Where are we going to go with the datacenter infrastructure?
Think about where we are now versus where we were 8-10 years ago.
Back in 2k I designed an infrastructure from the ground up. At the time it was completely state of the art. Pair of border routers, 4 load balanced firewalls, 2 tiers of redundant load balanced cores with redundant leaf (rack) switches hanging off of each core. There was also a secondary set of LB FWs inbetween the app server layer and the DB layer. It was incredible. Foundry (majority of the equipment) actually took a number of companies in to look at the architecture, including Deutsche Bank.
The app and DB servers were all compaq proliants and we had an 1 TB EMC Sym as our SAN. All of this was globally redundant to a mirror datacenter on the east coast.
It was a really amazing setup. Completely fault tolerant on all levels.
From some rumors I am hearing, which may or may not be correct, we will be able to basically duplicate the same storage and power into a single 42U rack. 2 REDUNDANT architectures in the same rack.
In 2 chassis.
Lets take a look at how this (in theory) could happen.
Cisco recently released the Nexsus 7K switch, which just won Best of Interop this week in Vegas.
It has a 15 Tbps backplane to be future-proofing for either 40Gbps or 100Gbps ethernet (whichever one finally wins the next standard but that is another topic.) I have seen one of these bad boys in action and it is a really cool box. I am still going to have some issues with Cisco and oversubcription of ports. The current big 10Gig blade has 32 ports. Cool. Would be really cool if it wasnt 4:1 oversubscribed. The blade only has a 80Gbps connection to the fabric so fully populated you will only get 4Gig out of each 10Gig port.
This is not where it gets interesting.....
This is where it does. I have heard that there is going to be a blade that will house SAS disks right into the chassis. 32 2.5" SAS disks. That is over 9TB raw. This would be so cool. Even cooler? I have heard that you are going to be able to put server type blades into the chassis as well.
So....we have a 10 slot chassis. What should we do with this?
Ok....we need to have a sup engine. We also need some connectivity. 2 blades down, 8 remaining.
Now storage. 2 slots 32 disks upto 9TB of disk. 2 more down and 6 to go.
Now I am not sure how the server blades would work. Looking at an IBM or a HP blade chassis and the size of the blade I could imagine you could get 2 blades per slot. Perhaps using something like the SPA adapter used in the Catalyst 6K line. Lets even pretend you can only get one 4 proc blade per slot.
6 blades with 4x4 core procs and lets say 32 Gig of RAM per slot.
That is 64 cores with 192 GB of RAM that can work with a 15Tbps backplane to communicate with the network and the storage.
A box like this would have the same have about the same horse power in 21U that I used to have in 10 racks (420U) in a datacenter.
This has been the problems I have had with Blade Centers....you can only do so much with them. They are only good for raw cpu cycles. You still need a separate switching and storage environment.
In what could be available from Cisco you have, well, a nexus. You have full layer 1 to 7 in one box.
Everything in one box. If there was intelligent virtualization you could spin VMs up and down as needed and as a chassis becomes overloaded you could spin up idle chassis to take over the load.
Rumor has it Cisco spent about 250,000,000 on the development of this box. It is going to be FCoE capable and it will be ready to take the next generation ethernet standard whether it is 40 or 100 Gig. If they add storage and intelligence we are really going to be moving forward. Even if you dont want to put all of your server resources into the chassis you will be able to run some very interesting network analysis with a blade. As the cpus are right in the chassis you can span a huge amount of traffic for any DPI application.
I think this could be really cool.