Session is starting… just met the other Scott Lowe
We’re actually going to start with lunch…
12:41 PM – Presentation is starting. We’ll be talking about Unified Computing.
12:44 – Technology solutions to business problems. Business focus, but there is a lot of back-end stuff that has to work in order for the front line stuff to work. i.e. want BI? You need underlying storage and architecture to enable that service. Current fastest growing costs in IT are in the area of server management and administrative overhead. Cisco is working on reducing these expenses.
12:46 – Power and cooling costs are also increasing; reducing servers = lower power and cooling costs.
12:47 – Virtualization takes it to one level, but also increases complexity, increasing operational expenses. Other vendors simplify this challenge by adding a management layer and providing professional services. Keep adding software on software to manage these challenges adds yet more complexity and cost. Creates what Cisco calls an “accidental architecture”.
12:51 -Cisco sees an eventual migration from mainframe > mini > client/server > web > virtualization to cloud. To get there, tie it all together in a way that’s easy to manage end to end.
12:52 – Today’s Datacenter = trusted, controlled, reliable, secure. Or, go to a cloud provider to get flexibility, dynamic, on-demand and efficient services. Instead, combine these goals so we have compute and control.
12:54 – A lot of the technology behind UCS is also found in other Cisco products, such as Nexus. Unified fabric is one example.
12:59 – Cisco dismisses Gartner’s pessimistic outlook on convergence of data and storage networks as a controversial ploy to gain readership.
1:02 – Single server, very small environments aren’t going to benefit from this financially… only when getting to the rack level will this start to show promise.
1:04 – Legacy virtual switches – tied to the vswitch and not mobile. So, some folks trunk every VLAN down to each server, thus negating a lot of policy decisions. New: Cisco VN-Link.
1:06 – What does UCS solve? UCS at first glance looks like a blade system. Contains blades with mezzanine cards. Servers slide in the front; connections in back. There is a fabric interconnect “virtual backplane” ties to other networks – LAN, SAN (FC, etc). This is also where UCS Manager runs. Differs from other vendors with less cabling and simplified architecture.
1:12 – Wire once for bandwidth, not connectivity and then just allocate bandwidth to different services as needed.
1:16 – Virtualization scalability, reduced TCO. I/O, CPU and memory are the three primary resource needs. How do we take virtualization problems (i.e. app 1 needs to much I/O). There becomes resource contention at the CPU to handle all of the needs (i.e. soft switch). Make sure that the CPU is spending its time serving VMs, not infrastructure needs. Solution: Hypervisor Bypass.
1:20 – Customers are running out of RAM in their virtual environments. Xeon 5500 memory architecture – traditional: 12 to 18 DIMM slots (96GB to 144GB RAM per system). Cisco developed a mux/demux technology supporting up to 48 DIMM slots, increasing RAM capability to 384GB. Helps organizations better match resource needs to equipment purchases. No more need to buy more complete servers just to get more RAM.
1:23 – Embedded unified management – Unified management domain with automatic discovery and dynamic provisioning of resources. Traditional blades “are very stateful in nature” – i.e. BIOS revisions, etc. Cisco’s blades are different. All of it is centrally controlled… a “stateless server”. You end up with a stateless bare metal server with no need to manage the hardware directly.
1:29 – The tool is a best effort tool… if it can’t manage a system for some reason, the management software lets you know.
1:40 – Quite a few questions in the room… been listening, so I haven’t been typing. Sorry!
1:49 – Had to leave the room for a minute… back now. Some specific questions are being answered about the impact of Cisco’s overall management philosophy. Cisco is getting some good press on UCS. Cisco believes in making sure that they can support UCS by hiring people from non-Cisco server-centric environments. Where they don’t offer a service, they’ve partnered with others, including EMC, 3PAR, MS, VMware, HP, Oracle, SAP, CSC, and more.
1:53 – That was the UCS presentation…
1:57 – Cisco is seeing the most common replacement scenario be on the server replacement cycle. The speaker was unwilling to identify the vendor they are displacing the most.
1:59 – Question: UCS is a lower margin market for Cisco than they are used to. Can they sustain it and become am industry leader in the space? Cisco has more than 2,000 people dedicated to UCS.
