Jumbo Frames on Internet ?

As a professional in networking domain, I am sure lot of people have question in their brain that if Jumbo frames perform so much better on Internal Network, Then why doesn't IEEE makes it a standard across Internet.

What is a Jumbo Frame?

? Technically, the IEEE 802.3 Ethernet standard defines the maximum frame payload (MTU) value at 1500 bytes.

? Supporting anything larger than 1500 bytes is outside of this standard, and we call it a “Jumbo Frame”.

How much bigger?

? Nobody really knows. It’s non-standard, remember.

? The “rough guideline” for most people is around 9000.

? Most vendors today actually offer a different number.?4

Efficiency Numbers are taken from Wikipedia Article - Courtesy Wiki

Technically, if IEEE makes a standard of 9K MTU on internet we can earn greater efficiency and yield really high throughput.

? Until recently, jumbo frames have primarily been an “internal network only” thing at best.

? But now some IX operators are starting to roll out Jumbo Frame VLANs at major exchanges.

? This could eventually lead to the ability to deliver a > 1500 byte packet end-to-end. ? And many people are cheerleading this effort.

? With a lot of idealism about improving the efficiency of high-speed end-to-end flows, which is a good thing.

Time to See the Other side of Picture, why it can be the worst idea.

Picking an MTU standard is very hard looking at the deployment level can effect every vendor and every user on internet. Remember no standard value defined anywhere.

Another variable is PMTUD is how industry deals with MTU Mismatch.

? When a router encounters an MTU mismatch and a frame that is too large, it drops the packet and sends an ICMP.

? The host receives the ICMP, and reduces the packet size for the retransmission and the flow. Hopefully it now fits.

? Routers can only generate these ICMP packets so fast.

? It’s incredibly vulnerable to Denial of Service attacks too.

? ICMP packets are often limited/blocked by ISPs or users. ? The only reason stuff works today IS the 1500 byte MTU.?

I really think if everyone supports jumbo frames, it will be irrelevant because ultimately transport ship is way bigger than its payload, so if we fit 1500 or 4600 or 7800, until its under 9000 ( if IEEE makes a standard it should be fine ).

Now we reduce overhead of fragmentation and reassembly at lot of devices. Definately, will cause some abruption for whole industry and specially ISP's however I see it step closer to the moon ( or perfect internet here ).

Larger packet sizes increases jitter.??

? Making 100 Gigabit flows more efficient is a noble cause.

? But a single 9000 byte packet down a 10 Mbps link has 7.2ms of serialization delay alone.

? The benefits are non-existent until you have a 100% Jumbo enabled end-to-end path.

? No content network will ever enable a service which has minimal benefit and which risks breaking the flows for any percentage of their customer base.??

This one is real concern and nightmare for every network engineer.?however again if IEEE has standard around supporting jumbo on Internet, I dont see the why engineer community cant solve this problem.

Modern NICs help eliminate most of this.

? With techniques like Large Segment Offload (LSO)

? Instead of making it a big hassle to move data around in 1500-byte chunks, you just hand the NIC a 64k buffer.?

There are multiple ways to solve the problems, however again NICs are not industry standard as well, its like a workaround I see, it still doesn't solve the problem of fragmentation and reassembly. To me this seems like we will have a storage tank, however we will not increase the size of transport payload.

IEEE has many different ways to solve this problems, however I am sure there are way more smarter people working behind the scenes as well. This article is just for awareness purposes and shouldn't be used as a concrete evidence for implementing anything or changing industry standards.

Author -

Saurabh Gera