Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[fordem]

Gary,

I actually started posting a response based on a "super car" analogy, but didn't - I didn't think you'd appreciate it but since you bring it up ...

I've got two NAS here, the first is a Taiwanese made Encore Electronics unit, the second is my D-Link DNS-323 - the Encore Electronics is a single disk linux based NAS with a 10/100 mbps port, it's good for ~3.5MByte/sec tops, the D-Link on 10/100 will easily triple that 9~10MByte/sec, on gigabit maybe 18~20MByte/sec, and on gigabit with jumbo frame (9000 byte) another 50%, 27~30MByte/sec.

Let's turn these into cars - that Encore would be your Geo Metro and the D-Link maybe a Toyota Turbo Supra (the gigabit port is the turbo charger) - they can both run on the I75 and the Geo will struggle to make the speed limit, whilst the Supra will do it with ease - now let's put them on the Autobahn in Germany (where there is no speed limit), the Supra looks reasonable maxed out at 150 mph until some dude comes along in his Bugati Veyron and blows by at 250.

So ...

What sort of money did you spend - Geo Metro money?  Toyota Supra money?  Bugati Veyron money?  A Veyron by the way retails at USD$1.5M

[fordem]

Read very carefully - I didn't say you could get 50MB/sec from a DNS-323 (or DNS-321), I said the NIC can handle that, and that is based on personal testing.  Unfortunately the NIC is not the bottleneck so if you transfer a file from one computer to the DNS-323 (or vice versa) you're not going to see 50MB/sec, you're going to get the maximum speed that the bottleneck can pass - whatever that bottleneck is or wherever it happens to be.

By the way - yes - the throughput when backing up my IDM server is to the order of .5MB/sec and trust me that IBM cost me more than the DNS-323 did, and is capable of reading/writing a lot faster (the NIC there is good for 800 mbits/sec or 100MByte/sec.

Confused?

I did say the idea was to open your eyes, didn't I.

Transferring a data file from one disk to another across a network is really not that simple a process.

You start by reading data off of the first disk, passing it through a disk interface across an expansion bus and through a network interface, all in one device - then out on to the network, which may be as simple as a cross-over cable, or may be complex and include switches and routers and wireless - and then into the second device, through a network interface, through an expansion bus, through a disk interface and finally write to the disk.

Your transfer speed will be as fast as the slowest of the items mentioned.

Reading a single large contiguous file from a disk is fairly quick, but if that disk is fragmented and the head has to jump around to retrieve the data - a block here, a block there, the transfer speed is going to take a major nose dive, and the same thing happens if the data being transferred is a large number of smaller files (now you know why my backup takes as long as it does).

Where is the bottleneck?  It's hard to tell, but it's in the DNS-323/321 and probably the processor, which has to calculate checksums etc. during the transfer process.

Thanks for your reply. 

Ok, if I'm running a tad over 100 mbps, then it's gigabit performance?  If that's how it's definate, then I won't argue that point, but that is damn poor performance, don't you agree?  To me it's like saying I have a Porsche twin turbo... a super car, but for some reason I can't go over 60mph when I floor it.

Now, you're saying that I'm having issues elsewhere and I should be able to get "roughly 50 MB/s".  When I achieved over 20MB/s between two computers, I was using the same switch, the same type of wires, yet I'm only getting around 10MB/s with the dns-321.  I'll be happy to be around 20MB/s with the dns-321.  If i have a bottle neck what do you suggest I try? It's a simple network here.  I have two gigabit switches and I can swap them out but why? They are both brand spanking new and the switch indicates I'm in gigabit mode. The cables are cat 5e.  I can swap all these if you like, but the question is more than 1 forum member is experiencing the same issue.  I just verified last night I can get a transfer rate of about 3-4x faster than my computers equipped with 100mbps NICs.  I just can't do it with the dns-321.

You're saying with jumbo frames I should be seeing 30-35Mb/sec.  That I haven't tried yet, but a forum member here said that jumbo frame was good for 2-3MB/sec only.  If this is case, then I currently should be running at 27-28MB/sec and I'm not anywhere near that ballpark.

What size jumbo frame are you using? Unfortunately the disadvantage of the new gigabit nic card I have is that jumbo frames are limited to about 7K whereas most gigabit nics allow 9K frames.

Gary

[garyhgaryh]

Fordem,
Using your analogy with our dns-321/323, then there is no way your veyron will go over 60mph .
I like your supra analogy - I have a real old one that puts out 450HP , but if it was like my dns-321 it won't go faster than a geo.

