Q&A for TechWiseTV Workshop: HyperFlex

Q&A
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public. Page 1 of 5
Cisco HyperFlex Systems: Agile, Efficient, and
Adaptable Infrastructure
April 27, 2016
Q. Does HyperFlex only work with VMware? Is Red Hat Enterprise Virtualization (RHEV) an option?
A. VMware only today. RHEV and other KVM-based hypervisors are a roadmap item.
Q. Does this solution work with Hyper-V or Azure stack?
A. No Hyper-V today. vSphere is the only supported hypervisor, with other hypervisors being roadmapped.
Q. Architecturally speaking, why has Cisco elected to go with distributed I/O parallelism instead of data
locality? It has been said that all the public cloud vendors (for example, Google and AWS) use data
localility technology.
A. The IOvisor serves as a broker that distributes I/O to local or remote controller VMs. The IOvisor can read
locally or remotely, so it can take advantage of NVMe and 3D NAND. We use the network. This gives us additive
performance at the cache and capacity tiers. It also gives us multiplicative I/O performance on reads from cache.
At the end of the day, the bottleneck isn't the network; it is the I/O being queued to storage. Architectures using
data locality aren't using an IOvisor-like technology.
Q. Can the nodes deliver a SMB/CIFS share critical to a VDI environment without additional servers?
A. The HX platform doesn't provide native file services today. However, you can of course create an OS-level
share for a VDI solution.
Q. What actual VMware products are preinstalled?
A. vSphere 6.0 Update 1.
Q. Is there a time frame on the storage sizing calculator so you can figure out usable space? I work for a
company where our customers keep asking how much usable space there is per node, and we don't know
the answer.
A. Rough sizing with RF3: HX220s are approximately 2 TiB per node, and HX240s are approximately 7.7 TiB per
node.

Q. Is this a pure virtualzation platform? Can Cisco UCS® servers be used for physical workloads?
A. Today it's supporting just the vSphere hypervisor. The HX appliances can't be used for bare metal
nonhypervisors, but there is potential to support it, and it's being considered.
Q. When can we expect Hyper-V support? Can we connect to external LUNs using a Cisco UCS fabric
interconnect?
A. Contact your Cisco account rep for an NDA roadmap session for HyperFlex. Yes, you can map external LUNs.
Q. Can the data platform do erasure coding?
A. The HyperFlex data platform doesn't support erasure coding today. Erasure coding is a balance between
capacity and performance. With our native dedup/compression, which is always enabled, we can negate the need
for erasure coding and the effects of using such a feature.
Q. Is there native replication?
A. We support Vsphere replication with SRM, and we have solution briefs with Veeam and Zerto. Native replication
is coming in the future.
Q. Can we also deploy the HyperFlex service profile through Cisco UCS Central for existing domains?
A. Cisco UCS Central support is on the roadmap. Today the installer pushes service profiles directly to the fabric
interconnects.
Q. Can the striping feature be turned off, maybe per virtual machine? For instance, if I'm running low on
space, can I only stripe data for the critical ones while restricting the less critical ones to their nodes to
save space?
A. The replication factor is configured on a per cluster basis. This would not be a good way to save space. Striping
is a native part of the file system and cannot be turned off.
Q. How many copies of data does the HyperFlex data platform support? RF3?
A. RF3 (for production) or RF2 (for nonproduction).
Q. Does the data rebuild on remaining nodes, or is there a lag or defined maintenance mode?
A. Data is rebuilt into the free space in the cluster. As long as there is free space, we can continue to suffer drive
losses and rebuild. Node failure will either resync when it comes back or fully reprotect the data if the node has an
extended outage.
Q. If a customer buys a SmartPlay HyperFlex bundle with a one-year subscription, is there an SKU/process
to upgrade to a three-year subscription? What part number do we use to renew the subscription after the
first year or third year is complete?
A. The upgrade SKUs will go live soon in Cisco Commerce Workspace.

