|
[HELP] Switch Buffers - Quick QuestoinSo I was always concerned about our 2960 running our 3 HyperV hosts, and Core database server as far as the buffers are concerned on the 2960. I've heard they suck. Queries and such take some time, so I'm trying to troubleshoot this yet again.
I did a sh int on that switch and looked at those interfaces, but I see 0 for input errors, 0 for late collisions, 0 for collisions, 0 for output errors. Everything looks good to be honest.
Would this imply that the switch should be doing fine? If it was a buffer issue for us, I would see discards, no? |
|
|
On the interface stats, you'd also want to do "| i drop," specifically any input (host(s) to 2960) and output (2960 to host(s)) drops present. ...and I'm guessing this question comes from your old thread here? ? If so, did you ever set up SNMP / PRTG monitoring for the CPU, memory and interfaces like we suggested? ...I'll reiterate again... if you have no way to measure / trend any sort of performance variables, how are you supposed to tell what this setup can / should do? My 00000010bits Regards |
|
|
ya I just installed PRTG actually about 15 min ago (before I even read your post... odd....). Pretty sweet. Only get 10 monitors, but I got one of the troublesome switches up on it right now. |
|
1 recommendation |
to chugger93
You shill put a PRTG logo on your webpage, they'll give you a code for 30 (additional) sensors. See hereRegards |
|
|
That's pretty sweet, thank you!
Ya I don't notice anything unusual. 23% or so CPU...memory plenty left. Not a whole lot of traffic on any given port. No errors or discards form what I can tell. hmm |
|
chugger93 |
I believe this is how its basically connecting.. its a bad mockup..but gives u an idea |
|
cramer Premium Member join:2007-04-10 Raleigh, NC Westell 6100 Cisco PIX 501
|
to chugger93
Yes, the buffer space is a) limited, and b) shared across ports. As long as you aren't doing jumbo frame, it'll be fine. We ran 9k iSCSI across one for about a year. It works at low traffic levels, but drops will happen, and eventually, they'll be bad enough to make a mess. (We use an HP A5500 now. Not perfect, but cheap enough to be good) |
|
|
to chugger93
One quick point about monitoring desktop switches: looking at the INTERFACE statistics for microdrops due to ASIC/buffer issues is unlikely to yield good results.
You CAN look at the internal hw counters for drops but its platform specific and hard to know what each counter means without a decoder ring from TAC. |
|
|
Here's the stats on that interface. This pipe goes right to the DB server. The DB server has a teamed NIC operating at 2Gigabytes. It's never really taxed by any means...and I see no drops. Maybe I just need SQL tuning... not sure. |
|
|
to chugger93
Glad you got that sensor working...
Have you had any reports of performance issues in that timeframe that PRTG's been running?
If not, I'd continue to let that sensor run AND NOT TURN IT OFF FOR WHATEVER REASON, just to build up a baseline. I'd also continue to gather full "show int" snapshots every now and then... just as a sanity check.
One question... you indicated the DB NIC was teamed, yet you're only monitoring on interface Gi0/45... did I miss something here, and there's another interface in a portchannel on your switch?
My 00000010bits
Regards |
|
TomS_Git-r-done MVM join:2002-07-19 London, UK |
to HELLFIRE
Or just use MRTG and have add many as you want. |
|
2 edits |
to HELLFIRE
dunno? I figured the team nic has the one pipe out to the switch. Local Area Connection 1 & 2 are the ISCSI connection back to the SAN, 3/4 arnt being used. Team is for the network interface Port-channel1
switchport access vlan 101
switchport trunk native vlan 101
switchport trunk allowed vlan 1,4,10,89,90,101,111,121,145,150,912,1100,1500
switchport mode trunk
!
interface Port-channel2
switchport access vlan 10
switchport trunk native vlan 101
switchport mode access
!
interface Port-channel3
description ICRHV003 virtual trunk
switchport access vlan 101
switchport trunk native vlan 101
switchport trunk allowed vlan 10,90,121,150,912
switchport mode trunk
!
interface Port-channel4
description ICRHV002 virtual trunk
switchport access vlan 101
switchport trunk native vlan 101
switchport trunk allowed vlan 10,90,121,150,912
switchport mode trunk
!
interface Port-channel5
description ICRHV001 virtual trunk
switchport access vlan 101
switchport trunk native vlan 101
switchport trunk allowed vlan 10,90,121,150,912
switchport mode trunk
!
interface GigabitEthernet0/45
switchport access vlan 10
switchport trunk native vlan 101
switchport mode access
channel-group 2 mode active
!
interface GigabitEthernet0/46
switchport access vlan 10
switchport trunk native vlan 101
switchport mode access
channel-group 2 mode active
!
Also, I wonder if I should have jumbo frames turned on, on this switch |
|
|
to chugger93
..."so the "team nic" has only one physical interface to the switch" I guess is my question at this point. Based on the config, it looks like it.
As for jumbo frames, far as I know, turn em on only if you know switch and endhost supports it. I could be wrong...
My 00000010bits
Regards |
|
|
no you're right, theres two ethernet cables coming from two nics that go into ports 45/46. |
|
|
to chugger93
...oh derp, missed that. So to be sure, you'll want to put a similarlly configured sensor onto Gi0/46, and onto the port channel interface(s) as well, just to keep a God's-eye view on things.
My 00000010bits
Regards |
|