NASLite Network Attached Storage

I am starting this thread with the goal of getting some hard, cold facts about NASLite servers and their performance. In specific, what is your DETAILED hardware configuration and what is the real world performance of NASLite running on said hardware?

To this end I put forth the following:

Provide the Make and Model of motherboard as well as the processor and any configuration changes you might have made. IE. is it over or under clocked, is the Cache off or is it write back or write thru?

The amount of memory, bus speed if you know it?

The network interface card you are using or the NIC chip the motherboard has built in?

The RAID card you use and details like the amount of cache RAM on board, that is if you have one?

The drives in the box, be specific about make, model, speed (RPM), cache, interface type (ATA, SATA, SAS, FC-AL, SCSI, and the speed the interface is used at IE. The drive may have an ATA100 interface and be plugged into an ATA33 interface?)

The power supply make and rating (how many Watts?)

Are you on an UPS and/or power filter?

What does your network look like? IE. The switch/s you use and of course the speed of the switch.

Some basics about the client machine used to gather performance info about the the NASLite server.

Also include the number of days that your sever has been up for and the GB of transfer it has had.

This thread is not intended as a discussion forum for the merits of one processor or NIC over the other. It is simply for the posting of hard data so others looking can try and find the sweet spot, hardware wise for the best performance vs. price for a NASLite implementation. Please keep this in mind as you post.

Here is the software that Superboss used, seems simple enough and the results will be relevant between test and systems. I downloaded it from the following site and just used the same command line as SB did so my numbers would compare. Please use the same command line parameters so we can get a true picture of performance.

http://www.xdt.com.au/Resources/Downloads/

I will amend this thread with further links to software that is both free and easy to use to benchmark the performance so we are all on the same page as far as numbers.

My setup is based on an Intel D845EBG2 motherboard with a P4 running at 2.0GHz on a 400MHz FSB All devices not needed are turned off that serve no use. ACPI is on I think.

Memory is 512MB of generic Kingston DDR 400 RAM, 4MB is used by the on board video.

NIC is a 3Com 3C985SX running full duplex over Fibre to my switch.

Drives are as follows and each has a Antec 3 fan slot cooler blowing over it, temps all run in the 25-28 degree C range:

Disk-0 shares the bus with the Boot CDROM and is a Maxtor STM3500630a V.3.AAE Firmware, 500GB 7,200RPM, 16MB of cache and an ATA133 interface it sits on an ATA100 bus and has a total up time of 3565 hours.

Disk-1 is on the secondary ATA bus with Disk-2, it is a Seagate ST3300831A V.3.03 Firmware, 300GB 7,200RPM 8MB cache and an ATA100 interface on an ATA100 bus. Uptime is 13,413 hours

Disk-2 is a Seagate ST3200822A V.3.01 Firmware, 200GB 7,200RPM 8MB cache, and an ATA100 interface on an ATA100 bus with Disk-1. Uptime for the drive is 6,878 hours.

Power supply is a generic Deer electronics 300Watt ATX with standard cooling.

Power is filtered by a Tripp-Lite IsoBar premium power stripp, no UPS right now.

The network consists of the NAS going into a Allied Telesyn AT-9410GB Managed Gigabit switch into one of the GBIC ports via Duplex Fibre. All Gigabit connections are handled by this switch. Out of the second GBIC port via Fibre to the SFP port on a Dlink DES-3350SR switch for any 100Mb connections. The second Gigabit port is open. Cable runs are all short and all are just CAT5 or CAT5e with a Leviton 24 port patch bay in the rack and Leviton CAT5e connectors in the wall.

Here are my runs with it:







Test machine is a Dual Opteron285, 10K SCSI, FC-AL RAID0 W/5 15K disks, 8GB RAM with Nforce Gigabit NIC and Vista Ultimate. The performance of the array in my machine under Vista using software RAID0 is shown in the following links. The first is using the same test we use on the NAS box over the network but with 1000 frames there was an problem, most notable is the fact that the data rate is far to high on the read The second test is the same but with a total of 10,000 frames rather than just 1,000. The third is the same as the first but again with 1,000 frames at HD resolution.







I do love fibrechannel and 15K drives!!!! Just have to find a hardware RAID card now.

