NASLite Network Attached Storage

1. improve performance
2. Improve Performance
3. IMPROVE PERFORMANCE

Oh, and did I mention, IMPROVE PERFORMANCE

I have built a new NASLite2-USB box running v2.02 with a P4-3.0 processor, 512MB ram and a new AOpen i865Gm-IL mobo with built-in GB lan. I am very disappointed with the throughput speeds I have been able to achieve. It is significantly worse than I can get going to another box I have that is a P4-1.8, 384MB, PCI GB lan card, running Windows 2000 Workstation.

The PASMARK disk dest, run on my main computer and accessing the NASLite box and the Win2000 box as mapped drives, gives:

Win2000 sequential read 22.9 MB/sec
NASLite2 sequential read 12.1 MB/sec

Win2000 sequential write 32.6 MB/sec
NASLite2 sequential write 22.4 MB/sec

Win2000 DiskMark score 153
NASLite2 DiskMark score 102

Win2000 machine: 2/3 the hardware, running a bloated M$ OS, still gets 50% better disk throughput that the latest NASLite2-USB v2.02!

?????

.

Hi lmoseley,

I've done extensive benchmarking of NASLite v1 and v2 against just about everything out there and my results are not quite the same, could you please, in detail, tell me step by step on how you came to these benchmarks.

Please provide your network topography and the hardware used as well.

Thanks.

This is difficult not to comment on...



The above statement is pretty ignorant at best. I'm sure Ralph will have plenty to say on the subject, but for argument's sake, let me bring the following to your attention:

With SMART enabled, NASLite-2 launches a complete hard disk SMART self test that starts at boot time and continues for a duration that varies based on disk type, size and firmware. That alone can often reduce performance to as low as 20%. In addition, if you are to compare performance, you should do that on the same hardware and not on different machines. Just because you are using all new components for your NASLite machine, that is no guarantee that they all work well together.

That said, your inital comment about performance is a bit out of place. NASLite is all about performance and has been since the beginning. We take great care in making sure that we move as much data in or out of that NIC with as little hardware effort as possible.

In your "benchmarking", if all things are equal, NASLite will hold it's own pretty well. If it didn't, Ralph and I will be out of a job

As I have mentioned in my other current thread, although SMART should be on, it apparently is not, so this is probably not a factor. Also, after the NASLite box starts up, the disk activity light remains out, which argues against any disk test going on in the background.



Probably so, but that isn't really practical. But, in order to install Win2000 on the NASLite machine, I'd have to reformat one of more drives. And to test NASLite on the Win2000 box, the same is true. There aren't really any "components" to not work together in the NASLite box... motherboard, CPU, memory, drives. No add-on cards, etc.



It doesn't seem unreasonable to expect a P4-3.0 with 512 MB and mobo GB lan to outperform a P4-1.8 with 384 MB and a PCI GB lan...

Ok, Apparently you are severely confused about benchmarking "network" filesystems, I would suggest you do a little research and maybe you'll understand why your comment above and your rant is "unreasonable".

I don't know about you but there are a couple of things that I would think about before I was to declare a winner in this battle.

First of all the processor doesn't matter much at all. I get just as fast performance from the old 400MHz Celeron as I do now from an 800MHz Athlon. Back in 95 when Network appliance was fairly new they put out a machine that used a pentium 200 (non MMX) and it would easly flood almost three 100Mb network links. Your claim of the faster processor hardware performing worst than the slower is questionable and rather moot. Another thing that I keep in mind when evaluating Non MS based servers running Samba is that MS does a fine job of trying to break or at least make the link from a MS based box run slower if it is accessing a non MS file server.

One more thing that comes to mind is are you running a legal copy of 2000 server and do you have the license for all the clients to be connected to it to truly test it's performance against NASLite2.0. Or are you just running a pirated version so you can run your suck.

Come back when you have read about benchmarking and remember that there are ATA, SATA, FC-AL, SCSI, and SAS shoes and drive recievers out there for cheap. Try that and use the exact same hardware for the testing and then we may be open to listening to you run your suck again.

Flawed testing doesn't hold much weight around most forums except those at MicroSoft.

Mike

I put a intel 1000GT in my NAS box to beat up on lmoseley, but now I think he's got a point.

For me, XP (as SMB server) is 'writing thru' way faster (twice as fast) on a large file transfer.

There are many ways to measure network file performance, but in a low volume, backup box scenario... gigabit performance isn't stellar (for me) if my files or file sets are bigger than my NAS RAM. (on Windows copying to NAS).
I've got a LSI IDE card coming in, so maybe that'll change things.

As I clearly stated above, the second machine is running Win2000 Workstation (not Server). It is an older machine and is running a LEGAL copy of the OS, not that that should matter. Are you suggesting that a pirated copy is somehow FASTER than a legit copy?

