NASLite Network Attached Storage

After a few struggles with kickers, BIOS- and jumper-settings, two NASLite-2 boxes are operational now, the mapping is being updated and storage capabilities are greatly expanded from NL 1.4 in my network.

I bought a Netgear Gigabit network switch (GS108) and one bit of info on the datasheet made me curious in my need for speed:
"because Gigabit Ethernet is a full duplex standard, you get up to 2000 Mbps on each port. The GS100 Series provides Jumbo Frame support up to 9,000 byte frames."
(Some PCs and the two NAS boxes here are equipped with the Intel Pro/1000 MT desktop adapter. )

Not being an expert, I am asking myself now:
How does the given NL-2 Linux configuration fit in this picture - does it support the full duplex standard and Jumbo Frames?
And what can I do for the settings on the WinXP PC end to maximize performance in terms of network transfers?

Any insights from the experts are greatly appreciated! TIA, Thorsten

Thorsten,

If my tests are at all representative, a NASlite server is dramatically slower than what Gigabit ethernet can handle. The speed of your switch has essentially zero impact.

Gigabit is theoretically capable of 125 MB/sec. You won't get anything near that. In similar setups with a fast IDE drive, NASlite serves at about 6-10 MB/sec; while FreeNAS does 20-23 MB/sec. I don't understand why NASlite is so much slower, but in either case the speed is still far below the maximum. (for comparison, I've got an old Mac G4 server that delivers 35 MB/sec.)

I agree you will not get anywhere near Gigabit speed, however, my moderate setup with NASLite 2.0 is getting about 18-20MB/sec not 6-10MB/sec as you report. My speeds doubled after updating the Marvell Miniport driver on my XP machine, so its not only the NAS which has an effect on transfer speed its also the machine you are transferring to/from.

Thanks for the feedback.
How do you reliably measure the transfer speeds? "Eyeball", i.e. take a 300 MB file and clock it on your stopwatch, or are there any utilities that will establish and test this? I have read of diskwriggler, but did not get far with it.
Anybody else with experience in transfer speeds for NL-2?

For rough measurements, I use a Mac tool called MenuMeters, that gives realtime data. But for critical measurements I time the copying of a 10GB file, and a 10GB folder full of small files (the speeds can vary a lot depending on the size of the files.)

The stop watch method is good enough.
There is also the LAN performance monitor on WinXP task manager.

I currently use Naslite+, not V2, on a 100Mbit network and on a motherboard that only supports UMDA33 for each disk. I get about 7Mb/s reading from just one disk and if I start a copy from another disk in the same box I get to a total of 10Mb/s overall so it is evident that NASlite can utilise my 100mbit LAN almost fully. I too had initial problems with the marvel LAN driver on the WinXP box.

I believe that with NASlite 2 and reading from UMDA133 or SATA disks or even better from a RAID card, speeds will be much higher....

I am in the process of setting this up and will report back.

Just setup V2 on my hardware with an adaptec 2400A controller and a 838Gb RAID5 array.

A transfer of 177Gb took 6hours and 18minutes translating to a transfer rate of 8Mb/s which is pretty good on a 100Mbit network.

Hello, I now did some clocking too:
49000 MB took 182 minutes to copy with bit by bit verifying, using Allsync, on a Gigabit Network. If my math is right, this translates to 4.49 MB/sec. Not exactly breathtaking, but the verifying may have slowed things down considerably, so this may not be a valid comparison.

Are there any suggestions to improve things on the "Windows end" in terms of settings for the NICs? (I guess the Linux side can't be messed with anyways.)

Search the forum for Disk Wriggler. It is an easy tool to measure speeds. Data verification will slow down the measured transfer rate.

The verification slows down things a lot.

From experience I can't justify its use. It's not like burning a file on DVD or CD....

I havent seen bit by bit verification used at all.

So I skipped verification by now, since I am working in a stable network, but the question remains, what might be done on the windows end of configuring the NICs to reach an optimal performance. Namely such settings as for Jumbo-frames and full duplex standards.
Since I can't make changes on the NASLite end, I at least envision having the "windows end" be confirgured for best performance and matching.

Any ideas anyone?

Well, I am still unsure about the optimal Windows/ NIC settings, nobody seems to know here.

I have clocked some more transfers (of 200 and 150 GB, without verfication) and came up with a calculated throughput speed of around 6.6-6.8 MB/ sec for the transfer. Still not exactly breathtaking...

Meanwhile I also looked for some more benchmarking tools, since I could not figure out how to use diskwriggler without suitable video files at hand, and found SANDRA from http://www.download.com/SiSoftware/3260-20_4-6284746.html, which offers benchmarking as well.

Here, I am benching and average of 36 MB/ sec throughput with a local drive, and 19 MB/ sec (averaged) for mapped to NASLite-2 Network drives (both using the "Windows cache option" switch UNchecked). Whatever that's worth.

Ralph has been benchmarking NASlite-2 since early beta, so he has a much better handle on how to properly conduct such tests. I think he plans on submitting some numbers in the near future, but I’m not sure about the details.

Benchmarking results were instrumental in our development process and did expose some startling latency numbers in common networking hardware such as hubs and switches. Even high-end managed switches tended to introduce huge latency in transfers.

When he’s ready, he’ll try and shed some light on that issue.

Great, that would be helpful! Since things can't be manipulated on the Linux end for us end-users, I would want to be sure to optimize the NIC on the other end, to get the best possible match.
Thanks for the hint, am looking forward to Ralph's results/ input/ instructions.