Original URL: http://www.theregister.co.uk/2009/08/07/netex_hyperip/
Where does WAN acceleration by virtual machine actually get us?
A tour of the estate
Opinion NetEx is claiming its virtual machine-based WAN acceleration product, HyperIP, supports more software applications than any other similar product. Software like this is taking us towards a virtual mainframe environment.
NetEx's VM-based WAN acceleration converges WAN optimisation and servers. It also complicates the server software estate.
Data acceleration across a wide area network is mostly carried out these days by appliances that sit on the IP link and reduce TCP/IP chatter while also reducing the amount of data sent across the wire. NetEx, Riverbed and Silver Peak all have boxes and technologies that do this.
One NetEx product is NetEx/IP. Like the other two technologies it is software running in an embedded server. NetEx took the software functionality and launched it in October 2003 as HyperIP software that could run on a Linux/X86 server. HyperIP was packaged as a VMware virtual machine (VM) in the first quarter of last year.
The latest NetEx announcement says that the VM version of HyperIP works with more third-party apps than anyone else in the WAN optimising game. Supported apps include ones from all the main providers of disaster recovery, data migration and replication software, such as Dell/EqualLogic, EMC, including Data Domain, FalconStor, Hewlett-Packard/LeftHand, Hitachi Data Systems, IBM, Microsoft, Network Appliance and lots of others.
NetEx says: "Virtualizing applications for VMware eliminates the need for specialized appliances."
So let's envisage a set of physical servers that have been virtualised and run an application software estate. Whenever these apps need to squirt data across a WAN they set up a connection and pass it through the network infrastructure to the far end of the link. Unbeknownst to them a stealthy interloping appliance would squat on the link and massage their data communications so the far end receiver got the data quicker and with less network bandwidth required.
The data centre management structure for this was, in theory, simple. Application people looked after the apps, server people the servers, networking people looked after the networking while storage people looked after the storage. These boundaries are blurring. The NetEx HyperIP VM collapses a network appliance's functionality into the VMware server estate.
HP's Left Hand Networks VSA product runs a SAN storage application as a VM. So too does DataCore's Virtual SAN Appliance software. The storage array controller functionality becomes, like the HyperIP functionality, just another system app in the VMware server universe. Cisco has developed a software Ethernet switch that runs as a VM. Here are three examples of previously separate networking and storage technologies, running on their own dedicated hardware, that have their software functionality extracted and turned into VMs.
In the VMware universe everything is an app. We are seeing suppliers thinking that, where their product has software functionality running on a dedicated server, X86 hardware can do it faster and cheaper than dedicated hardware, unless that is you use specially developed ASICs and FPGA chips to accelerate past current X86 hardware limits; BlueArc and its filers for example.
Once these suppliers transition their software to an X86 hardware platform then the door is open to a further transition to running it as a VM on X86 hardware. The thing is, many people will realise, that a dedicated hardware appliance isn't busy all the time. It sits there, costing money and doing nothing when idle. Whereas, the accountants say, if it was a VM it would use a shared X86 server platform base and cost less when it was doing nothing.
It means though, that data centre management and admin has to change to reflect this. Just because everything is an app in VMwareland doesn't mean that existing application and server admin people can handle networking and storage system apps. The data centre admin organisation is going to have to get networking and storage smarts to cope.
We're also able to get to the point where larger enterprises could buy substantial unified bladed server, networking and storage systems, virtualised mainframes, from the likes of Cisco (UCS) and HP (Matrix) and, no doubt IBM, with one-throat-to-choke support. But smaller enterprises could end up heading towards the same mainframe-like approach if they go full throttle into server virtualisation and adopt VM-based storage and networking functionality as it becomes available on the "one box fewer to manage is goodness" principle.
The counter to this saving is that there are still many throats to choke when service gets interrupted. There is a reseller opportunity here, to unify the potentially fragmented and divisive support scene.
Another thought; if a customer has to buy specialised hardware and software to get a needed data centre facility, such as network routing, SAN fabric switching, filer storage or WAN optimisation then changing suppliers is not so easy. If that storage and WAN optimisation functionality is disaggregated into software running in a VM and a storage JBOD (just a bunch of disks) or router hardware, given an identity or functional personality by that software, then changing the software becomes easier. It's just, say for argument's sake, a SAN iSCSI block access layer between apps needing block data and a JBOD array providing the disks to do it, with standard interfaces either side.
So you could swap in another supplier's VM-based iSCSI block-access software product quite easily. Ditto filer. Ditto networking functionality, in theory, so long as you have a JBOD networking equivalent. We don't of course. Virtually all data storage outside of a server's DRAM is hard disk drive based. Fundamentally every storage array is a JBOD with ornamentation and uses substitutable hard disk drives from the same small set of HDD suppliers. The same isn't true for networking. There isn't the router or switch equivalent of a hard drive, more's the pity.
We won't see Cisco collapsing its networking functionality into a set of VMs and abandoning basic network hardware supply to the equivalent of Hitachi GST, Seagate and Western Digital any time soon. Sun's hardware and software strategy, with its open storage and open networking elements, was aiming in this direction though. Whether Oracle will continue that thrust remains to be seen. ®