Monday, 30 March 2009
Nehalem Day - start of a new tin era?
Well today marks the day when new x86 server ranges from the core market players HP, Dell and IBM all released new generation servers based on the i7 Nehalem chipset. Its probably the first time that chipsets and server ranges have been built and tailored for Virtualisation and associated workload and not the other way round where VMware is plonked on top of a big beefy multi core server with loads of Layer 3 cache.
Nehalem has changed the game, it is built to cater for Virtualised workloads and most importantly to work effectively with VMware and i expect in future other Hypervisor vendors (kinda helps VMware have Intel as an investor though).
Various features within the chipset will most certainly benefit, some examples when compared to the older chips includes improved memory controller access and speed due to Quickpath technology being implemented onto the CPU Die rather then externally on the Northbridge, flexible vmotion migration between generations of Intel Chipsets to ensure we don't need to go and buy massive clusters worth of kit, general power reduction and Hyper Threading comes back from the dead.
For Virtualised environment newer architectures bring benefit with Extended Page Table (NPT in the AMD world), this feature was shown with a whitebox prototype way back at Vmworld 2007 on a keynote. This evolves upon how current Shadow Page Table algoritms work to allocate memory register within ESX's. SPT has brought benefits which has allowed us to obtain higher consolidation ratios with lower utilised machines in the first phase of VMware maturity in most organisations. Next level demand for higher intensive workload can be catered for in most cases by NPT/EPT. Both nehalem and AMD barcelona/shanghai are able to offer built page table management within the actual CPU and remove associated overhead which was previously incurred when new page requests were required at the hypervisor layers. Applications and environments which have large amounts of Processes running within the box i.e. Citrix/terminal server will benefit from this, but not all environments will benefit as greatly as these.
When I was at the Intel stand in Cannes i asked the geezer if Nehalem was 8 Core, he said "oh yes its eight core blah blah", being the skeptic I am I investigated, it turned out it was just dual threaded or more better known as Hyper Threaded. I seem to remember in the early days hyper threading caused more hassle than it was worth, it kind of didn't actually matter then, OS and application was able to interact. This new generation I feel will change the mould, instead of the yesterday era of single core processors which were just logically split and presented to a host OS, we now have a friendly more environmentally, easier to manage etc etc layer of indirection called the Hypervisor, couple this with neat new features in the vSphere technology and this will hopefully work better than previous incarnations did.
I don't believe that adding continuous amounts of cores is the answer, look at Sun UltraSparc and this works in a similar fashion http://www.sun.com/processors/UltraSPARC-T2/, it basically has multiple threads making use of memory lag, this allows more CPU cycles to be made available to the virtual machines. Will be a shame if IBM buy Sun as I think they will just EOL this architecture.
Overall Nehalem benefits are application/workload dependant, be very wary of buying it with the intention that it will solve your problems, like most things in todays world its mostly hype and isn't a silver bullet, you may find that going with AMD will benefit, they have offered Nested page table support with RVI for a long time now and it is mature.
So...now onto arguing whether its worth waiting a week or two for this kit and also badgering vendors/resellers to find out prices and lead time!
Nehalem has changed the game, it is built to cater for Virtualised workloads and most importantly to work effectively with VMware and i expect in future other Hypervisor vendors (kinda helps VMware have Intel as an investor though).
Various features within the chipset will most certainly benefit, some examples when compared to the older chips includes improved memory controller access and speed due to Quickpath technology being implemented onto the CPU Die rather then externally on the Northbridge, flexible vmotion migration between generations of Intel Chipsets to ensure we don't need to go and buy massive clusters worth of kit, general power reduction and Hyper Threading comes back from the dead.
For Virtualised environment newer architectures bring benefit with Extended Page Table (NPT in the AMD world), this feature was shown with a whitebox prototype way back at Vmworld 2007 on a keynote. This evolves upon how current Shadow Page Table algoritms work to allocate memory register within ESX's. SPT has brought benefits which has allowed us to obtain higher consolidation ratios with lower utilised machines in the first phase of VMware maturity in most organisations. Next level demand for higher intensive workload can be catered for in most cases by NPT/EPT. Both nehalem and AMD barcelona/shanghai are able to offer built page table management within the actual CPU and remove associated overhead which was previously incurred when new page requests were required at the hypervisor layers. Applications and environments which have large amounts of Processes running within the box i.e. Citrix/terminal server will benefit from this, but not all environments will benefit as greatly as these.
When I was at the Intel stand in Cannes i asked the geezer if Nehalem was 8 Core, he said "oh yes its eight core blah blah", being the skeptic I am I investigated, it turned out it was just dual threaded or more better known as Hyper Threaded. I seem to remember in the early days hyper threading caused more hassle than it was worth, it kind of didn't actually matter then, OS and application was able to interact. This new generation I feel will change the mould, instead of the yesterday era of single core processors which were just logically split and presented to a host OS, we now have a friendly more environmentally, easier to manage etc etc layer of indirection called the Hypervisor, couple this with neat new features in the vSphere technology and this will hopefully work better than previous incarnations did.
I don't believe that adding continuous amounts of cores is the answer, look at Sun UltraSparc and this works in a similar fashion http://www.sun.com/processors/UltraSPARC-T2/, it basically has multiple threads making use of memory lag, this allows more CPU cycles to be made available to the virtual machines. Will be a shame if IBM buy Sun as I think they will just EOL this architecture.
Overall Nehalem benefits are application/workload dependant, be very wary of buying it with the intention that it will solve your problems, like most things in todays world its mostly hype and isn't a silver bullet, you may find that going with AMD will benefit, they have offered Nested page table support with RVI for a long time now and it is mature.
So...now onto arguing whether its worth waiting a week or two for this kit and also badgering vendors/resellers to find out prices and lead time!
Subscribe to Posts [Atom]
Post a Comment