I am retiring some of our 6900's and 3600's to migrate to a Viprion 2400. Does anyone use vCMP yet? If you virtualize the instances, is the performance less than not virtualizing the instances? Has vCMP been out long enough that you would feel comfortable running production on it? I'm just looking for some actual real life user feedback on it.

Good questins. I'm looking forward to doign this at some stage this year or next myself. Withr egards to performance, tmm itself shouldn't be any slower as the CPU cores themselves are dedicated to a particular tmm instance. IIUC network interfaces (Physical) are shared too. But they come with 10Gb NIC's (2x sfp+'s with a blade) so that hsoluldn't be any more of an issue that sharing the 10Gb links between your core switches (If you're lucky enough for a 10Gb backbone :). The only thing I'm not confident of is the SSL offload hardware. The encryption hardware is going to be shared between instances. As is the license for it. And I THINK that the license is enforced globally. So if you have qa 500TPS license, and one of your vCMP's bursts 5 SSL negotiations in a 10ms window, you will ikely get license failures in the other vCMP's... vCMP itself has been out as long as v11... So if you're comfortable with v11, you should feel comfortable with vCMP... But as always, test, test and retest... H H

We are actually consolidating several platforms (6400/3900/1500) onto the VIPRION C2400 platform right now using vCMP. At first, I encountered several problems - mostly not due to vCMP specific problems but because of general v11 issues. If you haven't made the jump to v11 on any boxes yet, be prepared for some pretty drastic changes - especially in the HA department (with the introduction of traffic/device group/sync groups/etc). It would appear that with 4 vCMP guests virtualized on a single B2100 blade, each vCMP guest should have the equivalent processing power of somewhere in between a single 6900 and 8950. The B2100's really excel in L4, and can push ~40gbps of L4. This is actually what the BIG-IP 11000 is spec'd at. Real world performance - I can't yet tell, because I only have a minimal amount of traffic on the VIPRION platform at this time. Regarding SSL TPS licensing. Each core on the B2100 blade has 500TPS included with the base license. All of this stated capacity (2000 SSL TPS/blade) is available to any and all vCMP guests. Any guest that is not currently using any SSL will not contend for any SSL resources. If one guest needs 2000 TPS, and the others aren't using any, one guest will get the entire 2000 TPS (even if that guest is using one blade). This is how it was explained to me by my SE. My beef with vCMP at this time is resource allocation. There is a 1:1 ratio of CPU:vCMP guest, meaning that if I have a B2100 that only has 3 vCMP guests, I am basically wasting an entire CPU core in the B2100. I can not allocate 2 CPU cores to 1 vCMP guest, therefore I am wasting a processing core. The only way to provide more processing power for a vCMP guest is to add another blade in to the chassis. Once you do this, you only then have the option to make a vCMP guest span "one" or "multiple" blades. Not very flexible, in my opinion. Josh

I think Josh's summary is quite good and accurate. Not very flexible... I'd encourage you to also give this feedback to F5 Support and your account team. PD is actively working to address the flexibility issues with vCMP. But opening RFEs would help with prioritization. Thanks, Aaron

Interesting, so you can only assign one core per guest per blade? So I could do 2 cores for a guest if I have 2 blades? I just bought 2 blades per chassis at the moment. So I was only planning on splitting it up into 2 guests, so that would be a lot of cores wasted at the moment. I'm in the process of racking it up in my lab at the moment to start POC'ing it out.

Nick, Yes, you can currently only assign one core per guest per blade. If you have two blades, you can have one guest span both blades which would give that guest 2 cores (1 core per blade). Yes, unfortunately, you need at least 4 guests to fully utilize all of the processing power of each blade. You could have 4 guests that each spanned both blades - 2 cores per guest. I do have an RFE open regarding this "issue." Josh

vCMP on Viprion 2400

Hamish
Cirrocumulus
Mar 07, 2012
Good questins. I'm looking forward to doign this at some stage this year or next myself.

Withr egards to performance, tmm itself shouldn't be any slower as the CPU cores themselves are dedicated to a particular tmm instance. IIUC network interfaces (Physical) are shared too. But they come with 10Gb NIC's (2x sfp+'s with a blade) so that hsoluldn't be any more of an issue that sharing the 10Gb links between your core switches (If you're lucky enough for a 10Gb backbone :).

The only thing I'm not confident of is the SSL offload hardware. The encryption hardware is going to be shared between instances. As is the license for it. And I THINK that the license is enforced globally. So if you have qa 500TPS license, and one of your vCMP's bursts 5 SSL negotiations in a 10ms window, you will ikely get license failures in the other vCMP's...

vCMP itself has been out as long as v11... So if you're comfortable with v11, you should feel comfortable with vCMP... But as always, test, test and retest...

