Storage Magazine blog - laatste entry
09-02-2012 - Numbers don’t lie
I have been going through my annual ritual of checking all of the published statistics from industry ...
Lees meer Storage Magazine activiteiten
Geen evenementen gevonden.
Storage Magazine poll
STM November 2011
All-flash enterprise arrays hebben de toekomst.
LAN Magazine nieuws
Detail
Jon Toigo  |
 |
Jon William Toigo is directeur en eigenaar van Toigo Partners
en voorzitter van The Data management Institute LLC. |
12 september 2011 - Chokepoint
Whether from reading history or from watching Hollywood epics, most folks know about king Leonidas. He and three hundred Spartan warriors, aided by about 4.500 other Greeks who rarely get a mention, made a stand against a vastly superior Persian force at Thermopylae in 480 BC. Their heroic defence, which involved exploiting a geographical bottleneck or chokepoint on the Greek mainland and was ultimately circumvented by treachery, is the stuff of legend. Thermopylae has been the site of several similar battles since Leonidas, all less auspicious than the Graeco-Persian engagement and almost always fatal to the defenders. Conclusion: even the most celebrated strategies ultimately fall prey to reality. I fear this is also what is happening with server virtualization. If we don’t hold the line now, the entire land of storage standards may fall. Let me explain.
Funnel
Many IT planners who use server virtualization are beginning to realize that their hypervisor has created a Thermopylae-like chokepoint. This bottleneck obstructs LAN and storage traffic and slows the performance of guest applications. Workloads are aggregated into fewer servers and their collective I/O is funneled through a virtual SCSI controller, which provides no real traffic shaping or quality-of-service provisioning. This is creating havoc in many virtualization projects.
Consequently, storage hardware vendors had to spend even more ink on marketing brochures. They are already extolling array-based, cabling-based or driver-based work-arounds for the other storage issues created by server virtualization. But this one, the hypervisor-imposed I/O bottleneck, is beyond the scope of what switch vendors and box peddlers can address. Server hypervisors were meant to improve the efficiency of resource utilization by consolidating applications on virtual server hardware. In the real world, however, they are producing the opposite effect.
VAAI
Things have gotten so bad that VMware has implemented VMware vStorage APIs for Array Integration (VAAIs). This is a set of primitives intended to “more tightly integrate storage with virtual hosts.” VMware has arbitrarily added these commands to the SCSI set – that is, without normal consideration, review or approval by the ANSI committee that is responsible for the SCSI standard. The VAAIs are, at best, a way to reduce the I/O burden for things like data replication by about 20 per cent and offload it to smart arrays. And by ‘smart’, VMware means arrays that can parse the commands and perform the desired I/O functions without the involvement of the hypervisor itself.
Every vendor I talk to these days is scrambling to embed VAAI support into their array controllers in order to be classified as smart in VMware’s eyes. The new additions, however, only contribute to the already costly functionality bloat. One hardware manufacturer recently told me that VMware is driving engineers nuts. Not only are their SCSI add-ons inefficient, VMware has also come out with at least two versions of them in a very short time span. Both iterations were wildly different from each other and each necessitated another round of emergency engineering retrofits. One fellow said that normal quality-assurance testing of the results would probably have to be done by consumers. He was only taking into account the impossible matrix of quality assurance tests that vendors would have to complete prior to shipping. They just don’t have the time to do so within their short time to market.
Multihypervisor
Another big issue a storage vendor raised has to do with multihypervisor environments. Neither Microsoft or Citrix have bought into VAAI. Let’s say a customer has different hypervisors, such as VMware and Hyper-V or Xen. In that case, all traffic to the storage array will have to be inspected to discover which streams feature the non-standard SCSI primitives and which do not. But this kind of stateful traffic inspection will cost more I/O. The only alternative is to put the VMware guest traffic, and the storage which supports its workload, apart from everything else. This, however, creates more isolation and complexity in the storage infrastructure.
History shows that the Thermopylae strategy didn’t work out very well for just about anyone who tried it. So why embrace it again with server hypervisor technology? Moreover, why do we let hypervisor vendors make arbitrary changes to standard storage protocols? And why do we let them treat those fixes as innovations, when they are simply inefficient and imperfect patches to problems that the hypervisor vendor created in the first place? And, finally, why is no one even asking these questions aloud? It is said that those who ignore history are doomed to repeat it. So with all due respect to Leonidas: this is not Sparta.
22-02-2012 - Europees MKB benut opslagcapaciteit onvoldoendeVan alle kleine Europese bedrijven met 100 of minder werknemers heeft slechts 2 procent alle bedrijfstoepassingen naar de cloud gemigreerd. 85 procent heeft bedenkingen over een dergelijke stap. Dit blijkt uit gezamenlijk onderzoek van Dell en Intel.
Lees meer
Infosecurity nieuws
Storage Magazine zoeken
Laatste editie
Storage Magazine agenda
Geen evenementen gevonden.
Lees meer
Klik
hier om een evenement op te voeren.
Storage Magazine bloggers
Blog categorieën
Geen categorieën gevonden.
Blog jaren
Blog tags
Storage Magazine sponsor