Virtualization Technology News and Information
Article
RSS
VMblog Expert Interview: Gary Quan of Condusiv Technologies Talks I/O Bottlenecks, Performance Impacts and How to Accelerate

interview condusiv quan 

VMblog recently had the chance to catch up with Gary Quan, SVP, Product Strategy, at Condusiv Technologies to talk about the impact of I/O bottlenecks, how they impact performance and how to accelerate throughput by 30-40%.

VMblog:  Tell us what you mean when you say organizations should solve I/O inefficiencies at the source to increase throughput, reduce latency and increase data transfer rates.

Gary Quan:  I would be glad too. First of all, when indicating ‘solving at the source', I mean where the I/Os are being created from. In this case, the Windows operating system where the system and other applications are generating the Write I/Os. 

In an ideal environment, Write I/Os would be created in efficient, sequential I/Os rather than fractured, random I/Os. For example, when writing out a larger piece of data, it is more efficient and faster to do it in a single sequential I/O rather than 20 or more smaller, fractured, random I/Os. Not only is it more efficient on the OS side to process a single I/O rather than 20, you are also gaining the optimal performance from your storage. If you ever look at benchmarks on storage, both HDDs and SSDs, sequential I/Os always outperform random I/Os.

Unfortunately, the Windows file system tends to break up Write I/Os into smaller, fractured, random I/Os. The reason for this is that when a file is created or extended, the file system does not know how large that creation or extension is going to be, so it basically looks for the next logical allocation to write the data. If that allocation is not big enough, then it has to look for another allocation, and so on and so on. This results in many smaller fractured write I/Os being written and, of course, being read back in the same inefficient manner.

VMblog:  What is the "I/O Blender Effect" and how does it penalize application performance and VM density?

Quan:  The "I/O Blender effect" is when you have multiple virtual machines from the same hypervisor host, all sending their I/Os through that same host at the same time.  The hypervisor acts like a blender. It mixes these all these I/O streams, which causes a randomization of the storage I/O packets, before sending out what is now a chaotic mess of small, fractured and now very random I/O streams out to the storage controller.

The storage is now only receiving data in small chunks at a time and won't understand the relationship between the packets, so it now only has the opportunity to create very small stripes across its media, and that, unfortunately, can mean more storage operations are required before it can send an acknowledgement of the data transfer back up to the Windows operating system that originated it.

Then, if the VMs are sending small fractured I/Os to begin with, it can make the problem and performance degradation even worse.

VMblog:  What types of applications are hit hardest by these I/O inefficiencies?

Quan:  Although we like to point to MS-SQL workloads, it is really any systems that are data throughput intensive. This includes Oracle on Windows, ERP, VDI, EHR, Business Intelligence (BI) apps, CRM, Exchange, SharePoint, file servers, backup, and more.

VMblog:  Does buying newer, faster storage such as SSDs, all-flash, hyper-converged, or moving to the cloud solve these issues you have been talking about?

Quan:  This can help, but essentially "no", hahaha it does not solve the root issue. Windows is still Windows... No matter where you run Windows and no matter what kind of storage environment you run Windows on, Windows still penalizes optimal performance due to severe write inefficiencies in the hand-off of data to storage.

VMblog:  What benefits can be achieved by transforming I/O into contiguous, large, sequential I/O and eliminating the I/O blender effect?

Quan:  There are a multitude of benefits, but basically improved performance for the users. This includes, but not limited to:

  • By improving latency & throughput by 40%+, applications perform much better
  • Better overall system performance due to tuning I/O efficiency
  • Extended hardware lifecycles, allowing you to sweat your assets longer and save overspending on expensive new hardware
  • Optimal performance from your storage infrastructure by delivering and requesting more efficient, sequential I/Os
  • Increased network bandwidth by decreasing the I/O traffic on the network
  • By increasing the I/O capacity, you can increase the VDI density (# of VDIs per host)
  • Reduced cloud-compute costs due to decreased I/O traffic

VMblog:  What technologies are recommended to accelerate throughput by 30-40%?

Quan:  On the first question, I discussed the inefficiencies of Write I/Os on the Windows file system. DymaxIOTM solves this issue with a technology called IntelliWrite®. This technology gathers intelligence in the background on the file creation and extension sizes of files, file types, and applications. It then provides this intelligence back to the file system to help find the best size allocation which will enforce the more efficient, sequential I/Os to occur rather than smaller, random I/Os.

Then for Read I/Os, DymaxIO contains another technology called IntelliMemory®, which is a very intelligent DRAM caching technology. It is only using available DRAM that is left unused on your system and will dynamically give it back if any user or system process needs it. By satisfying Read I/O requests through DRAM caching, it eliminates the I/O having to travel to the storage and back to get satisfied. With DRAM 10-15x faster than even SSDs, you can see the huge advantage...actually providing performance to users that is better than when new.

Combined, both of these technologies have shown to increase throughput by 30-40% and more.

VMblog:  What solutions does Condusiv have to dynamically accelerate data for maximum I/O performance?

Quan:  Condusiv recently released DymaxIO. DymaxIO is fast data software that returns 40% or more of your throughput that is being robbed due to the I/O inefficiencies that I described. It's fast, easy, fully transparent, intelligent, and automatic. It can fix application slows, freezes, timeouts, slow SQL queries, and more, at the source - right where the I/Os are being generated from, the Windows Operating system.  Hardware is a big investment, now with DymaxIO you can get your speed back.

By installing DymaxIO, organizations can expect an immediate performance boost in virtualized environments, MS-SQL workloads, Oracle on Windows, ERP, VDI, EHR (MEDITECH), Business Intelligence (BI) apps, CRM, Exchange, SharePoint, file servers, backup, and more. You don't have to take my word for it, I recommend that you install the software and watch your performance problems disappear. You can get started with a free 30-day trial of new DymaxIO.

##

Published Thursday, July 30, 2020 7:34 AM by David Marshall
Filed under: ,
Comments
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
top25
Calendar
<July 2020>
SuMoTuWeThFrSa
2829301234
567891011
12131415161718
19202122232425
2627282930311
2345678