Kitz ADSL Broadband Information
adsl spacer  
Support this site
Home Broadband ISPs Tech Routers Wiki Forum
 
     
   Compare ISP   Rate your ISP
   Glossary   Glossary
 
Please login or register.

Login with username, password and session length
Advanced search  

News:

Pages: 1 [2]

Author Topic: Lower MTU users out there?  (Read 2843 times)

aesmith

  • Kitizen
  • ****
  • Posts: 1098
Re: Lower MTU users out there?
« Reply #15 on: February 21, 2020, 06:44:33 PM »

Are you using an IP MTU below 1492 bytes? Is that for IPv4, IPv6 or both? What are the circumstances ? Any problems encountered in the Internet as a result
Not me but a customer, on their WAN they have MTU 1476, MSS 1360.   They had some odd malfunctions until we forced that MSS.  On their Internet VPN they have MTU 1434 with the same MSS over-ride.  It's been a while so can't remember, maybe we set MSS deliberately low on the WAN so it's the same as the VPN.  Wouldn't swear to that.  I just remember we had to set that MSS over-ride when the started using DMVPN over the private WAN around 5 years ago.
« Last Edit: February 21, 2020, 07:14:34 PM by aesmith »
Logged

burakkucat

  • Global Moderator
  • Senior Kitizen
  • *
  • Posts: 34418
  • Over the Rainbow Bridge
    • The ELRepo Project
Re: Lower MTU users out there?
« Reply #16 on: February 21, 2020, 07:03:32 PM »

I use an MTU of 1478, <snip>

I need a memory refresh, please.

I know that an ATM cell is 53 bytes, made up of a 5 byte header and a 48 byte payload. Dividing the "magic number", 1500, by 48 and rounding down to an integer gives 31. Multiplying 31 by 48 gives 1488.

So what is the reason for (or the significance of) the 10 byte difference?   :help:
Logged
:cat:  100% Linux and, previously, Unix. Co-founder of the ELRepo Project.

Please consider making a donation to support the running of this site.

Weaver

  • Senior Kitizen
  • ******
  • Posts: 10294
  • Retd s/w dev; A&A; 3x7km lines; Firebrick; IPv6
Re: Lower MTU users out there?
« Reply #17 on: February 21, 2020, 07:21:55 PM »

Should that MSS be the IP MTU minus the IPv6 and TCP headers = -60 ? Why isnít TCP subtracting this automatically ? Presumably it doesnít know about the tunnel, which comes later? And PMTUD isnít helping, for whatever reason.

What happens if your TCP is with timestamps which adds another so many bytes? Need to adjust the MSS down even more in that case?
Logged

aesmith

  • Kitizen
  • ****
  • Posts: 1098
Re: Lower MTU users out there?
« Reply #18 on: February 21, 2020, 07:29:01 PM »

My example was all IPv4.  As for why it wasn't done automatically I don't really know, all I know is that we've seen occasional funnies with tunnels or VPNs, for example, one customer everything worked except RDP.  I'm going blame Microsoft IP stack and applications, given that I've seen Lync think it's a good idea to send video in 9K jumbo frames - with do not fragment set!
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 10294
  • Retd s/w dev; A&A; 3x7km lines; Firebrick; IPv6
Re: Lower MTU users out there?
« Reply #19 on: February 21, 2020, 08:19:36 PM »

Lync doesnít understand PMTUD properly then? MS only tested it across their own jumbo LAN, never through a router?  ;D What a shower.
The thing is, if you go to work for Microsoft probably everyone wants to go and work on something prestigious like operating systems or Azure or even Word or Excelís innards (which are really hard imho) and not on printer drivers or setup programs. Thatís why such peripheral software items are often rubbish, because they donít have the best people or most driven people maybe, just the B team who didnít make the cut for the sexy teams. So much for my theory, Iíve never worked for Microsoft, just in another such o/s dev environment, albeit a minute version relatively speaking.
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 10294
  • Retd s/w dev; A&A; 3x7km lines; Firebrick; IPv6
Re: Lower MTU users out there?
« Reply #20 on: February 21, 2020, 09:26:20 PM »

@burakkucat I was asleep so missed your post. The 10 bytes is I expect 2 bytes for PPP + 8 bytes for the ATM AAL5 CPCS trailer - an overhead at the end of the last cell. That 10 = 2 + 8 bytes is for the PPPoA case, so I assume thatís what our user is running.

