BTW, I just did some speed testing and experimentation.

 

My benchmark qualifies as simple, though certainly not ideal: time to copy a DVD ISO file from the Unraid to my main PC using simple windows copy/paste.

 

I am running 4.0 (unless there is a critical feature, I am a "two releases behind" kind of guy) with a five 300GB SATA-300 drive system.

 

* Stock = 3:30 for a 4.4GB file

* After the read-ahead tweak = 3:02 (http://lime-technology.com/forum/index.php?topic=965.0)

 

That calcs to moving from 21MB/sec to 25MB/sec - a 15% pop in performance.

 

I also did a write test, but only after the tweak: 8 min or 9MB/sec.

 

As has been pointed out elsewhere, there are a lot of "troughs" in the write transfer graph that indicate that the network is not the limiting factor.  If that were smoothed out, the write-results would have been closer to 15MB/sec.

 

 

Bill

Funny that this is coming up... Just last night I was wondering why I was only getting like 14-18% utilization on my new SATA NAS. This was moving data from one NAS to another, divided in half it seemed AWFUL close to 100megs. Sure enough it was plugged into a 100meg port. It was showing few troughs like I'd seen on my other NAS. I switched it to GigE and sure enough the troughs showed back up. <sigh> I'm hoping that some of this is just because I'm pulling from my slower IDE NAS. I'm honestly not 100% convinced that they issue lies with the NAS and wonder if perhaps some network tweaks might help smooth this. Are folks seeing the same troughs with say Linux or OSX transfers, Vista? Tried tweaking the network settings? I tried using Cablenut on my XP system but it would lockup each time I tried to write to the Registry - grr. Just wondering if it's a buffer in XP or if it's really the NAS software. Since faster CPUs and more memory don't seem to help the issue it makes you wonder. Adding buffers on the NAS seems to help though so room for improvement for sure, anyone tried going even bigger? I looked around for some SATA cards with cache but they are all a million bux plus are RAID models. Any stack tweaks for Linux maybe? More disk tweaks maybe?

I'm not sure the troughs are really a problem, rather they seem to be a reflection of reality: some parts are faster than others which translates into the fast parts (network, in this case) waiting for the slow parts.  The obvious seems to be true: Unraid can't pump out 125MB/sec as is theoretically possible with GigE.  This could be a factor of the raw drives, the SATA bus, PCI/PCI-E, or simply the Unraid software itself.

 

I am not overly concerned.  However, I will continue to tweak the settings on my PC (i.e. is my GigE card optimized, are my drives putting out all they can, etc.), the network (is my GigE switch the problem), and the Unraid box (GigE, SATA configuration, drive settings, ...)

 

 

Bill

Well, SATAII is supposed to be like 300MB/sec yes? Or is this a Bit vs Bytes thing? Honestly I'd be happiest to see a steady level of performance, as it stands now I'm not sure that having a GigE port is an advantage since pushing past 100megs steadily doesn't seem possible right now. If I knew where the bottleneck was I'd could try to address it but right now it's a puzzle.

 

Tried pushing the disk buffering to 4megs. Saw no consistant different worth keeping it there on my system. I'm presently moving a large number of DVD from one NAS to another and from my workstation to the NAS so I'm getting to watch this and try to tinker. Very frustrating to see all network activity cease and then start again over and over.<sigh>

Yes, SATA-2 is 3.0 Gb or roughly 300MB/sec.  However, that is the bus capacity, not the drive's speed.  Not sure what drives you are running, but I bet they are somewhere in the 50-70MB/sec range so forget about the SATA2 spec as it is not the weakest link.  Other than when calculating parity, Unraid's M.O. is to have only one drive spinning so it won't take advantage of the higher speed bus like other NAS' do.

 

So, if 50-70MB/sec is the peak between the drive, the drive bus, and the network, then what else is limiting speed?  Two things come to mind: the system bus (PCI or PCI-E) and Linux/Unraid/Samba/Windows/....  PCI is 132MB/sec (IIRC) but that includes a lot more than the one-way traffic of, for example, movie from unraid drive to the network.  PCI-E is quite a bit faster than PCI.

 

Add to the "can't be higher than 50-70MB/sec" argument the vageries of data transfer means the average speeds are rarely, if ever, achieved, so that probably brings the real-world average to under 40-50MB/sec.  Given that I am experiencing about half of that, there seems to be hope for some improvement, but not exponential potential.

 

BTW, I am sure this exact conversation has been had ten times on this board, probably with me participating in all of them - hopefully I am not contradicting myself each time.  LOL!

 

Regarding your "not sure if GigE is an advantage" comment, I am getting FAR better than even theoretical FastEthernet speeds.  FE caps out at 12.5MB/sec where I am getting about double that.  Just remember that your 14-18% GigE utilization means 40%-80% higher than FE.  FastEthernet is 100Mb/sec (bits, not Bytes).

