three blocks

Analysis

Flash-enhanced servers: Intel and Microsoft

posted on 18 July 2008 08:29


Unix and Linux too - and VMware

Flash-enhanced servers promise much and Sun, Dell and HP are aboard the flash server train. What about Intel, Microsoft, Linux and Vmware?

If Sun is right that flash-enhanced server hardware and flash-using amendements to system software are going to substantially boost a server's I/O processing then where does that leave Intel, Microsoft, IBM (SuSe), Red Hat and VMware?

Currently an x86 server is an x86 server and an x86 server O/S is an x86 server O/S. Sun wants to change that. A flash-enhanced SunFire server with flash-amended Solaris+ZFS+what-else system software could go an awful lot faster than an industry-standard x86 server running Windows or Linux.

Noting that Intel has a flash memory solid state drive (SSD) operation from which exuberant performance noises have been coming, we saw a Microsoft employee Usenix paper (downloadable below) about design tradeoffs in flash SSDs. Why should Microsoft be interested in .... of course ..... suppose Microsoft is interested in ensuring that Windows Server does not cede any performance advantages to Solaris and wants to help Dell and HP get the most flash-assisted performance out of their servers?

Dell and HP engineers will be very cognizant of system software changes that will keep their servers performance status up to the mark compared to flashed SunFire/Solaris boxes. They don't write system software and so will be exploring flash system software changes with their main operating system suppliers - Microsoft and Red Hat.

We might hope to expect therefore Windows Server extensions next year designed to take advantage of a flash memory SSD added to a server between its DRAM and the hard disk drives.

Intel will need to be involved to standardize the architecture. It won't want there to be multiple versions of flash-enhanced server hardware as that will weaken the x86 industry standard architecture. We might envisage a flash version of the Storage Bay Bridge working group which brings together Intel, Dell, HP, IBM, Fusion-io, Spansion, AMD, RedHat, IBM (SuSE) and other interested parties to define a flash addition to the x86 architecture.

A prize for Intel would be the opportunity to influence an architecture to add a flash SSD to a server motherboard and so give it the ability to add its SSD to the server chipset bundle.

Red Hat and IBM would want to be involved with flash-enhanced servers so that their Linux software doesn't get left behind.

Another interesting thing here. Such flash-enhanced servers would become the basis for Windows Storage Server-based network-attached storage (NAS) products giving such filers an I/O boost prospect.

Lastly, VMware, which is now a prime piece of system software, will face the prospect, perhaps, of Microsoft's Hyper-V using flash-enhanced server software. Oops. EMC will probably be alert to this possibility already and will want VMware to take advantage of flash-enhanced servers too. If this comes to pass then EMC storage hardware people will stop dissing flash as a server I/O boost and start talking about different horses for different courses as they do with deduplication.

[Chris Mellor.]



Download file: Desig Tradeoffs for SSD performance.pdf


tags:  flash SSD