Getting started with PernixData Architect

So how well does your application perform? This is the question that everyone want to know the answer to, its kinda like the meaning of life (from an IT-pro perspective)

In order to understand how an application performs we can of course look at the memory and cpu usage, but the question that remains uanswered in most cases is, WTF is it doing with the storage system? Enter PernixData Architect

Now I have previously written about PernixData and FVP Architect uses the same agent on the ESX host, but instead of doing acceleration it does data analysis.

Now many companies that do data acceleration or sell storage often tell you how much IOPS it can do with a specific block type. Now using Architect we can actually see what kind of block sizes our virtual machines are generating and latency / troughput.

The installation process is pretty much the same as FVP we installed the host extension, install the management server which runs on SQL server.

(NOTE: I have a pretty small lab at the moment, running nested ESXi on Vmware workstation) so I only have a few virtual machines

Have to say that the UI is quite an improvement


We can use Architect and FVP within the same UI (Just switch between the dashboard menu in the top) We can do drilldown into different metrics


Which allows us to see latency/iops and what kind of workloads that are running. We can also see the block size breakdown which average IOPS


Now eventually when this is integrated with the Cloud solution that PernixData is building, it will give awesome insight into how we can configure our storage properly and what we can expect of our storage system. And we can eventually get some knowledge on how to properly configure applications and how they operate on a storage level.

#architect, #pernix-data

Regular VM vs Pernix FVP write back vs FVP write trough

So since I’ve been working on FVP the last couple of days, I decided to go a simple test. First of I have one VM which is not accelerated by anyway, then I move it to a write trough mode (meaning that writes are not accelerated since Pernix needs to maintain writes to the datastore. Last I decided to setup a Write back (meaning that writes are stored on the cache and then back to the datastore.

So just a regular file benchmarking test….


Now using write back we can see that IOPS on writes are close to the same as it was before, but the reads are accelerated by the cache


This was the test using Write back, on creating a file on 2 GB I was closely to 6000 IOPS 4K.


The problem with write back is that writes are stored on the ram cache in this case, but Pernix has a feature called Fault tolerant write back, meaning that all writes are replicated to another host in the cluster.

And note you can use Add-PrnxVirtualMachineToFVPCluster -FVPCluster pernix -Name felles-sf -NumWBPeers 1 –WriteBack

to move a virtual machine to a writeback cluster.


Software defined storage and delivering performance

I had no idea what kind of title I should use for this post, since this is more about to talk about different solutions which I find interesting for the time beeing.

The last couple of years have shown a huge growth in both converged solutions and software defined X solutions (Where the X can stand for different types of hardware layers, such as Storage, Networking etc)

With this huge growth, there are alot of new “player in the field” which are in this space, this post is more to show some of these new players and what their capabilities are, and most importantly where they fit in. Now I work mostly with Citrix/Microsoft products and as such there is often a discussion of VDI(meaning stateless/persistent/rdsh/remote app functionality)

and a couple of years ago when deploying a VDI solution you needed to have a clustered virtual infrastructure running on a SAN, and the VMs where constricted to the troughput of the SAN.

Now traditional SAN’s mostly run with spindel drives since they are cheap, and has huge storage spaces. For instace a PS6110E Array Has the ability to house up to 24x 3,5” 7,200 RPM disks.

Which can then be upwards to 96TB of data. Now if you think about it, regular spindel disks have about roughly 120 IOPS (Depending on buffers, latency and spindels) and we should have a kind of RAID set running on the array for redundancy across disks as well. Using 24x drivers with RAID 6 and double parity (not really a good example but just to prove a point) gives us a total IOPS of 2380, which is lower then my SSD drive in my laptop. Now of course most arrays come with buffers and caches in different forms and flavors so my calculation is not 100% accurate. Another issue with using a regular SAN deployment is that you are dependant on having a solid networking infrastructure and if you have some latency there as well it affects the speed of the virtual machines. So in summary

  • regular SAN’s are built for storage space and not for speed
  • SAN’s also in most cases need their own backend networking infrastructure

And based upon these two “issues” many new companies have their starting grounds. One thing I need to cover first is that both Microsoft and VMware have both created their own way to deal with these issues. First Microsoft has created a solution with Storage Spaces with SMB 3.0. Storage Spaces is a kind of software raid solution running on top of the operating system and with features such as deduplication and storage tiering which allows data to be moved from fast SSD’s to regular HDD depending on if the data is hot or not. Storage spaces can either be using JBOD SAS or internal disks depending on the setup you want.  And with using SMB 3.0 we have features such as multichannel, RDMA. Both of these solutions makes it easier for us to build our own “SAN” using our regular networking infrastructure. But note that this still requires we have a solid networking infrastructure, but this allows us to create a low cost SAN with a solid performance.

Vmware has choosen a different approach with the VSAN technology. Instead of having the storage layer on the “other” side of the network, they built the storage layer right into the hypervisor.

Meaning that the storage layer is on the physical machine running the hypervisor meaning that we don’t have to think about the network for the virtual machines performance (even thou it is important to have a good networking infrastructure for the VM’s to replicate across different hosts for availability)

Now with VSAN, you need to fullfill some requirements in order to get started, since this solution runs locally on each server you need for instance to have a SSD drive for just the caching part of it, you can read more about the requirements here –>

So its fun to see that for one,
* Microsoft still has the storage layer outside of the host but dramatically improves the networking protocol and add storage features on the file server.
* VMware moves the storage layer ontop of the hypervisor to move the data closer to the compute roles.

Now based on these ideas there are multiple different vendors which in essence bases their solution on the same.

First of we have Atlantis ILIO, which is a virtual applicance which runs on top of the hypervisor. Now I’ve written about Atlantis before but in essence what it does is create a RAM disk on each host, and has the ability to use the SAN for persistent data (of course after the data has been compressed and deduped leaving a very small footprint) Now this solution allows virtual machines to run completely in RAM meaning that each VM has access to huge amounts of IOPS. So Atlantis also runs ontop of each hypervisor so it runs to close to the compute layer as possible and is not dependant on having high-end SAN infrastructure for persistence.

Atlantis has also recently released a new product called USX which is a more software-defined storage solution which allows to create pools of storage containing both local drives and or SAN/NAS (and not just a place to dump persistent data for VDI)

Secondly we have Nutanix, which unlike the others is not a regular software based approach, they deliver a hardware+software platform which has a kind of lego based approach, where you buy a node and compute and storage are locally and you can add more nodes to scale upwards. With Nutanix there are controller VM’s running on each node which are used for redundancy and availability. So in essence Nutanix have a solution which resembles alot of VSAN since you have the storage locally to the hypervisor and you have logic which is used for redundancy/availability.

And we also have PernixData which has their FVP product, which caches and accelerates reads & writes to the backend storage. Writes and reads are stored on the aggregated cache (which consists of either a flash drive such as Fusion-IO or SSD drives locally on each node) which allows IO traffic to be removed from the backend SAN.



Now there are also a bunch of other vendors, which I will cover in time. Gunnar Berger from Gartner also made a blogpost, showing the cost of VDI on different storage vendors But most importantly this post is to give a bit awareness of some of the different products and vendors out there which allows you to think differently. You don’t always need to invest in a new SAN or buy expensive hardware to get the performance needed. There is a bunch of cool products out there just waiting for a test-drive Smilefjes

#atlantis, #dell-dvs, #nexenta, #nutanix, #pernix-data