There is no way our NAS will hit 27-30mb/sec for a big file - over 1 gig.  Correct me if I am wrong.  Even these guys

http://www.smallnetbuilder.com/content/view/30521/75/1/3/

couldn't get over 20MB/sec with large files.  They're stuck around 15-16MB/s.  So unless you run some sort of military spec dns-321 on a mil network, I don't see how you can get that throughput.  I'm not saying it's impossible, I just can't see how you can do it.

Are you running vista? If so can  you do a little experiment for me.  Find a large file on your NAS drive.  Perhaps an DVD iso image.  Drag and drop that from your NAS drive to your desktop.  When the copying dialog comes up tell me what it says for the transfer rate after it settles down (30-60 sec).  I'm just curious what your transfer rate is.

[garyhgaryh]

I did read that carefully.  You didn't say our NAS will go 50MB/s.  You said the NIC on these nas is roughly capable of 50MB/s:


Just in case you're curious the NIC in a DNS-323, which I believe is the same in the DNS-321 is capable of 400 mbit/sec transfers - roughly 50 MByte/sec - so there's a pretty big difference between what it can pass and what you have measured - might be an idea to find out why.

Read very carefully - I didn't say you could get 50MB/sec from a DNS-323 (or DNS-321), I said the NIC can handle that, and that is based on personal testing.  Unfortunately the NIC is not the bottleneck so if you transfer a file from one computer to the DNS-323 (or vice versa) you're not going to see 50MB/sec, you're going to get the maximum speed that the bottleneck can pass - whatever that bottleneck is or wherever it happens to be.

By the way - yes - the throughput when backing up my IDM server is to the order of .5MB/sec and trust me that IBM cost me more than the DNS-323 did, and is capable of reading/writing a lot faster (the NIC there is good for 800 mbits/sec or 100MByte/sec.

Confused?

I did say the idea was to open your eyes, didn't I.

Transferring a data file from one disk to another across a network is really not that simple a process.

You start by reading data off of the first disk, passing it through a disk interface across an expansion bus and through a network interface, all in one device - then out on to the network, which may be as simple as a cross-over cable, or may be complex and include switches and routers and wireless - and then into the second device, through a network interface, through an expansion bus, through a disk interface and finally write to the disk.

Your transfer speed will be as fast as the slowest of the items mentioned.

Reading a single large contiguous file from a disk is fairly quick, but if that disk is fragmented and the head has to jump around to retrieve the data - a block here, a block there, the transfer speed is going to take a major nose dive, and the same thing happens if the data being transferred is a large number of smaller files (now you know why my backup takes as long as it does).

Where is the bottleneck?  It's hard to tell, but it's in the DNS-323/321 and probably the processor, which has to calculate checksums etc. during the transfer process.

[fordem]

Gary,

I don't know if this is going to work - I've never used imageshack before - anyway, let's see how it goes.



If it works, it's going to need a bit of an explanation.

This is a composite of three screen captures assembled in MS Paint, because I couldn't capture all three images in one shot.

The background is an SNMP graphing tool called PRTG (www.paessler.com), it reads the bandwidth figures from my network switches, the one that's open is the graph from the switch port that feeds my DNS-323.

Directly below that is the output from a little utility called NASTester - if you search the DNS-323 forum, you should find a link to another site that you can download it from.  Basically what it does is create a test file, of a size that you choose (up to 2GB) on a local drive and then measure the time it takes to transfer it across the network to a mapped drive, it then calculates the average transfer speed and displays it - you can tell it how many iterations and it will loop as required and list each run and then give you the average, and then it reverses the two end points and repeats, so you get both write speeds & read speeds.

To the left of that is the Windows task manager displaying the network tab and the bandwidth that Windows is measuring.

From these three you can see here that with a 2GB file size, I can read approximately 250 mbit/sec from my DNS-323, achieving an average of 27.96 Mbyte/sec over two runs.  I have seen transfers averaging in excess of 30 Mbyte/sec and instantaneous peaks (reported by DUMeter) exceeding 35 Myte/sec.

There is nothing "mil spec" about my DNS-323 or my network - in fact, it's all entry level off the shelf stuff - a DNS-323 Hardware Rev A, firmware 1.06, with a pair of Seagate 250GB, Barracuda 7200.9 desktop SATA drives in a RAID1 configuration, an IBM xSeries 206 server (this was entry level three years ago, cost me all of $746),a desktop 3.0 GHz Pentium IV, with 1GB RAM, two 250GB desktop SATA drives - 1 x Seagate Barracuda 7200.9, 1 x Maxtor QuickView 6L250S0 - in a RAID1 configuration on the integrated Intel ICH SATA controller, integrated Intel PRO/1000CT network interface, running MS Windows Server 2003 standard edition.  These are connected by CAT5 (yes CAT5, not CAT5e, not CAT6) cables to a Netgear GS108T 8-port SmartSwitch - jumbo frame is enabled and set for a 9000 byte/frame size.