I ran the test in a home office environment. When I ran the test, only the XP box (where the test software was running) and the particular server being tested were running (ie, no competing use of the network). The three machines are both directly connected to a gigabit hub.

I understand the comments regarding running benchmarks on identical hardware. In a lab environment, I would do that. Rather, what I am doing is comparing two real-world computers which I think should show AT LEAST comparable performance, but do not.

So how much latency do you think your hub adds to your poor performance?

Maybe your benchmarks should have been titled NASLite-2+Hub vs Win 2000+Hub? Why don't you try and run a cross over cable from both test machines nic to nic and see some performance gain.

Among other issues, I've not seen any mention of rebooting the test boxes in between tests (unless you want to give a *caching* advantage to someone) and a host of other things.

Considering the bottleneck of a NAS system is the nic card and the harddrives, I think it's pretty unfair to test NASLite against another solution not using the same two drives and nic's etc.

We appreciate constructive criticism, but quite frankly, lmoseley has given me no legitimate data or precedure here to warrant anything he claims to be dismal performance other than a poorly executed test precedure.

The Intel 1000GT card now works with NASLite?

I for one would be very disapointed if the Win2000 to WinXP box link were not to be faster than the NASLite2.0 to WinXP solution. Microsoft has the advantage of having the code and being able to tune it for the best transfer rates. The Samba group get the advantage of working in the blind and playing catchup all the time as MS moves the target around to maintain advantage.

Another point I will make is that the development team here is 2 guys releasing a VERY compact NAS solution with limited drivers. Tunning is also likely minimal for such things as the ATA or SATA drivers as well as the NICs. I would surmise that if they were to have say 10 talanted coders and developers versed in the the inner workings of the windows SMB code and give them a few months to tune up the code that performance would be far better than that of Win2000.

As far as a hub being a problem I would be willing to argue that it has little effect on this one. He is likely using it for both tests so any delays are common to both runs. Also I doubt that he is using a HUB as I don't think that there are and Gigabit HUBS. He is using a switch, correct?

As to your copy of 2000, Pirate or not they both perform the same. A silly extraction. My reference was to the fact that you pay for true server function in the MS environment on a per user basis. We get around this with NL.

I for one trust NL alot more than I do Win2000 for the function that I use it for. If I take a bit of a performance hit for my choice then so be it. I also use NL because it is not contributing to the MS juggernaut.

So again your argument, at least for me, is moot. the performance of the product is F-IN great considering all they are trying to do with it.

Mike

As far as the hub/switch, during my benchmarking of NASLite-2, my performance nearly doubled when going with a crossover cable direct, from 35mb/s to 90mb/s. Clearly shows the latency in network topography.

Anyone who has crossover cable can verify my findings.

I believe that the culprit is not Naslite V2. It is probably the WinXP client.
Here some more information to make my point.

BTW I am quite happy with performance but a couple of things have caught my eye. I run a 2.01 Naslite box with Gigabit LAN card BUT on a 100Mbit LAN and with a 3ware RAID card for a RAID5 2 TB array. This is my media file server.

I also have a WinXP PC from which I have recently been copying large multiGB files to the NASlite box.

I use 2 methods for copying files over. The first is using Win explorer using copy paste or even a DOS copy command. With this method I get about 8-9Mb/s writes to the NAS box. Network Utilisation from the Task Magager on the WinXP box shows around 86%. Clearly the network is not being flooded by the WinXP box.

The second method is using a program called HDTVtoMpeg2 for copying and joining by .TS files into 1 file on the NAS box.

This manages to write to NAS at about 10.75 Mb/s. Network Utilisation from the Task Magager on the WinXP box this time shows around 96%. Clearly the network is in this case flooded or almost by the WinXP box and I am almost certain that if I moved to a Gigabit switch it would go much higher.

It seems to me that HDTVtoMpeg2 uses a more efficient method for writing.

To me NASlite is not the culprit but rather the way WinXP utils write to the network.

Hi,
during reading all forums letters depending performance what hardware to use I noticed that all people name boards cpus and harddisks. Nobody talks about the nic and the cables between. Performance depends in "all about" will say: if you take 1GB 64bit nics a 64bit sata2 raid controller high-performance harddisks for the nas you still have to check cables switch patchpanel and connectors for cat7. And don`t expect lowcost hardware to reach 1000Mbits at any point of this chain - even when it`s sold to perform like 1Gbit.
I tried a rtl8139 and intel 100 - that was 2.5 times faster!
That I experienced with nas until now and I`m shure that software is the last to check.
I`m looking forward to a sata2 64bit controller, but I have to wait until Tony get`s the driver in the kernel. (3ware 9xxx).
Hope I have done no mistakes and when buying a bike everybody knows that there will be no way to speed than stroke volume...
Roland

Now that is strange! Can enyone confirm that the 1000GT works. To the best of my knowledge, the NASLite-2 pro/1000 driver does not support the GT