2:08 – It was pointed out that HP is happy to bypass a reseller and sell direct to a customer whereas Cisco won’t, so the VAR channel is happier with Cisco.
2:10 – Short break.
2:24 – Chris Naddeo is up. Talking about multi-tenant environments.
2:25 – Challenges – individual tenants on a multitenant environment need to play nice with one another.
2:26 – Tenant defined: Was intentional ambiguous, but could be down to the VM guest level. SMT 1.0 is a reference design. It’s an end-to-end design that ties together Cisco, NetApp and VMware products to ensure quality of service. It’s a design – not a product with a SKU.
2:29 – Solution brief is 4 pages, overview is 25 pages, CVD is 90 pages, deployment guide is 100+ pages.
2:32 – There is no compliancy requirement… just a reference design.
2:34 – Now looking at the components of the SMT reference design. Availability, secure separation, service assurance, management. Material being shown in the screen is right from the design guide freely available for download from Cisco’s site. http://www.cisco.com/en/US/docs/solutions/Enterprise/Data_Center/Virtualization/securecldeployg.html
2:39 – UCS stateless stuff works better if you are booting from SAN rather than from local disk and this is included in the SMT reference design
2:44 – Cisco is targeting the service provider with SMT of which UCS is a key component. Was UCS developed for this/others/both?
2:49 – A key component of SMT is NetApp’s MultiStore.
3:00 – Ok. I hate marketing lies. Misleading, dishonest.
3:16 – VCE presentation – from “the other Scott Lowe” and Ed Saipetch
3:34 – Why VCE? Help customers get to lower costs by moving to both internal and external clouds.
3:36 – VCE is integrated sales, engineering & support teams from Cisco and EMC/VMware. This is their focus. Engineer tested and validated packages. Accelerated virtualization option. They were asked if EMC has become more partner friendly… answer was yes.
3:40 – Scott Lowe indicates that he’s seen a major change in EMC’s stance on partnerships.
3:42 – Why would I buy a Vblock? Answer is forthcoming. Vblocks are actually being sold.
3:47 – The questions was asked: Who owns the Vblock from a task perspective? Answer: Organizationally dependent and usually lands with the group that handles virtualization.
3:51 – Nuts and bolts of a Vblock – Design governance, compute, networking, storage, software
3:52 – Design governance – Three companies coming together to create a product. Pretested, integrated, ready to go/grow. Balanced configuration with regard to compute power, storage I/O, network capacity, virtualization, simplified deployment and management.
3:54 – Nuts and bolts – compute layer = Cisco UCS, VBlock Type 1 is 16 to 32 UCS B-series blades (128 to 256 cores, 960 to 1920 GB RAM) , Vblock Type 2 is 32 to 64 UCS B-series blades (256 to 512 cores, 3072 to 6144 GB RAM)
4:01 – Type 0 Vblock. Smaller than a Vblock 1, but no specs beyond that at present.
4:10 – Network – 10GbE, FCoE and Fibre Channel, 2:1 oversubscription ratio for each chassis. Vblock 1 (2) = 4 (8) GbE and 4 (8) FC uplinks per fabric; Predictable network oversubscription ratios min 4:1 to 8:1 max. Need to have 10GbE upstream to connect the Vblock.
4:12 – Vblock type 1 uses either dual MDS 9222i or dual MDS 9506, 8 4 Gn N-port to each FI, 4 to 8GBE
4:13 – Storage – Type 1 uses CX4-480 with a blend of EFDs, FC and SATA sales to deliver from 43 TB, 41000 IOPS to 68 TB, 50000 IOPS. Type 2, Symmetrix V-Max, 8 to 16 4Gbps fibre channel. 140 TB/92000 IOPS to 211 TB/140000 IOPS.
4:20 – Storage details are up on the screen. I’ll post them later.
4:25 – Software = Vblock = VM container based on Enterprise Plus. Nexus 1000V is used to manage virtual switching across the Vblock. PowerPath/VE to enable path management. Ionix Unified Infrastructure Manager (UIM) is strongly recommended.
4:43 – Ok… I really like this Vblock concept.
5:06 – Done, folks! Listening too intently to keep typing!