True Gigabit NAS?

[kjcdude]

I'm looking for a small to medium business nas device that can deliver true gigabit+ speed.

I'm using 10,000rpm wd raptor drives in raid 5 and here are my results on two devices i've used so far.

On both devices i've configured them for raid 5, the network connection is over 802.3ad via a gigabit network that can sustain full gigabit with no problem. I'm transferring around 10 gigs of data to around 25 clients.

Thecus N5200B Pro

QNAP TS-639

At the beginning you will see a green line that indicates the upload to the device, the green/blue lines from then on out is the outbound traffic.
The TS-639 made it over a gig for a second or two, but wasn't able to maintain that type of data transfer speed.
The system stats at the time of transfer was showing the CPU to be at 100% for almost the entire duration of the transfers.

Thanks

 

[kjcdude]

I went ahead and ran some more tests and was extremely please/confused with the results.
The test was identical the one ran above, but produced a much better transfer speed.

CPU data from the qnap interface. Time frame: last 3-5 minutes of the above graph.

 

[thiggins]

The performance you get is highly dependent on how the files are being tranferred.

In the Fast NAS series, I kept hitting a wall at ~ 70 MB/s no matter what I tried in the way of hardware, drives, NAS OS. I found that the block size used in the network transfers was the limiting factor. The use of > 64 KB block sizes by Vista SP1 file copy was the key.

Supporting multiple clients will also limit your speed due to drive seek times.

Bottom line, you're not going to get sustained 100+ MB/s performance to 25 clients from any consumer / prosumer NAS.

 

[kjcdude]

What would you suggest would be best for my situation.

I need to transfer anywhere from 500mb to 10 gigs to each client with around 25 active clients at a single time.
I can support 802.3ad.
I have a decent supply of 10,000 rpm raptors that I can use.
I have chasis, mobos, ram.
Clients are xp pro sp2

Can anyone suggest a solid raid card and network cards that I can use to create a distribution server that can move 1gb/s+.

 

[etolsen]

Relative to disk transfer speeds, the network will most always be the bottleneck; as surely the transfer speeds of your disk subsystem are far greater than the 1Gb/s network.

While not a network engineer, I've always been told never to expect more than 70% of the wirespeed bandwidth; Tim's results of ~70MB/s reflects this advice. Putting those limits aside for a moment, if you are in agreement that 100MB/s would be the absolute best throughput for a 1Gb/s network, then your goal is an additional 30MB/s.

The easiest method to gain the additional 30MB/s bandwidth is to add an additional network card to the NAS. Without first-hand knowledge of the two NAS products you mentioned, the how-to is not available.

Focus on the network bandwidth as the bottleneck and you'll find the solution you're looking for.

 

[00Roush]

An Intel PCIe based network card would be my choice. Either this or two of these. Also could try two of the desktop cards if you wanted to cut costs.


00Roush

 

[pbls]

i have tle later desktop adapter you mentioned and i almost it the 100mb/s between the two dsktops with windows 7 and a DIR-655 in the middle,and i don't have any Raptor just a few WD 1 Tb green drives.:

 

[beisser]

atto diskbench uses relatively small files for its benchmark.

its easily possible that the data atto writes to the network share fits into the ram of the target computer. that would limit the speed only to the netwrok link and remove the harddrive from the equation.

if you really want to test it, reduce the ram of both computers to 512mb (to avoid caching) or use a benchmark utility that uses really big files (like iometer or iozone). if you let that one uses 8 or 16 gig files, the target system wont be able to cache it and you will see the true speed.

 

[thiggins]

What would you suggest would be best for my situation.

I need to transfer anywhere from 500mb to 10 gigs to each client with around 25 active clients at a single time.
I can support 802.3ad.
I have a decent supply of 10,000 rpm raptors that I can use.
I have chasis, mobos, ram.
Clients are xp pro sp2

Can anyone suggest a solid raid card and network cards that I can use to create a distribution server that can move 1gb/s+.

Are you simply moving files, or are the clients interacting with a server-based application? If just doing file copies, then move the clients to Vista SP1 or higher or Win 7. Sever/NAS should be Server 2008. This will provide the much improved file copy performance.

If files are small, then put as much RAM in your server as it will take. This will increase performance via caching.