I'm sorry - I don't have a Vista system on hand - I'm still running XP Pro, a couple of my kids have Vista laptops, but nothing with a gigabit port, so I don't know that borrowing one of those would serve a purpose.

By the way - this is the same exact hardware/software/network that chugs along at 0.5 MB/sec doing a backup, so file size does have a significant impact on throughput.

[Freeman]

Wow, I come back and the discussion continues!

I think the technical back and forth can go on for quite some time but one thing can be agreed upon.

That the DNS 321 can transfer at gigabit speeds although at the lowest end if only briefly in some cases.

This is really kind of beside the point, though, isn't it? It's faster than the fastest 100Mbps network adapter. It's also as fast as it's going to get on a DNS 321 with its limited hardware, its an immensely underpowered little computer which is sufficient for a NAS device. Low powered CPU, small amount of memory...There are so many possible bottlenecks in there that it really isn't worth the effort arguing about it. It's not much competition comparing it to a modern PC or laptop that has far beefier hardware.

What it boils down to is price versus performance and in my opinion, other than the current buggy firmware which we all hope D-link will fix very soon it's currently the best NAS in its class. Try to find a NAS from another manufacturer that offers the same features at this price point ( D-Link, fix the bugs please! ).

I like the car analogy so I'll try my own, bear with me. I think it would be better to introduce a Honda Civic to represent a 100 Mbps card and the Porsche can continue to represent the 1000 Mbps card. Except you bought the Porsche for the cost of a Honda Civic ... or a geo metro. Now you're driving it at 140 mph and you think, damn, why can't I go at its maximum speed, 180 mph?! The Honda Civic, BTW, has a max of 120... it's all relative... and the Porsche is still fast.

Just out of curiosity, are you comparing the DNS 321 to the performance of any other NAS'? I'm just wondering if there are other gigabit NAS' with gigabit cards, at the time I bought my DNS 323, my first NAS, it was the only one with a gigabit port at its price.

BTW, I set my jumbo frames at 4k, I read in a review that it was the DNS 321 sweet spot and there was no real difference at 9k.

[fordem]

Freeman

Whilst I agree that these represent "the low end" of gigabit speeds, there is nothing brief about it - the duration of the time that I can hit the ~28 MByte/sec transfers shown is limited only by the size of the files and the available space - the largest file I've ever had on my DNS-323 was ~60GB, and it will sustain the rates shown.

[garyhgaryh]

fordem:

Running the NAS performance tester, I'm getting the following:

Dell D620 Laptop with gigabit docking station:
Avg (W): 12MB/sec
Avg (R): 9.41 MB/sec


Dell Inspiron530 with an inexpensive gigabit NIC:
Avg (W): 13.93 MB/sec
Avg (R): 9.47 MB/sec

My WRITES are much faster than my READS.

I just changed my network this past weekend and I go through two gigabit switches (both dlink DGS-2208) to get to my NAS.  Do you think that will slow my throughput?

If I start using jumbo frames, I think I'll be in your ballpark in the WRITEs, but your READs are amazing.  At first I thought perhaps the high performance writes was due to caching but that didn't make any sense.

About the "milspec" stuff, I was only being sarcastic .  You are running the intel PRO NIC and they do have rave reviews so I wonder if that combined with the jumbo packets are the two major reasons why you network is faster than mine .

As a newbie to gigabit networks, for me to enable jumbo frames, do I need to do anything else to my computers or is that automagically supported by the gigabit nics? Will this affect 100mbps networks with jumbo frames enabled or only gigabit networks? I have my network speed set to "auto" on my NAS.
I'll google jumbo frames right now.

Freeman:

Yes, i agree that performancewise we are at the _VERY_ low end of the gigabit speed.

[fordem]

Gary

I'm guessing that your laptop hard drives are the lower performance 5400 rpm ones - the fact that your write sppeds are higher than your read suggests that your laptop cannot write as quicky as it reads - you might also want to try defragging your disks it can make a major difference.

Jumbo frame only works on gigabit networks, you need network cards and a network switch that support jumbo frame and you'll need to specifically enable jumbo frame and in most cases set the frame size.  Some switches it'e either on or off, with no frame size settings, and some low end switches that support jumbo frame have it permanently enabled.

