The Arista 7500E can have up to 64 member links in a LAG. MLAG two of them together, and that's a port-channel consisting of 128 links. Why you would want to do this - I don't know. But that's one fat pipe. XD
My Virtual/Storage guy would absolutely love that.
What's the speed on each port? I have an Oracle guy that's interested in maximum throughput.
Saw the Cisco 9K can also handle this number.
If Im not mistaken isnt this a feature of the Trident 2?
Quote from: that1guy15 on October 27, 2015, 12:55:00 PM
Saw the Cisco 9K can also handle this number.
If Im not mistaken isnt this a feature of the Trident 2?
Not that it's a big deal, but do you have a document that supports this? The 9500 data sheet says 32 max, which would make for 64 in a vPC pair I'd imagine.
Reference: http://www.cisco.com/c/en/us/products/collateral/switches/nexus-9000-series-switches/datasheet-c78-729404.html (http://www.cisco.com/c/en/us/products/collateral/switches/nexus-9000-series-switches/datasheet-c78-729404.html)
The 7500E is an Arad-based platform. Not Trident. But yes, I didn't think about it before and had to look back at my notes and T2 can support up to 64.
Quote from: deanwebb on October 27, 2015, 12:31:22 PM
What's the speed on each port? I have an Oracle guy that's interested in maximum throughput.
Speed isn't really a limiting factor here. To my knowledge its the # of bits that can be used to hash across a number of member links. So on the 7500 it's 10/40/100G - up to physical port count limitations of course.
Quote from: AspiringNetworker on October 27, 2015, 06:06:42 PM
Quote from: deanwebb on October 27, 2015, 12:31:22 PM
What's the speed on each port? I have an Oracle guy that's interested in maximum throughput.
Speed isn't really a limiting factor here. To my knowledge its the # of bits that can be used to hash across a number of member links. So on the 7500 it's 10/40/100G - up to physical port count limitations of course.
Please make them all 100G, thank you. Having 12800G of throughput sounds like it will keep the Oracle guys happy... for a week... :developers:
Quote from: AspiringNetworker on October 27, 2015, 06:01:46 PM
Quote from: that1guy15 on October 27, 2015, 12:55:00 PM
Saw the Cisco 9K can also handle this number.
If Im not mistaken isnt this a feature of the Trident 2?
Not that it's a big deal, but do you have a document that supports this? The 9500 data sheet says 32 max, which would make for 64 in a vPC pair I'd imagine.
Reference: http://www.cisco.com/c/en/us/products/collateral/switches/nexus-9000-series-switches/datasheet-c78-729404.html (http://www.cisco.com/c/en/us/products/collateral/switches/nexus-9000-series-switches/datasheet-c78-729404.html)
The 7500E is an Arad-based platform. Not Trident. But yes, I didn't think about it before and had to look back at my notes and T2 can support up to 64.
I am going mostly off memory and didnt have a way to look it up as I was on the run. Yeah the 64 on the 9K could have been from a pair. Honestly I have heard so many product briefings in the past several months and all the CCIE studies, shit is starting to blend together.
Ill dig and look.
Quote from: that1guy15 on October 27, 2015, 06:28:18 PM
Honestly I have heard so many product briefings in the past several months and all the CCIE studies, shit is starting to blend together.
Hahaha - story of my life homie.
Quote from: deanwebb on October 27, 2015, 06:24:16 PM
Quote from: AspiringNetworker on October 27, 2015, 06:06:42 PM
Quote from: deanwebb on October 27, 2015, 12:31:22 PM
What's the speed on each port? I have an Oracle guy that's interested in maximum throughput.
Speed isn't really a limiting factor here. To my knowledge its the # of bits that can be used to hash across a number of member links. So on the 7500 it's 10/40/100G - up to physical port count limitations of course.
Please make them all 100G, thank you. Having 12800G of throughput sounds like it will keep the Oracle guys happy... for a week... :developers:
LOL! My aren't we optimistic!
I actually typed Orable and then corrected it.
Now I realize that, since it rhymes with "horrible" and sounds like the British "orrible", I should have left it that way.
I feel dirty reading this thread. Any sort of mlag is bad mkay.
Quote from: burnyd on October 27, 2015, 10:57:00 PM
I feel dirty reading this thread. Any sort of mlag is bad mkay.
But, but... vPC can now support L3!
Quote from: burnyd on October 27, 2015, 10:57:00 PM
I feel dirty reading this thread. Any sort of mlag is bad mkay.
Interesting statement. Why do you feel that way? Does that include between ToR and host?