Mike

Motherboard: ABIT KT7A-RAID (flashed with custom bios to allow disabling of ACPI) pretty much everything disabled in BIOS.
PSU: 300W noname with 120mm fan, works fine.
CPU: AMD Duron 800MHz
RAM: 1GB PC133 SDRAM
Raid-card: LSI Logic Megaraid SATA 150-4 (64MB cache and I think write-back is enabled, will have to check)
Drives: 4x Samsung Spinpoint P120 250GB in RAID-5 (no hotspare, all in one array)
Network card: Intel PRO/1000 MT
Network cables: CAT6
Switch: NetGear GS105, 5-port gigabit switch

The box is not hooked up to any power filter or UPS. The uptime is <24 hours as I turn it off every night (my electricity bill was absolutely MAD last few months).

These are the numbers I get when running diskwriggler between my NASLite and my workstation (Intel PRO/1000 GT with Western Digital Raptor drives running Windows XP Pro SP2):



I'm a little dissapointed with these numbers and would like to identify the bottleneck but have been unable to do so..

I would be allot dissapointed with the numbers you got, the RAID5 should perform at least 50% faster than what I got even when taking into consideration the extra reads and writes involved. My guess is that the problem lies in the hardware of the RAID card and maybe drivers more than anything else. Try and flash the card with the latest firmware and set the cache to writethru. Also try moving it to another PCI slot.

Mike

NASLite v2.06 tested from WinXP/HE 3.2GHz P4 (Prescott) with GigaBit PCI card
Gigabit LAN, Cat-6 cabling
US Robotics 8-port GigaBit hub (some PCs are 100mbps)
From System Status Page
CPU MODEL VENDOR MHZ BOGOMIPS CACHE
i686 Mobile Intel(R) Pentium(R) 4 - M CPU 2.00GHz GenuineIntel 2395.973 4784.12 512 KB (Note that this Mobile chip is overclocked from 2.0 to 2.4.)
FSB is 800MHz
1GB RAM (DDR PC-3200 ECC unbuffered)
SuperMicro P4SCE motherboard
Disk-0 IDE 200GB Seagate (8MB cache READ-ONLY) (no testing)
Disk-1 IDE 300GB Maxtor 6L300R0 (16MB cache)
Disk-2 SATA 500GB Seagate ST3500641AS (??MB cache - not shown in ServerMessageLog)

I'll edit this post with more details later ... but for now, please explain the following so that I can provide meaningful results ....

J:\xxx\diskWriggler-1.0.1.win32>try.bat 1

J:\xxx\diskWriggler-1.0.1.win32>echo OFF

Running DiskWriggler on Disk-1
Frame Resolution : 720 x 486
Bytes / Pixel : 2.00 (packed) (8 bit, 4:2:2 sampling)
Frame Size : 699,840 bytes
File Type : All frames in one container file (699,840,000 bytes).

Writing 1000 x 699840 byte frames into file \\192.168.1.24\Disk-1/framestream.dat
389 frames in 5.00 secs ==> 77.80 FPS (51.93 MB/s)
262 frames in 7.27 secs ==> 36.06 FPS (24.07 MB/s)
349 frames in 4.13 secs ==> 84.61 FPS (56.47 MB/s)

Write Summary : 1000.00 frames in 16.39 secs
: 36.06(min) 61.01(avg) 84.61(max) FPS
: 24.07(min) 40.72(avg) 56.47(max) MB/s

Reading 1000 x 699840 byte frames from file \\192.168.1.24\Disk-1/framestream.dat
read_frames:read_frames: CreateFile failed: 32.


J:\xxx\diskWriggler-1.0.1.win32>try.bat 2

J:\xxx\diskWriggler-1.0.1.win32>echo OFF

Running DiskWriggler on Disk-2
Frame Resolution : 720 x 486
Bytes / Pixel : 2.00 (packed) (8 bit, 4:2:2 sampling)
Frame Size : 699,840 bytes
File Type : All frames in one container file (699,840,000 bytes).

Writing 1000 x 699840 byte frames into file \\192.168.1.24\Disk-2/framestream.dat
269 frames in 7.03 secs ==> 38.26 FPS (25.53 MB/s)
292 frames in 5.00 secs ==> 58.40 FPS (38.98 MB/s)
107 frames in 5.08 secs ==> 21.07 FPS (14.06 MB/s)
301 frames in 5.00 secs ==> 60.20 FPS (40.18 MB/s)
31 frames in 0.39 secs ==> 79.36 FPS (52.97 MB/s)

Write Summary : 1000.00 frames in 22.50 secs
: 21.07(min) 44.44(avg) 79.36(max) FPS
: 14.06(min) 29.66(avg) 52.97(max) MB/s

Reading 1000 x 699840 byte frames from file \\192.168.1.24\Disk-2/framestream.dat
read_frames:read_frames: CreateFile failed: 32.

