5G) to DNS323 though samba, and cannot wait that long, so I switch to NFS and much faster.
Offline
fordem wrote:
Out of curiosity - how do you know it's not the gigabit interface - or for that matter, the disk interface?
I don't - just guessing :-)
It doesn't matter much, does it?
Cheers,
Emacs
Offline
I am wondering if there is anyway to enable jumbo frame on the DNS? I saw quite a few benchmark of other brand NAS and it seems jumbo frame in some case will double the transfer rate. Is there anyway we can hack the firmware and increase frame size?
Offline
dickeywang wrote:
I am wondering if there is anyway to enable jumbo frame on the DNS? I saw quite a few benchmark of other brand NAS and it seems jumbo frame in some case will double the transfer rate. Is there anyway we can hack the firmware and increase frame size?
Whilst I would certainly like to know if it can be done - I don't think that enabling jumbo frame will result in significant throughput enhancement - out of the box the unit cannot saturate a 100 mbps connection, my best on 100 mbps has been around 70 mbps and on gigabit has been roughly 150 mpbs, on a network that is capable of handling colose to 98 mbps on 100 mbps, and over 300 mbps on gigabit, without jumbo frame.
Offline
zeroday wrote:
what is the best software to measure?
I use hardware to measure - my wrist watch :-)))
Copying a large file and getting the time used for that.
Cheers,
Emacs
Offline
Copying a couple of gigs worth of files on Vista machine to the NAS, Vista shows a transfer speed of around 9.2 MB/sec. It fluctuates a bit, but that is about the average.
Note - I have a 100 Megabit router.
Last edited by dan.crouthamel (2007-08-27 16:34:03)
Offline
fordem wrote:
dickeywang wrote:
I am wondering if there is anyway to enable jumbo frame on the DNS? I saw quite a few benchmark of other brand NAS and it seems jumbo frame in some case will double the transfer rate. Is there anyway we can hack the firmware and increase frame size?
Whilst I would certainly like to know if it can be done - I don't think that enabling jumbo frame will result in significant throughput enhancement - out of the box the unit cannot saturate a 100 mbps connection, my best on 100 mbps has been around 70 mbps and on gigabit has been roughly 150 mpbs, on a network that is capable of handling colose to 98 mbps on 100 mbps, and over 300 mbps on gigabit, without jumbo frame.
I would say a 10-20% increase is achievable if we can find a way to enable jumbo frame. Just look at this benchmark for the LinkStation Pro:
http://www.smallnetbuilder.com/content/ … 11/75/1/4/
The LinkStation Pro also could only reach about 10Mbyte/sec(about 80Mbps) with 100mbps network, but it doesn't prevent it from having a 20-30% speed increase under Gigabits network with 4k jumbo frame.
Last edited by dickeywang (2007-08-27 06:41:50)
Offline
Hi,
Have my DNS323 connected to a gigabit interface on my Dlink DIR645 router.
It has two 500GB disc in Raid 1 configuration.
While using ftp on my mac I get for both read and write (copying DVD images) ~18000 to 18500 MB/s
Using smb/cifs connection this drops to 14000 to 15000 MB/s
Seems like the 18500 is the max that the box can handle. Not sure if enabling jumba frames is going to make any difference.
Unfortunattely the smb/cifs seems to be more reliable then the ftp. While making a backup of directories with FTP lots of files my connection gets reset and I am loosing data. Not sure if this is the FTP program or something else. Using smb/cifs does not give these problems.
Regards,
edit: I use firmware 1.03 the ftp problems where also in the 1.01 firmware. Upgrading did not solve them but made them less frequent.
Last edited by Speijk (2007-08-27 15:36:18)
Offline
Pages: 1 2