On Mon, Feb 27, 2012 at 7:03 PM, George Herbert <george.herbert@gmail.com> wrote:
On Mon, Feb 27, 2012 at 3:45 PM, William Herrin <bill@herrin.us> wrote:
universe does $30/mo per customer recover that cost during the useful life of the equipment? As I stated, one can either do it with SANs or with alternate storage.
Should not assume Rackspace et. all provide any level of fault tolerance for extraordinary situations such as hardware failure beyond what they have promised or advertised. IaaS provided redundancy is not always necessary, and may be unwanted in various situations due to its cost; single redundancy means a minimum of twice the cost of non-redundant (plus the overhead of failover coordination). With various computing applications it may make a great deal of sense to handle in software -- should a node fail, the software can detect a node failed, eject it, reassign its unfinished work units later. Typical Enterprise level fault tolerant SAN manufacturer prices seen are ~ $12 to $15/GB usable storage, for ~50 IOPS/TB, data mostly at rest, and SAN equipment has a useful lifetime of 5-6 years; a typical 200gb server then exceeds $30/month in intrinsic FT SAN hardware cost. There IS a place for IaaS providers to sell such product, probably at four or five times the $/month for a typical server. Just like there is a place for Network service providers to sell transport and network access products that have redundancy built into the product, such as protected circuits, multiple-port, dual WAN , that can sustain any single router failure, etc. Those network products still can't reliably guarantee 100% uptime for the service. There is a place for IaaS providers to sell products where they do NOT promise a level of reliability/fault tolerance or performance guarantee that requires them to utilize an Enterprise FC SAN or similar solution. Just like there is a place for NSPs to sell transport and network access products that will fail if a single router, card, port fails, or if there is a single fiber break or erroneously unplugged cable. This way the end user can save on their network connectivity costs; a tradeoff based on the impact of the difference between their network with 1% downtime and their network with 0.001% downtime; VERSUS the impact of the cost difference between those two options. End users may also prefer to implement their redundancy through dual-homing via multiple providers. It is very important that the end user and the provider's sales/marketing know exactly which kind of product each offering is.
Amazon hit those price points with a custom distributed filesystem that's more akin to the research distributed filesystems than anything
Amazon is quite unique; developing a custom distributed filesystem is a rather extraordinary measure, that provides them an advantage when selling certain services. But even EC2 instance storage is not guaranteed. The instance storage is scratch space If your instance becomes degraded, and you need to restart it, what you get is a "clean" instance, matching the original image. EBS and S3 are another matter. The same provider that offers some 'unprotected' services, may also offer some more expensive storage services that have greater protections. -- -JH