 

 

Bill

I'm running both 7200RPM 1TB WD drives with 16meg cache and 750Gig Seagate drives also at 7200RPMs and with 16megs of cache. The Seagates are ST3750640AS models the WDs look like WD10EACS-00Z which is their "green" drive - supposedly the spidle is variable speed. I cannot recall the site out there that benches all the new drives, I did find some reports of Seagates hitting an average of 70MB/sec - that includes the bus blah blah. I suppose if the bus isn't a bottleneck than why set it to SATAII and not leave it at 150Megs? Apparently a difference is seen setting it higher so there must be an impact at just 150 right? Actually the test I found had it bursting to over 160MB/s yay. Looking at Wikipedia it does look like SATA 150 is adequate for most use, frustrating to say the least. http://en.wikipedia.org/wiki/SATA_II#Throughput I guess it is telling that a parity check with unRAId runs at around the mid 40s according to the interface. <sigh> If the drive were mounted in my desktop I feel sure that moving files around wouldn't take this long so how could the disk be what holds it back? Yeah, I'm frustrated but the SATA does seem faster and certainly parity checks way faster. I can certainly say I wouldn't prefer a striped data setup even if it might be faster...

I guess one question we should be asking ourselves is, "who cares?"  Driving to work in rush-hour traffic in a Ferrari or a Yugo makes no difference because the application doesn't value top speed.

 

The same can be said for a media server.  Other than copying files - where faster is better - the speed requirement is determined absolutely by the media in question.  So if 25MB/sec can get you four HD video streams and a bunch of music streams, why do we want 30MB/sec or 40MB/sec?  If these were database servers or were being used for movie-making (two race track applications) , then the additional speed would be required.

 

Yeah, I'd rather drive to work in a Ferrari too - the two cars may arrive at the same time but I want to be styling!  I would also like to get better speed out of the Unraid box, but it really is secondary to the cool features we have (easy expansion, unequal drive sizes, SATA+PATA, low cost, etc.)

 

BTW, I found a cool article showing read speeds of various NAS solutions.  Note that my 25MB/sec is pretty darn good.

 

http://www.smallnetbuilder.com/component/option,com_nas/Itemid,190/chart,15/

 

 

Bill

All very interesting... Whats most annoying is when watching the usage is during a copy (a read from the unraid seems quicker not surpisingly) and the network goes down to 0 and can sit there for a few seconds before starting up the next burst when it jumps to 15-18% then 25% and then back down again.. whereas as mentioned on the 100mb link it cruised at 80-85% utilised all the time.

 

BUT it is alot quicker moving files about which is good in my book.. as copy 100GB on a 100mb network was so painful.

 

As a slight aside.. is anyone using the PCI-E Adaptec Sata 4 port board with unraid ok ? I am one drive away from needing a expansion board

I'll read the article in a minute Bill, honestly if ouir average speeds is anywhere in the ballpark then complaining is indeed kind of silly. I'd still like a smoother data flow though and not this herky jerky thing we've got going.

 

As for fast enough - check out the issues the TIVX user is having streaming to his device. Midway through the box is running out of data judging from the description. I personally haven't hit this but I'm also still looking for a good HD front end for HD files, MKV files in my case. TIVX doesn't support it unfortunatly and looking at the files it does support perhaps those files require a higher datarate. In any case from the sounds of it at least one of us may be having issues with throughput and that concerns me since at some point we might all be looking to stream something more data intensive. <shrug> If we've maxxed out the disk subsystem then it is what it is but if some other server can do it better than that might point to room for improvement - that said I do not want striped data so that will remain a limitation.

 

A friend is building a new NAS sometime soon. He's going to try ZFS, WHS, and I'm going to try and get him to try unRAID too. If I can get him to do some speed testing I will, I would be REAL interested to see where unRAID falls in the mix.

IIRC, HD requires 20-30Mbit whereas the typical Unraid (SATA on GigE) pumps out 20-30MBYTE, so we should have plenty of headroom.

 

For synchronous applications such as media, what is often more important than average transfer rates are minimum transfer rates.  Even if Unraid pumped out Petabytes per second on average, if it dropped down too low for a few tenths of a second and a couple of frames were lost, you would notice.  This is a similar issue with graphics cards, where it is the floor, or minimum, transfer rate (MB/sec with us, Frames/sec with graphics cards) that often makes or breaks the user experience.  Of course, with media you can buffer it on the receiving end, something not possible with graphics due to the interactive nature where latency is a killer - game-speaking, literally.

 

 

Bill

Erm (sorry to divert the conversation again)... I'm running 720p/1080p MKV files from my unraid over the lan to a TVIX machine.. so it does play them (and very well overall) on the latest beta release....

Erm (sorry to divert the conversation again)... I'm running 720p/1080p MKV files from my unraid over the lan to a TVIX machine.. so it does play them (and very well overall) on the latest beta release....

 

I'll sidetrack again... TIVX will play MKV files?!! I looked specifically for that since that is something killing me and did NOT see it in the features list. I've been working on hacking an AppleTV to use as a frontend but a TIVX just might do the deal.

http://drivers.softpedia.com/get/FIRMWARE/Others/Dvico-TVIX-HD-M-41-5100SH-Firmware-Support-MKV-with-Tuner-1-2-23-Beta.shtml

 

May or may not be the latest - do your own research.

 

 

Bill

I've posted more details into the tvix thread...