The golden rule is that the entire network path between the two devices using jumbo frame MUST support jumbo frame AND the frame size to be used.  You may see this statement corrupted to say the entire network, but that is incorrect.

You can mix gigabit with jumbo frame, gigabit without jumbo frame, 100 mbit, 10 mbit and wireless all in the same network and have

[peas]

I certainly appreciate fordem's input and the measurements he's taken time to capture.  One major fallacy with his assertions is he's measuring with the DNS-323, while this thread is about the DNS-321.  While they are similar in performance, they use different chips and have different firmware.  Controlled environs...

I have to agree with garyhgaryh here.  I've benchmarked the '321 on my Gb network and it tops out at 16 MBps (MegaBytes p/s).  No way it can hit 27 MBps when transferring to its internal HD.  The network isn't the bottleneck; I and others who've posted in this thread have performed numerous tests between other systems using the same network that the '321 sits on, and results have proven that the network can handle substantially more traffic than the '321 can push.  The limitation is the '321.  For its price I can't complain too much

An oddity I've noticed is that jumbo frames reduce performance significantly.  I have a D-Link Gb switch which claims jumbo frame support, and my PC's NIC supports jumbo frames.  I've tried all settings of jumbo frame sizes on the '321 and my PC.  None offer any improvement in speed, and most all jf settings reduce performance 10-30%.  This looks to me like an issue with the 1.01 firmware which added a UI selection for jumbo frames.  Hopefully they will improve jf performance in future firmware releases.

[fordem]

Hey peas

I'm not sure I would describe it as a fallacy - I made it clear from the outset that the figures came from a DNS-323 and the measurements were originally intended to illustrate the point I was making, that anything over the 12.5MB/sec performance possible with a 100 mbps network should be considered as gigabit performance, which I believe is apparent if you take the time to read the thread in its entirety

I'd also question the rationale behind calling it a fallacy when you yourself say ...

While they are similar in performance...

- isn't that exactly what my point?  I never said they were identical, just similar.

Last - I don't know if you saw this, but here's someone whose jumbo frame experience on a DNS-321 doesn't match yours - I'd say, if you can't get an increase in throughput when you enable jumbo frame, any increase at all, you're not doing something right.  My first foray into jumbo frame about eighteen months back was equally disappointing, but I've since had much better results, using exactly the same equipment - I never did figure out what I was doing wrong.

[peas]

I'd also question the rationale behind calling it a fallacy when you yourself say ...  - isn't that exactly what my point?  I never said they were identical, just similar.

I'd say we're splitting hairs here.  You're throwing around numbers like 27 MBps while people are achieving ~15 MBps. It adds to the confusion at the very least. Instead of pursuing the "I'm right you're wrong" angle, we'd all be better served by staying on topic and understanding the issue at hand.

I agree that any rates above 100 Mbps (12.5 MBps) indicate that the link speed is 1 Gbps and that the '321 is trying as hard as it can to fill that bandwidth.  It just has an underpowered CPU; no sense bashing people's network setups.  I know a bit about networks, from the TCP layer to IP to raw Ethernet frames.  I've modeled Ethernet's CSMA/CD architecture and worked on TCP termination. My network is kosher.  I can transfer between 2 PCs on my GbE network at approx 350 Mbps with jumbo frames. That's using Windows' inefficient copy-paste GUI.  I don't see an increase in throughput with '321 jumbo frames. In fact I get a slowdown.  That's a pretty clear indication something is wrong with the 321's jumbo frame support.  Sure there might be one particular config where jumbo frames might work well, but neither I nor numerous others have witnessed it.  Mind you, I'm using one of D-Link's own products for the GbE switch (DGS-2208).  I'm disappointed that D-Link hasn't tested more thoroughly with at least their own products.  Regardless, I've turned off jumbo frames and am satisfied with the '321 as it is.  I hope that D-Link will optimize the 321's performance, but it serves its purpose well in the here and now.

[LSDave]

I believe the DNS 321 is VERY limited in transfer speeds.

I have 2 machines running windows 7 and i can get sustained 70-80MBs but 15MBs max transferring to or from my Dlink NAS..

ill be selling it and building a FreeNAS machine with aggregated Gigabit nics.

[gunrunnerjohn]

ill be selling it and building a FreeNAS machine with aggregated Gigabit nics.

I considered that, but decided to try the Synology DS209, and my benchmark speeds came up to very usable numbers.