One has to add the overhead including always the 8 bytes CPCS trailer and then round up as you divide by 48, either that or pre-add 47 and divide by 48 with the division rounding down. Then multiply by (48+5) to get the total number of bytes on the line. My PPPoEoA total overhead is 32 bytes  :'( including everything and then calculate the number of cells required to hold that by division that rounds up.
« Last Edit: February 21, 2020, 09:33:29 PM by Weaver »
Logged

burakkucat

  • Global Moderator
  • Senior Kitizen
  • *
  • Posts: 34418
  • Over the Rainbow Bridge
    • The ELRepo Project
Re: Lower MTU users out there?
« Reply #21 on: February 21, 2020, 10:18:37 PM »

@burakkucat I was asleep so missed your post. The 10 bytes is I expect 2 bytes for PPP + 8 bytes for the ATM AAL5 CPCS trailer - an overhead at the end of the last cell. That 10 = 2 + 8 bytes is for the PPPoA case, so I assume thatís what our user is running.

Thank you. I was just trying to understand the 1478 bytes MTU that ejs has set.
Logged
:cat:  100% Linux and, previously, Unix. Co-founder of the ELRepo Project.

Please consider making a donation to support the running of this site.

Weaver

  • Senior Kitizen
  • ******
  • Posts: 10294
  • Retd s/w dev; A&A; 3x7km lines; Firebrick; IPv6
Re: Lower MTU users out there?
« Reply #22 on: February 22, 2020, 06:02:14 AM »

Ejs has presumably picked an optimal value 1488= 31 * 48 an integer multiple of 48 for the total on-line byte count, and then subtracted the PPPoA overhead of 10 bytes to get the PPPoA payload which is the correct optimal IP MTU. So IP MTU = n_bytes_on_line - header_overhead = 1488 - 10, where 10 bytes = 2 + 8. Thereís always that one-off final-cell trailer overhead of 8 for the CPCS, and so the last cell can only hold 40 bytes of AAL5 payload, not 48. Any more and an additional cell is required.

Another example: My own numbers are 1440 - 32 = 1408 where 1440 instead of 1488 is an integer-multiple of 48 low enough to be within the MTU of my 3G USB NIC, including protocol overheads, and 32 bytes is my PPPoEoA overhead, instead of 10 bytes. Hugely bloated is PPPoEoA. This gives me 1408, the IPv6 MTU I use now.

See also Wikipedia on PPPoE overhead, which I wrote some years ago.  ???

Does that make sense? I used to use this perfect MTU many years ago, perfect for one-box ATM ADSL users.

Should of course simply ask Ejs  :)
« Last Edit: February 22, 2020, 06:27:04 AM by Weaver »
Logged

aesmith

  • Kitizen
  • ****
  • Posts: 1098
Re: Lower MTU users out there?
« Reply #23 on: February 22, 2020, 01:49:58 PM »

Just out of interest I checked our LTE interface at home, never having done so before.  It reports 1500 bytes MTU, and if I not losing my marbles that's confirmed by being able to ping with data length of 1472 without fragmentation.  Are you factoring in L2TP for you 3G, or does it just have a low MTU for some reason?
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 10294
  • Retd s/w dev; A&A; 3x7km lines; Firebrick; IPv6
Re: Lower MTU users out there?
« Reply #24 on: February 22, 2020, 09:01:55 PM »

The 3G USB NIC just has a really annoyingly low IP MTU anyway, 1440 iirc.
Logged
Pages: 1 [2]