In my previous post (link here), I discussed 2 types of performance issues that we help fixing : storage-centric and application-centric and how infrastructure and application teams view them.

I would now like to discuss a different beast : Service delivery performance issues

These issues fall into the “It takes to long to provision this new service / application / environment to my users / customers” bucket.

There can be many storage-related reasons the delivery of a new service takes too long :

  • Creating new volumes just takes too long / involves such complexity that it cannot be done quickly
  • Copying data or VMs takes too long and/or has such a performance impact on production systems that I can only run it out of hours
  • Service levels in terms of time to deliver the new service/server/VM cannot be guaranteed because it might take days to provision storage

The infrastructure teams are usually painfully aware of these issues, since they are the ones who have to do all this manual work, or work out ways for the provisioning to happen out of hours. Application teams simply grind their teeth while waiting for the service to be provisioned, wondering why it takes so long to provision what sounds like a simple request…

I think it is fair to say that these issues are recognised by both teams and that they would both like them to be resolved.

Again, a Pure FlashArray is a great way of easily solving those challenges. Creating new volumes takes seconds and does not involve any complexity at all (this is a storage system with no options!). Copying data is instant thanks to our dedupe-assisted volume or VM cloning, and has absolutely no performance impact. Also because it takes minutes to install a new shelf and no rebalancing of storage is involved, new capacity can be provisioned extremely easily…

Thanks to our instant volume cloning, we can make copies of production databases available for development and test purpose instantly, in order to streamline the workflow of data within the enterprise. For an example of how we can make these copies available instantly and with no performance impact on the production systems, watch the following video (link here) where our very own Chas Dye demonstrates this with Oracle

We also provide metadata assisted virtual machine cloning which can be used to dramatically improve the delivery time of new virtual machines, as they allow the hypervisor to clone VMs with existing data (or templates) while copying only metadata within the array.

For example one of our customers had the following problem : they have an internal cloud used mainly for continuous integration by their developers. This continuous integration requires cloning new virtual machines quite often. This used to have quite an impact on their shared disk storage system : during the 45 minutes it took for the 50 VMs to be cloned, all the production systems would suffer from increased storage latencies.

By moving this Private Cloud platform to Pure and taking advantage of its deduplication assisted VM cloning with VMware vSphere, this customer has been able to improve the time it takes to provision these VMs by 36x! (going from 45 minutes for 50 VMs to 4 minutes 15 for 200 VMs) while at the same time making the storage performance issue during this operation simply go away…

Enabling Private Cloud with 100% software-defined service levels

Finally, Pure FlashArrays allow our customers to design and automate Private Clouds in a simple and future-proofed manner : through 100% software defined service levels

With traditional storage platforms, the way service levels have been defined is through hardware defined storage classes of service (different disk type, RAID level…) and an automation layer above it which determines where a resource is created.The challenge with this approach is that it’s only good as far as the present requirements are concerned. As soon as the requirements change it becomes very hard to have this infrastructure evolve…

With Pure FlashArrays, it is now possible to build a Private Cloud infrastructure with 100% software-defined service levels, where all the data sits on economical Flash all the time. With the performance never being a concern (whether for normal operations or during provisioning) and at an affordable cost, it is now possible to define, provision, automate and change service levels purely in software, without relying on hardware-defined classes of storage