3750X jumbo frames support & latency paranoia with ISCSI config

Unanswered Question

/* Style Definitions */ table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-qformat:yes; mso-style-parent:""; mso-padding-alt:0in 5.4pt 0in 5.4pt; mso-para-margin-top:0in; mso-para-margin-right:0in; mso-para-margin-bottom:10.0pt; mso-para-margin-left:0in; line-height:115%; mso-pagination:widow-orphan; font-size:11.0pt; font-family:"Calibri","sans-serif"; mso-ascii-font-family:Calibri; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:"Times New Roman"; mso-fareast-theme-font:minor-fareast; mso-hansi-font-family:Calibri; mso-hansi-theme-font:minor-latin;}

So I want to use one pair of 3750x Gb switches in an dual network/iscsi data configuration. I want to create a non routed L2 vlan for the iscsi traffic, running just on this 3750x pair. I want to enable jumbo frames support for this traffic, but I have to enable globally on the 3750 switch (cannot configure per interface). Next I want to plug in things like 2 ASA firewall, with some interfaces running at 10/100 (non gig). I will hardest the speed duplex on these ports and on the ASA. I may connect some ACE LB’s in the future. The ASA (8.2) appears to support jumbo frames. I think the ACE would probably too.

Question1: If I plug in a server or 3rd switch to the core 3750 stack, and these devices don't support jumbo frames, will connectivity cease? I understand the L3 MTU will step down, but the 3750 uplink ports to other devices will not run at L3. They will be L2 ports on separate vlans.

Question2: Fear was set in motion by the storage guy regarding the 3750 lack of ability to do static cache buffering. In other words, the 3750 switch packet latency (based on ASIC configuration) was deemed not acceptable and may cause problems for the disk I/O. I am not a storage expert so researched this – I don’t really want to fork over more money for Nexus switches at this point. And I want to benefit form common switch base infrastructure. In any case, seems like the 3750 switch has packet latency between 6-24 us (that’s microseconds); and disk IO latency in an ideal environment should be no more than 2 ms (that milliseconds). Now 2 ms = 2000 us at least the last time I was in science class. Therefore I conclude that the 3750 switch should exceed the disk IO requirement regarding latency by at least 2000/24 ~ 83 times. Now I understand if I exceed the switch backplane bandwidth I may have issues, but that’s 64 Gb/s. So will I have problems? Can I configure L2 QoS of some sort such that the 3750 switch will favor the ISCSI vlan throughout, including on the backplane ring/cable connecting the two switches together?

Many thanks in advance!

I have this problem too.
1 vote
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
teater Fri, 12/06/2013 - 11:21

Did you ever resolve your question?  I am the same scenario I am considering.

hey teater, we just turned on jumbo frames globally on the 3750x and used the switch for both ISCSI (on ISCSI vlan), voice and data on the approriate voice/data vlans. everythign worked well. even non-jumob frame traffice traversed switch fine. NOTE: we had not routed path into the ISCSI vlan of a firewall or router. it exsited just ont he 3750x stack. that kep potential jumbo frame issues to a minimum. ill ping the guy at that location to see how its working these days, aws im not there any more. intial rollout went well though.


This Discussion