Cisco SDS – Chan's Blog

Cisco has just announced their newest datacentre infrastructure solution code named HyperFlex. This is a fully integrated Cisco proprietary Hybrid Hyper-Converged solution offering (similar to Nutanix / Simplivity / VMware VSAN) that consist of the followings

Cisco UCS C series rack mount servers (C220 or C240) with local storage (SSD+HDD)
Software Defined Storage virtual appliance (VSA)

HCI market is quite busy at the moment and there’s lot of demand as it was only natural that Cisco would join the party with their own offering to compete with the incumbents such as Nutanix, Simplivity and VMware’s software defined HCI solution based on VSAN.

While the UCS C series servers are nothing new and are the same familiar rack mount servers, the real introduction here is the SDS solution which effectively comes from their strategic partnership with Springpath and is what is worth considering. (For those who aren’t familiar, Springpath was supposedly founded by ex VMware engineers so naturally you’ll find many similarities between this SDS solution and VMware VSAN)

Now I’m not intending to cover the HyperFlex offering in depth here within this article but will highlight few key points and what I think of the solution and compare it to some competition.

Hardware

As mentioned above, the hardware consist of 2 configuration choices on day 1 as follows,

HyperFlex HX220c M4 (UCS C220 rack mount server based)
- CPU = 2 x Intel Xeon E5 2600 v3 processors
- Memory = 256Gb to 512 Gb 2133 MHz DIMMs
- Caching layer = 480GB high-endurance (Intel 3610) cache SSD
- HDD = 6 x 1.2 TB 10,000 RPM 12-Gbps SAS disks
- Network = Cisco VIC 1227 (10gbe x 2)
- Software
  - VMware 5.5 or 6.0 u1
  - Cisco HyperFlex HX Data Platform Software version 1.7
- Cluster
  - Nodes: Minimum of 3 nodes, Maximum 8 nodes (Initial version. Will increase in future)
  - Management: Cisco UCS Manager and vCenter plugin
HyperFlex HX240c M4 (UCS C240 rack mount server based)
- CPU = 2 x Intel Xeon E5 2600 v3 processors
- Memory = 256 Gb to 784 Gb 2133 MHz DIMMs
- Cache = 1.6-Tb high-endurance (Intel 3610) cache SSDs
- HDD = 15 x 1.2 TB 10K RPM 12gbps SAS disks (8 additional disks supported through SAS expander)
- Network = Cisco VIC 1227 (10gbe x 2)
- Software
  - VMware 5.5 or 6.0 u1 (VMware only on day 1. Additional hypervisor may follow)
  - Cisco HyperFlex HX Data Platform Software version 1.7
- Cluster
  - Nodes = Minimum of 3 nodes, max 8 nodes per ESXi Cluster (Will likely increase in future).
  - Management = Cisco UCS Manager and vCenter plugin

* Note that the hardware configuration on all nodes MUST be same (cluster validation fails otherwise)

Key Architecture Points

The HyperFlex SDS offering is a VSA (Virtual Storage Appliance)
- In other words, its very similar to Nutanix or Simplivity’s implementation as its just a VM running on a Hypervisor (ESXi) that acts as a virtual storage appliance.
- This is markedly different to VMware VSAN which is kernel based and may offer better scalability (though there are pros and cons to both in kernel and VSA type of architectures)
- Springpath software will create a distributed NFS datastore that is made available to all the hosts in the cluster (It will rely on a control VM residing on each ESXi host to serve the IO).
- Note however that unlike Simplivity’s offering, there’s no dedicated hardware to offload any de-dupe or compression work to so its all done in the SW (tax on CPU cycles)
Launch & Initial positioning
- Cisco launched this solution internally within Cisco on the 25th of February followed by the public launch on the 1st of March
- As a version 1 product, the initial target market for Cisco HF would be,
  - VDI deployments
  - Small to medium scale virtualisation (vSphere only) environments
  - Test and Dev environments
  - Branch office requirements
- VMware vSphere based only
  - Future roadmap may include other hypervisors naturally
  - Support VAAI
  - vVol support is not there on day 1 but on the roadmap
  - vCenter plugin available
- No native replication
  - Relies on application level data replication such as that of VMware vSphere replication or Zerto
- Inline de-duplication is available on day 1 (always on. Approx 20-30% savings)
- Inline compression (during de-staging from SSD to HDD) is also available on day 1 (variable block size, 30-50% approx. savings)
Scalability (day 1)
- Independent compute nodes independently from storage nodes
  - 3-8 HF nodes per cluster + up to 4 additional compute nodes (has to be UCS B200 blades only)
  - When adding compute only nodes, HF cluster will automatically push a software component, referred to as IOVisor on to the new compute nodes (in the form of an ESXi VIB)
- Hybrid HCI (SSD for caching and HDD for capacity) on day 1
  - No all flash offering available for now
  - Similar to competition, all disks are in pass-through mode (no local RAID)
Unique Selling Points
- Unlike other HCI offerings in the market that typically do not include the networking components, Cisco HF solutions include the Networking elements. Full content included within the HF bundles are as follows
  - Compute Nodes with local storage
    - Comes with ESXi pre-installed
    - A wizard driven installer VM is used to simplify the initial deployment
  - Software license subscription for SDS (Yes you read it correct, its not a perpetual license but only a annual or 3 year subscription that need to be renewed)
  - Cisco UCS Fabric Interconnects for server HW managements
    - Pre-configured for rapid deployment
    - Include single UCS domain license
    - 48 or 96 port option for FI’s
    - Complete UCS manager software (used for hardware management only and no Springpath SDS management capabilities will be included on day 1)
- Unlike all other competitive offerings, Cisco HF uses a dedicated 120GB SSD (separate from caching SSD’s) for data logging (meta data) on each host which should help with performance & scalability
Ordering Options
- Orderable through normal channel partners as per usual process
- Pre-Defined bundles available
- Configure to order option also available

It is important to note that Cisco HyperFlex is not offered as a replacement or an alternative to converged architecture solutions that Cisco already excels in, such as FlexPod or VBlock but only offered as another silo option for appropriate use cases. Industry analysts predict that the Hyper-Converged market may be worth in the region of $3 billion and this is Cisco’s answer for their customers.

The marketing message around HF is going to be focused around its Simplicity, Speed to provision and scalability (linear, node based) which is no different to other HCI vendors such as Nutanix.

My Take

I think HyperFlex is a good version 1 HCI solution from Cisco and I like number of things it has to offer such as its cheaper cost (in comparison to competition) and the fact that it automatically include the networking and Fabric Interconnect modules within this cost. Architecturally it looks solid too, however there are some minor things that need to be addressed / improved which I’m sure will get addressed as the product evolves (quite normal for a version 1 product). Its designed from ground up to ensure the availability and the integrity of your data which means if there are multiple simultaneous node failures for example, that takes the HF cluster beyond the configured availability (replication) levels (similar to VSAN FTT) it will offline the cluster to ensure the integrity of your data.

However being pragmatic, I would personally like to wait and see how this performs out in the field with real customer data under normal working conditions. While established solutions like VMware VSAN may provide fully integrated HCI solutions for vSphere at a much more deeper level than a VSA based solution would provide, if you are a Cisco house and are happy with UCS server hardware (who wouldn’t btw? they are just awesome…!!), this solution may appeal to you quite easily.

I would urge you to register for the webcast (link here) to find out more or reach out to your Cisco AM or reseller to find out more (my employer Insight can help too)

In the meantime, additional information can be found here

Image credit goes to Cisco…!!