 

[pbls]

atto diskbench uses relatively small files for its benchmark.

its easily possible that the data atto writes to the network share fits into the ram of the target computer. that would limit the speed only to the netwrok link and remove the harddrive from the equation.

if you really want to test it, reduce the ram of both computers to 512mb (to avoid caching) or use a benchmark utility that uses really big files (like iometer or iozone). if you let that one uses 8 or 16 gig files, the target system wont be able to cache it and you will see the true speed.

i have 2Gb of ram,the results above this are the same...

 

[beisser]

those atto-results are definitely cached, because no single harddrive (except ssds) can sustain that speed for a long period (i have raptors and they dont even come close to 100 mb/sec).

here is an atto disk bench i made on my readynas pro:

this is a drag and drop copy from my computer to the nas:

i had to use a ramdisk because at the time i did this test i didnt own a ssd yet.

these are all cached results though because my nas has 3 gb of ram..

i would need to reduce the ram on my readynas pro to 512 mb or so to get real uncached results.

what i mean is: of course you get super speed between your computers as long as the file copied is smaller than the ram of the target computer

 

[00Roush]

those atto-results are definitely cached, because no single harddrive (except ssds) can sustain that speed for a long period (i have raptors and they dont even come close to 100 mb/sec).

I respectfully disagree that a single drive is not capable of sustaining high speed transfers for files that exceed the size of the RAM on the server or client. With my current server (2GB of RAM) I regularly see file copy speeds of 90-100 MB/sec for files ranging from 1GB to 4GB. With my large test file that is 20.1 GB in size I see about 87 MB/sec for file copies to and from the server. This is with a single WD 320 GB SE16 data drive be utilized in my server.

Anyway I think we are off point here... I believe the original question was not about how fast a single transfer could take place but finding a NAS that could saturate a gigabit connection for hours on end with 25 clients requesting data at the same time. Iometer would probably be the program to use for this type of testing. If I have more time tomorrow I will look into what might be necessary to handle this type of workload.

00Roush

 

[beisser]

that is odd. then my raptor must be damaged pretty badly, because i cant get it to give me more than 65 mb/sec on drag and drop, no matter what i do.

i must admit, that the raptor is a good 4 years old and may be growing old. but even so it has to be faster than a 7200rpm drive (even if its a raid edition drive).

with my brand new ssd i get close to or above 100 mb/sec in drag and drop to my nas.

anyway as you have said, this is besides the point

 

[erack]

The performance you are seeing on your 4-year old raptor does make sense. Hard disk drives are very good at streaming performance, transferring very large files. When an HDD is transferring a large file it does not need to move its head about the disk much, so the performance is limited only by how much data is passing under the head.

Where SSDs can excel is random read / write performance. Since SSDs do not have mechanical heads to move when accessing different areas of a disk, they can achieve much higher performance for "random" types of operations.

The interesting thing about HDD performance is that streaming performance actually increases as the capacity of a single HDD platter does. So even though your raptor is operating at a higher rotational speed than a standard 7200 RPM desktop drive, a "new" drive bought today will have significantly better streaming performance. Even though the 7200 RPM disk is rotating at a slower speed than your raptor, way more data is flying under the head than on 4-year-old drives.

Where your raptor might still have some advantage over a new 7200 RPM drive is in access latency, which translates into better random read / write performance. Although with probably better use of NCQ on newer systems, a new 7200 RPM drive might win there as well.

 

[kjcdude]

Thanks for all the replys everyone. I've been busy and haven't had time to respond yet.

I'm working on some more tests with the QNAP along with a freenas box that i'm setting up.
Here's the specs of the freenas pc currently.

intel core 2 4300 @ 1.8ghz
2gb ram
built in nic
software raid 5
4 10,000 rpm wd raptors

Once i get these test done i'll have a good base as to where the freenas box sits in regards to the on board nic and software raid.
I can get the means to purchase a raid card and decent nics, but I need justification that they will improve performance before i make the purchase.

For the freenas, what type of format should I go with? On the qnap i'm running EXT3, on the freenas i'm running UFS currently.

 

[kjcdude]

Here are the results for an ATTO test from a client to the QNAP device.

 

[kjcdude]

I ran the same test as the original post on the freenas and got some very very weird results.

When it drops near the end down to near 0kbs there are actually still files needed to be transfered. It stoped transfering the files for a bit then started back up at a slow speed.

I'm running some tests on the drives now to see if any are failing.

 

[kjcdude]

Here's the results from another qnap test.