J:\xxx\diskWriggler-1.0.1.win32>

I've already tried every slot in the machine I'm afraid and the BIOS is flashed to latest version. I've also tried switching from write-back to write-thru without any major difference. I believe is the same model / make (perhaps different amount of ports) as Ralph used for his benchmarks. No IRQ conflicts or nothing so I'm out of ideas. Guess the only thing I havent tried so far is to replace the cpu + motherboard....

You might just be hitting a hardware problem with the LSI card, I hear that they are not the best of performers, I have an ATA100 model that I am going to give a try to this weekend and will get back with you on this. You're no moron so I figured that you had done all those things. You might try a different NIC as a last resort, I personally HATE the realtec based cards and dodge them like the plague, 3Com or Intel all the way!

Mike

Thanks Mike, I've tried multiple NICs already with no luck and the current one is an Intel PRO/1000 MT server adapter so should be more than sufficient. As I said earlier, I believe Ralph even uses the very same raid adapter so its not likely to be a compatibility issue but obviously there could be something broken with my specific card (although it doesnt seem likely as it is working, not just as fast as one would expect).

To be honest I'm thinking of ditching this box altogether and building a new one.. checked the local auction sites for some replacement hardware and quickly realised I can basically get brand new stuff for just a little extra. The downside is new hardware is more likely to lack proper support in NASLite so I might even convert to the dark side and go with a distro (Ubuntu Server) that support newer hardware. I'll wait with replacing the raid adapter until I have tried it in the new box.

Is your LSI card in a 32 or 64 bit slot?

Obviously performance will wane with the card in a 32 bit slot.

I will post some diskwriggler numbers here today if I get some time to my NL2 box. Also worth noting, latency in networks is huge, try benchmarking with a crossover cable from the client machine to the NL2 box, you'll be surprised at how much your losing in speed.

Thanks Ralph, I appriciate it. The card is indeed running in 32bit mode as the motherboard has no PCI-X slots. Even in 32bit mode it should be able to generate higher numbers, the drive in single mode can provide much higher numbers. On my setup there is 0 impact using the switch, I have tried with direct connection using a CAT6 crossover. I suppose this would indicate a bottleneck within the NASLite box.

It should, there's a performance hit when using raid 5 with any system.

I had mentioned that as well but with 4 drives spinning his performance should be at least 50% higher than mine even when you take into consideration the fact that RAID5 has two read/write cycles per block stored. I would expect that he get around 37-40MB/sec sustained sequential block access from his array.

It will be interesting to see the numbers your machine gets.

Mike

Spent a day trying to find a suiting setup for my new file server and ended up with the following:

Asus M2N32 WS PRO
AMD Athlon64 X2 Dual-Core 4000+
ANTEC SONATA III 500 BLACK MIDITOWER
Kingston DDR2 1GB PC2-6400

With this setup I can re-use my LSI card with increased performance (this board has 2 PCI-X slots) while staying with NASLite. I know the CPU is totally overpowered but it was really cheap Will post new numbers when the new box is built.

The dual processor will be good for the future, once we move to the 2.6 kernel we will support multiple cores/cpus.


Also, when benchmarking, something I don't see too many people think about, is the client machine your benchmarking from, this plays just as important role.

Realistically, the only real way to benchmark these types of servers is with identical hardware on both ends, or better hardware on the client side.

I think a machine with two Dual core Opterons running at 2.6GHz and 8GB of RAM should be more than fast enough to cover both read and write tests

Forgot to add, the drive I use for storage in the Opteron machine is a RAID0 array off of a Qlogic QLA-2300 made up from five 15K rpm 2Gb FC-AL drives.

How much longer till you guys roll out NL with a 2.6 based kernel? Are there going to be any new features in the package?

Mike

True the client machine is seldom mentioned when posting benchmarks. In my case I have used an Athlon XP 2600+ (1GB Ram, Intel PRO/1000 GT and raptor drives) as well as a Thinkpad T43 (Pentium M 750 1.86GHz 1GB Ram).