H

H
Josh_41258
Nimbostratus
Mar 07, 2012
We are actually consolidating several platforms (6400/3900/1500) onto the VIPRION C2400 platform right now using vCMP. At first, I encountered several problems - mostly not due to vCMP specific problems but because of general v11 issues. If you haven't made the jump to v11 on any boxes yet, be prepared for some pretty drastic changes - especially in the HA department (with the introduction of traffic/device group/sync groups/etc).

It would appear that with 4 vCMP guests virtualized on a single B2100 blade, each vCMP guest should have the equivalent processing power of somewhere in between a single 6900 and 8950. The B2100's really excel in L4, and can push ~40gbps of L4. This is actually what the BIG-IP 11000 is spec'd at. Real world performance - I can't yet tell, because I only have a minimal amount of traffic on the VIPRION platform at this time.

Regarding SSL TPS licensing. Each core on the B2100 blade has 500TPS included with the base license. All of this stated capacity (2000 SSL TPS/blade) is available to any and all vCMP guests. Any guest that is not currently using any SSL will not contend for any SSL resources. If one guest needs 2000 TPS, and the others aren't using any, one guest will get the entire 2000 TPS (even if that guest is using one blade). This is how it was explained to me by my SE.

My beef with vCMP at this time is resource allocation. There is a 1:1 ratio of CPU:vCMP guest, meaning that if I have a B2100 that only has 3 vCMP guests, I am basically wasting an entire CPU core in the B2100. I can not allocate 2 CPU cores to 1 vCMP guest, therefore I am wasting a processing core. The only way to provide more processing power for a vCMP guest is to add another blade in to the chassis. Once you do this, you only then have the option to make a vCMP guest span "one" or "multiple" blades. Not very flexible, in my opinion.

Josh
hooleylist
Cirrostratus
Mar 07, 2012
I think Josh's summary is quite good and accurate.

Not very flexible...

I'd encourage you to also give this feedback to F5 Support and your account team. PD is actively working to address the flexibility issues with vCMP. But opening RFEs would help with prioritization.

Thanks, Aaron
Nick_T_68319
Nimbostratus
Mar 07, 2012
Interesting, so you can only assign one core per guest per blade? So I could do 2 cores for a guest if I have 2 blades? I just bought 2 blades per chassis at the moment. So I was only planning on splitting it up into 2 guests, so that would be a lot of cores wasted at the moment. I'm in the process of racking it up in my lab at the moment to start POC'ing it out.
Josh_41258
Nimbostratus
Mar 08, 2012
Nick,

Yes, you can currently only assign one core per guest per blade. If you have two blades, you can have one guest span both blades which would give that guest 2 cores (1 core per blade). Yes, unfortunately, you need at least 4 guests to fully utilize all of the processing power of each blade. You could have 4 guests that each spanned both blades - 2 cores per guest.

I do have an RFE open regarding this "issue."

Josh
Frank_47355
Nimbostratus
Apr 03, 2012
Hello. Everybody!

I going to do the migration of BIG-IP6400 , toward VIPRION 2400 the VIprion has only one Blade. How are the best practice to configure the IPRION2400.

I need the Setting Up the VIPRION® Platform, swomeone has this guide ?
Nick_T_68319
Nimbostratus
Apr 03, 2012
All the guides are listed on the support site

http://support.f5.com/kb/en-us/products/big-ip_ltm/versions.11_1_0.html
boneyard
MVP
May 01, 2012
can't imagine this has changed since it was last mentioned, but still want to check. with one chasis with one blade, if i run 2 vCMP guests they will both only use one CPU and the other two are wasted?
Chris_Miller
Altostratus
May 02, 2012
Posted By boneyard on 05/01/2012 01:24 PM

can't imagine this has changed since it was last mentioned, but still want to check. with one chasis with one blade, if i run 2 vCMP guests they will both only use one CPU and the other two are wasted?

That's correct.
boneyard
MVP
May 02, 2012
Posted By Josh on 03/07/2012 05:15 AM

It would appear that with 4 vCMP guests virtualized on a single B2100 blade, each vCMP guest should have the equivalent processing power of somewhere in between a single 6900 and 8950.

Josh

each vCMP guest? I believe the whole blade to be between 6900 and 8950, wouldn't that mean that a vCMP guest is equal to a (6900 - 8950) / 4, so more towards a 3900?

thanks for the answer Chris