NASLite Network Attached Storage

Alot of the new gigabit NICs support Jumbo Frames. Is this enabled in the drivers provided with NASLite by default or is Jumbo Frames simply not supported in NASLite? Personally I would love to see a setting for this under Network Configuration if possible

bump....

Same question.

Don't need it, in a properly configured network, NASLite will saturate your NIC. In fact, it will probably decrease performance in most cases with things such as slow drives etc.

The very fact that you could "saturate" the nic is reason enough to use jumbo frames - more data in the same number of packets equals less overhead and faster throughput, and consequently, less likelihood of saturation.

Oh - and whilst you're about it - how about a tcp offload engine? That would make the ip stack even less of a bottleneck.

And yes, if you can transfer data to the NAS unit faster than it can write it to a slow disk, you could see a perfomance hit - but that's no reason to omit jumbo frame - all you have to do is make it configureable, and let the user enable or disable it as required.

Jumbo frames will not necessarily avoid saturation of the nic. All it means is that you get to push more data per packet. The packet size however will be larger so you get to push less packets.

In most practical applications of Naslite, jumbo frames are really a tossup. They will probably provide little or no noticeable improvement. That being especially true with common consumer grade routers most people have.

So, although you have a theoretical argument, in practice it matters little.

Just my opinion.

Check my post - I didn't say it would - what I said was it would reduce the possibility.



I see we're in perfect agreement - in fact that's the reason it's less likely to saturate the nic - you transmit more data, in less bits, because you reduce the overhead.



Again - we agree - your common consumer grade router is 100 mbps, jumbo frame is not a consideration.



In fact, my arguement, as intended, was just as ridiculous as Ralph's answer - however my point is - and it remains a valid one - Ralph's answer shows no good reason not to include jumbo frame support.

Browse the forum and you'll find that some of the folks have significant hardware investment in their NAS, do you really think these folks are running your "common consumer grade router"?

The people who're buying the product are the video aficiandos looking for storage for their collections, people willing to spend money to purchase RAID cards and high capacity hard drives, and with the price of a gigabit switch with jumbo support being less than that of a couple of 500GB disks, yes, I believe there is a need for jumbo frame support.

Make it a configureable option and you can eliminate Ralph's stated reasons for not supporting it.

OK,

So I have two 2GHz P4s I'm setting for a client. One is running Naslite 2 and the other one running Gentoo. Both have Intel Gbit nics and 2 SATA drives in RAID-1 on a 3ware. Through 3' crossover I'm getting 93MBs sustained transferring an 8G image. Are you telling me that with jumbo frames I'll be getting a higher throughput, essentially pegging the card at 125MBs?

Through my Linksys SD2005 switch however I'm only getting 34MBs sustained when moving the same file. Will Jumbo frames increase my throughput through the switch in any significant way?

According to your post above I would expect that to be the case.

I would also like a definitive answer on whether or not the latest Naslite2 can use Jumbo Frames, provided the h/w is capable.

2nd question: if I put a 2nd GigNIC into the NL server, connected to the same router, will I get any benefit?

I would like to run my photo database off NL across the GigLAN.

No and No.

i've never understood people's love affair with jumbo frames.

you are really only gaining about 2% in efficiency (38 byte overhead on 9000 bytes instead of 38 byte overhead on 1500 bytes) the saved time is measured in milliseconds, rather insignificant for the problems implementing jumbo (non-standrd) packet sizes on an established LAN. heck, a couple lost jumbo packets will more than destroy the "benefit".

sure, jumbo packets can lessen CPU load, but i'd rather invest in decent NICs and routers that will allieviate the CPU load themselves by offloading the checksuming and having direct memory access, etc etc. and wrt NL, the CPU is only being used for file serving, so having a little extra CPU load is negligible, and won't be causing any slow down. as is, i can have two clients connected to NL and simultaneously peg both transfers @ around 58-60 MB/s. thats all i need.




if the freight train has 100 40 foot cars or 10 400 foot cars, does it really make a difference?

sure there are less hitches, but the smaller cars will have an easier time navigating the curves on the older tracks, and in the end the seconds saved might not be worth the problems inccurred when one of those 400 foot cars de-rails.

There's no overall gain with jumbo frames, if there was, it'd be the standard today, and it's not. I spent an enormous amount of time researching this with a system engineer from Cisco. We tested jumbo frames and nic bonding, neither produced anything worthy.

And as the previous poster pointed out, a dropped packet? enjoy your stuttering/stopping video.

How do you arrive at that 2% efficency figure - throughput tests with a different NAS have shown as much as 20% increase simply be switching to jumbo frames.

Lost jumbo packets? Why should you have lost jumbo packets? This is supposed to be a local area network, lost packets should not be a concern.

And the router - what part is that playing in the jumbo frame transfer?



Your freight train analogy isn't a really good choice - it's the equivalent of running jumbo frame on an older CAT5 network that's not rated for gigabit ethernet in the first place.

You'd think with that kind of performance gain, Jumbo frames would be part of the IEEE 802.3 Ethernet standard, and common place, just another attempt by the man to keep us down.


GPL

And that's why there are so many GigE switches that treat jumbos like bad fish. Unless the client and server are on the same page things can get ugly and often do.

2% is about what the theoretical increase in efficiency is with jumbo packets. you gain a bit more data transfer and a bit more overhead, so in raw numbers 2% is the efficiency gain.

actual in the field numbers may differ, but there are so many different variables (just look at the differences in throughput amongst $5 enet cards versus good,usually more expensive enet cards. you can easily gain 20% increases in throughput just by purchasing a good card that implements worldwide standards in a precise way.

lost packets can happen for a number of different reason, i'm not saying lost packets will happen, just that if they do happen, jumbo packets lose any and all theoretical benefit.