Quote from: AspiringNetworker on October 28, 2015, 10:35:55 AM
Quote from: burnyd on October 27, 2015, 10:57:00 PM
I feel dirty reading this thread. Any sort of mlag is bad mkay.
Interesting statement. Why do you feel that way? Does that include between ToR and host?
:barf: :barf: :barf: :barf: :barf: :barf: :barf:
Quote from: AspiringNetworker on October 28, 2015, 10:35:55 AM
Quote from: burnyd on October 27, 2015, 10:57:00 PM
I feel dirty reading this thread. Any sort of mlag is bad mkay.
Interesting statement. Why do you feel that way? Does that include between ToR and host?
Yes for both. The Arista one is a bit cleaner as it just shares the sysdb state which is really neato but I generally do not like control plane fate sharing. I think anyone who has worked with mlag chassis has been burned by something in the past.
As far as host mode goes there are many ways now to get active active out of both links. Most end hosts are either virtualized in a typical VMware fashion or containerized and there are easy ways to get forwarding out of both links.
Wouldn't like to draw that network diagram :twisted:
Quote from: SimonV on October 28, 2015, 11:29:24 AM
Wouldn't like to draw that network diagram :twisted:
Haha... I learned that lesson a long time ago and now just draw a thick colored line with a legend that says something to the effect of, "4x40G". Unless of course you're talking about a rack elevation/cabling diagram.. then.. yyeeeeaaahhhhh that's gotta suck.
Quote from: burnyd on October 28, 2015, 11:22:12 AM
Quote from: AspiringNetworker on October 28, 2015, 10:35:55 AM
Quote from: burnyd on October 27, 2015, 10:57:00 PM
I feel dirty reading this thread. Any sort of mlag is bad mkay.
Interesting statement. Why do you feel that way? Does that include between ToR and host?
Yes for both. The Arista one is a bit cleaner as it just shares the sysdb state which is really neato but I generally do not like control plane fate sharing. I think anyone who has worked with mlag chassis has been burned by something in the past.
As far as host mode goes there are many ways now to get active active out of both links. Most end hosts are either virtualized in a typical VMware fashion or containerized and there are easy ways to get forwarding out of both links.
Agree, though I will say, sounding completely bias of course, in the now almost two years (Wow - combat veteran status right there ;P) I've been working on this stuff, I haven't seen a single MLAG issue. Even during a major upgrade between a 4.14 version and a 4.15 version. I'm not used to that at all ;P
At Websense, expecting anything to communicate between each other even between minor versions was a crap shot.... and you know what the "recommended" upgrade process was, half the time? Remove and reinstall... lol. Hopefully it's gotten better since those days...
Quote from: burnyd on October 28, 2015, 11:20:10 AM
:barf: :barf: :barf: :barf: :barf: :barf: :barf:
LOL
But haven't you heard? Switch stacks are the shizzle!
lol I also hate switch stacks.
Quote from: burnyd on October 28, 2015, 11:57:38 AM
lol I also hate switch stacks.
Every time a customer says to me 'we have a redundant stack as our core' I die a little inside.
They also die a little inside each time I have to tell them they have to down their nice 3750X stack for an hour to go from 12 to 15. 'But they're redundant!'. BZZZZT SINGLE MANAGEMENT AND CONTROL PLANE lol. No I'm not game to do a rolling stack upgrade.
I remember this one customer who I had to down their entire regional hub (3750X stack... nice....) as there was a memory leak (nice IOS 15 switch code!) that made it impossible to log in. On the console, it came up with 'no memory for AAA process' type syslog and their graphing showed a nice 95% utilisation flatline LOL (unrelated but a month after the reboot the utilisation hadn't increased, never got to the bottom of the reason for the original leak).
VSS is just as bad, probably worse as you're dealing with bigger networks and corresponding bigger kerfuffle per outage/incident.
Maybe its the NSX brainwashing but L3 ECMP core and L3 to the distro/edge or GTFO and no your user switches do not need active/active MLAG or FEX or whatever, show me your cacti graphs again for your access uplinks? LOLOL
but yes MEC is nice and one L3 device makes it really simple for the L1/L2 types to understand. They don't even have to remember to type in show standby. Finally I've actually never seen anyone run GLBP in prod, even the big boys (banks, big oil etc.), maybe its just my luck. Then again I've never seen anyone run VTPv3 either.
haha in some ways I would prefer that to some cisco dc networking gear.
Thats a good point. Sometime dead simple is the best solution.
Quote from: that1guy15 on October 29, 2015, 12:05:23 PM
Thats a good point. Sometime dead simple is the best solution.
Hello. I have some gear from Belkin that will absolutely baffle you.