Wednesday, February 9, 2011

Plant Bromothymol Blue Experiment

VDI storage solution "to high": the completely virtual local storage really cheap!

Brian Madden
See the original blog post

Last September I wrote an article describing a solution that I wanted to see, but that does not really exist: the possibility of having a local storage VDI virtual . In practice I was stressed because I did not like the SAN for VDI, and I wrote what would have been nice to have some sort of software that virtualizes access to local hard drives within a server host VDI. I was thinking of a solution that was able to put together the best of both worlds: a fast and flexible storage without the terrible cost of a SAN.

The new function "IntelliCache" present in XenServer 5.6 SP1 goes in this direction, although for now only works with disk images shared (while most of the current VDI solutions using persistent disks ).

In a new white paper produced by DataCore ( direct link to PDF), the company says that its SANmelody software solution that runs on a host VDI fulfills my fantasies on storage. It also states that with a full multi-server redundancy at a cost of less than $ 70 per user (or $ 70 for all that is needed: the host of the VM software, SANmelody, records necessary for the storage ... everything! ).

Regular readers know that usually do not re-released documents made by manufacturers. In this case, however, the document DataCore (written by Ziya Aral & Jonathan Ely) is very, very good. Authors talk of VDI storage without bias and validate their architecture using standard tools such as the benchmark of VSI Login Consultants.

from the document: Previous publications had shown configurations, for reducing the cost of these controllers, using thousands of VD. Reading between the lines, it is clear that the hardware costs for each VD grow rapidly As the configuration is scaled down. And it is obvious that the smaller VDI configurations are also the most important by most of the practical points of view. On the other hand, the configuration is referred to in this report can still be scaled upwards in a linear fashion, up to thousands of Virtual Desktop, eliminating bulky configurations created to try those "perfect point" artificial where costs are optimized.

Then continue:

The problem in the adoption of Virtual Desktop is that SANs are often implemented using large and expensive storage controllers combined with complex external networks for storage. While these tools offer the advantage of a reasonable scale, but introduces a significant barrier to adoption of the cost of VDI. To overcome this obstacle, the hardware vendors publish data referring to implementations that include a thousand to several thousand VD.

Many companies, realizing the benefits potential of this technology are now implementing pilot programs or are trying to integrate an initial adoption of Virtual Desktop into existing structures. If the granularity of these implementations were in the thousands, users would be forced to consume not only a "loaf", but an entire bakery in a single meal ... without even having time to see if the bread is good. The alternative is equally unappealing. The user "grits his teeth and is willing to confront the barrier formed by the cost and complexity of an entire SAN while using a dedicated non-optimal number of Virtual Desktop. In this way, the cost of desktop deployment becomes much higher than that there had been maintaining that the "old pattern" of the individual desktop. And in practice you get the opposite effect than that sought by the introduction of a new technology for the 'cost savings', as he has noticed a growing number of bloggers from the spirit practice.


The authors provide an overview of their test scenario, which started with two identical servers each configured to accommodate 110 virtual desktops. Each server had five drives: one start and two pairs of two pool records controlled by DataCore, one of which is used as a storage server and the second used as a mirror backup of the other pool. To have a base as simple as possible were used for the common drive SATA:

We are reluctant to use components that are not easily configurable standard SATA products. SAS, Fibre Channel, high-speed peripherals, SSD and hybrid drives have their own specific role and are ideal for performance superior to that of SATA. The inclusion of these devices in a reference configuration, however, would achieve the same effect as VDI architecture around a specific type of battery storage. An optimization of this kind would only create confusion, not only would make it difficult to compare, but this would lead to a kind of intrusion by a specific hardware architecture in areas where the stars should be the characteristics of VDI.

The hardware we have used it cost $ 3,848 per server (about $ 35 per user), with a total price comes to about $ 67 per user, taking into account the DataCore software. The authors explained:

The main innovation in the results of this benchmark is to use the system for storage virtualization DataCore on the same hardware platforms used to host the Virtual Desktop. While common sense suggests that the DataCore software could become a competitor to hunt the same scarce resources of the hardware platform, such as memory and CPU, numerous experiments have shown the opposite. In any case, configurations as easily have used that offered higher performance than companies with external storage.

retrospect, the reasons are easy to find. The Virtual Desktop application is not particularly heavy usage of I / O and proves to be easily supported by DataCore at the cost of a very limited number of CPU cycles. The elimination of most of the traffic channel outside help to further reduce this activity. In return, the latency of the cache block virtually disappear, as well as overloading the channel and latency I / O. SANmelody resident "pays off" by itself, without losing anything on the front of the functionality or portability.


Even if your test was limited to two servers, you can scale up by adopting a topology "star." The idea is that you continue to use local disks for VDI hosts (each with DataCore Software) as primary storage, and that the central hub in star topology contains a replica of each host VDI continuously synchronized, ensuring that no data is lost if a problem occurs in one of the hosts (and if you need high availability extreme enough to keep on hand a host of VDI reserve that can temporarily start the VM from the pool of storage ' central hub.) The authors also pointed out another advantage of local storage, saying that "... these types of configuration are inherently self-regulating." Every time you add a group of Virtual Desktop, the same happens with the storage infrastructure needed to support them. " DataCore

The authors are aware of how this approach is different than the typical architecture of storage for VDI:

Overall, the architecture under test is different than most of those used for this purpose from other providers storage. Rather than create a configuration around a VDI storage controller or pre-existing SAN architecture requires that the SAN is built around VDI to the same server.

Finally, they made fun of the gang freak (for another in the group), arguing that there is no "trick" that is able to improve the price / performance ratio of a typical VDI as their solution storage virtual

Better use of memory, more advanced hardware and various "tricks" to optimize provide the desired results. The increase in cost and complexity, however, leads to a system that is only slightly better in terms of price / performance ratio compared to that tested here. We "old freak I am sad to note that the configuration tested is sufficiently optimized to overcome the barrier that leads to a VDI practice to manage small and low-level optimizations can lead to only small improvements.

I must admit that this is one of the best business records that I have read for a long time now. What do you think? We missing something?

0 comments:

Post a Comment