Q. Are there any native cloud backup technologies?
A. No native cloud backup technologies today.
Q. Is all the data copied to other nodes, or just the metadata?
A. Both data and metadata are copied between nodes.
Q. Is it correct that storage is not centralized (as in a tray of disks) and instead is local to each node and
shared by the entire cluster?
A. Yes. Each HX appliance has local storage. Controller VMs and the IOVisor distribute the log structured file
system between nodes.
Q. What is the maximum cluster size?
A. Today it is 8 HX nodes or 8 HX nodes and 4 B200 compute-only nodes.
Q. Can writes go to memory and/or SSD?
A. Writes always go to SSD for persistence. There is never a data-in-flight issue. Reads can be served from both
SSD and memory.
Q. What happens if I need additional storage, but I am not adding any more ESXi hosts?
A. You can scale up in the HX240 by simply adding more disks. A new disk is immediately claimed and available
for use on your vSphere datastores.
Q. If I want to store 1 TB, do we need to have 1 + 2 (redundancy) TB storage?
A. The calculation is a bit more complicated because of the deduplication/compression. But there will exist three
copies of each block in the cluster.
Q. Is your deduplication inline, and what does that do to performance?
A. Deduplication is inline before being persisted to disk. There is no performance effect because this occurs during
destage, and it was specifically designed as a native feature of the file system.
Q. Are there any plans for erasure coding for RF3 Does that make sense in this architecture?
A. Erasure coding is a balance between capacity and performance. With our native deduplication/compression,
which is always enabled, we can negate the need for erasure coding and the effects of using such a feature.
Q. What hypervisors are supported? Is it only VMware? If so, are there plans (roadmap) to support other
hypervisors?
A. vSphere today and other hypervisors will be supported. Contact your Cisco account maneger for an NDA
roadmap session.

Q. How many nodes can effectively be in a cluster without challenges? My company runs tens of
thousands of servers that we are looking to VM.
A. We support 8-node clusters with HX appliances and then can go to 12 nodes by adding 4 B200 M4s. Each
fabric interconnect can support 32 HX nodes (4 clusters) or 48 with the B200 M4s.
Q. For updating firmware and upgrading versions, is that true one-click hardware updates, one-click
software udpates, and one-click hypervisor updates exist? Can we just click Update and go get a coffee,
and it will done when we return?
A. One-click upgrade is coming soon to HyperFlex.
Q. Why is there no performance impact when doing inline deduplication and compression? Is there a node
set apart that primarily does these activities?
A. There is no impact because the data has already been written and acknowledged in the SSD. Therefore, we can
deduplicate/compress as we destage to HDD.
Q. Is there a REST API? One of the slides showed a CLI for automation.
A. Yes, there is an API in the technical preview, and it is REST based.
Q. Can HyperFlex nodes be outfitted with only flash for cache and storage?
A. Today it's SSD and HDD or hybrid. All-flash HX appliances aren't available yet.
Q. Does it support Hyper-V?
A. Hyper-V is on the roadmap. Contact your Cisco account manager for an NDA roadmap discussion.
Q. What type of data protection do you recommend with your deduplicated and compressed datastores?
A. Works well with traditional backup data protection methodologies such as Veeam and so on.
Q. Is Cisco claiming there isn’t any performance hit with compression and deduplication turned on by
default? I do not understand how that is possible. Having those features turned on in its very nature is
overhead in some fashion, right?
A. It's software deduplication/compression using host CPU and RAM, so there are no large performance hit and no
effect on virtual machine density on the hosts.
Q. Is there built-in replication to a DR site?
A. No, there isn't built-in replication today. You would use third-party replication such as Veeam or Zerto or
vSphere replication and SRM.
Q. Do you have the option to turn off compression and deduplication?
A. No, the deduplication/compression is always on.

Q. Why are jumbo frames required, and what happens if we do not use jumbo frames?
A. Jumbo frames are a best practice but not required. Customers can elect to use regular size Ethernet frames if
they want to.
Q. How much would not using jumbo frames affect performance?
A. That would depend on your requirements. Vmotion and VM traffic would run more slowly on 1500-byte frames
than 9000-byte (or 9216-byte) frames.
Q. Springpath had a cloud monitoring service (like pure), but it doesn’t look like Cisco picked it up. Is that
true?
A. Contact your Cisco account manager for an NDA roadmap about the cloud monitoring service.
Q. Will users be able to use their existing 6296 fabric interconnects that they use with their existing Cisco
UCS chassis systems, or is this solution always going to require separate fabric interconnects?
A. Today customers cannot use their existing Cisco® fabric interconnects. With vSphere 5.5, it’s Update 3
minimum.

Q&A for TechWiseTV Workshop: HyperFlex

Recommended

More Related Content

What's hot (19)

Similar to Q&A for TechWiseTV Workshop: HyperFlex (20)

More from Robb Boyd (20)

Recently uploaded (20)

Q&A for TechWiseTV Workshop: HyperFlex