Condusiv Technologies Blog

Condusiv Technologies Blog

Blogging @Condusiv

The Condusiv blog shares insight into the issues surrounding system and application performance—and how I/O optimization software is breaking new ground in solving those issues.

A Deep Dive Into The I/O Performance Dashboard

by Howard Butler 2. August 2018 08:36

While most users are familiar with the main Diskeeper®/V-locity®/SSDkeeper™ Dashboard view which focuses on the number of I/Os eliminated and Storage I/O Time Saved, the I/O Performance Dashboard tab takes a deeper look into the performance characteristics of I/O activity.  The data shown here is similar in nature to other Windows performance monitoring utilities and provides a wealth of data on I/O traffic streams. 

By default, the information displayed is from the time the product was installed. You can easily filter this down to a different time frame by clicking on the “Since Installation” picklist and choosing a different time frame such as Last 24 Hours, Last 7 Days, Last 30 Days, Last 60 Days, Last 90 Days, or Last 180 Days.  The data displayed will automatically be updated to reflect the time frame selected.

 

The first section of the display above is labeled as “I/O Performance Metrics” and you will see values that represent Average, Minimum, and Maximum values for I/Os Per Second (IOPS), throughput measured in Megabytes per Second (MB/Sec) and application I/O Latency measured in milliseconds (msecs). Diskeeper, V-locity and SSDkeeper use the Windows high performance system counters to gather this data and it is measured down to the microsecond (1/1,000,000 second).

While most people are familiar with IOPS and throughput expressed in MB/Sec, I will give a short description just to make sure. 

IOPS is the number of I/Os completed in 1 second of time.  This is a measurement of both read and write I/O operations.  MB/Sec is a measurement that reflects the amount of data being worked on and passed through the system.  Taken together they represent speed and throughput efficiency.  One thing I want to point out is that the Latency value shown in the above report is not measured at the storage device, but instead is a much more accurate reflection of I/O response time at an application level.  This is where the rubber meets the road.  Each I/O that passes through the Windows storage driver has a start and completion time stamp.  The difference between these two values measures the real-world elapsed time for how long it takes an I/O to complete and be handed back to the application for further processing.  Measurements at the storage device do not account for network, host, and hypervisor congestion.  Therefore, our Latency value is a much more meaningful value than typical hardware counters for I/O response time or latency.  In this display, we also provide meaningful data on the percentage of I/O traffic- which are reads and which are writes.  This helps to better gauge which of our technologies (IntelliMemory® or IntelliWrite®) is likely to provide the greatest benefit.

The next section of the display measures the “Total Workload” in terms of the amount of data accessed for both reads and writes as well as any data satisfied from cache. 

 

A system which has higher workloads as compared to other systems in your environment are the ones that likely have higher I/O traffic and tend to cause more of the I/O blender effect when connected to a shared SAN storage or virtualized environment and are prime candidates for the extra I/O capacity relief that Diskeeper, V-locity and SSDkeeper provide.

Now moving into the third section of the display labeled as “Memory Usage” we see some measurements that represent the Total Memory in the system and the total amount of I/O data that has been satisfied from the IntelliMemory cache.  The purpose of our patented read caching technology is twofold.  Satisfy from cache the frequently repetitive read data requests and be aware of the small read operations that tend to cause excessive “noise” in the I/O stream to storage and satisfy them from the cache.  So, it’s not uncommon to see the “Data Satisfied from Cache” compared to the “Total Workload” to be a bit lower than other types of caching algorithms.  Storage arrays tend to do quite well when handed large sequential I/O traffic but choke when small random reads and writes are part of the mix.  Eliminating I/O traffic from going to storage is what it’s all about.  The fewer I/Os to storage, the faster and more data your applications will be able to access.

In addition, we show the average, minimum, and maximum values for free memory used by the cache.  For each of these values, the corresponding Total Free Memory in Cache for the system is shown (Total Free Memory is memory used by the cache plus memory reported by the system as free).  The memory values will be displayed in a yellow color font if the size of the cache is being severely restricted due to the current memory demands of other applications and preventing our product from providing maximum I/O benefit.  The memory values will be displayed in red if the Total Memory is less than 3GB.

Read I/O traffic, which is potentially cacheable, can receive an additional benefit by adding more DRAM for the cache and allowing the IntelliMemory caching technology to satisfy a greater amount of that read I/O traffic at the speed of DRAM (10-15 times faster than SSD), offloading it away from the slower back-end storage. This would have the effect of further reducing average storage I/O latency and saving even more storage I/O time.

Additional Note: For machines running SQL Server or Microsoft Exchange, you will likely need to cap the amount of memory that those applications can use (if you haven’t done so already), to prevent them from ‘stealing’ any additional memory that you add to those machines.

It should be noted the IntelliMemory read cache is dynamic and self-learning.  This means you do not need to pre-allocate a fixed amount of memory to the cache or run some pre-assessment tool or discovery utility to determine what should be loaded into cache.  IntelliMemory will only use memory that is otherwise, free, available, or unused memory for its cache and will always leave plenty of memory untouched (1.5GB – 4GB depending on the total system memory) and available for Windows and other applications to use.  As there is a demand for memory, IntelliMemory will release memory from it’s cache and give this memory back to Windows so there will not be a memory shortage.  There is further intelligence with the IntelliMemory caching technology to know in real time precisely what data should be in cache at any moment in time and the relative importance of the entries already in the cache.  The goal is to ensure that the data maintained in the cache results in the maximum benefit possible to reduce Read I/O traffic. 

So, there you have it.  I hope this deeper dive explanation provides better clarity to the benefit and internal workings of Diskeeper, V-locity and SSDkeeper as it relates to I/O performance and memory management.

You can download a free 30-day, fully functioning trial of our software and see the new dashboard here: www.condusiv.com/try

Which Processes are Using All of My System Resources?

by Gary Quan 17. July 2018 05:50

Over time as more files and applications are added to your system, you notice that performance has degraded, and you want to find out what is causing it. A good starting point is to see how the system resources are being used and which processes and/or files are using them.

Both Diskeeper® and SSDkeeper® contain a lesser known feature to assist you on this. It is called the System Monitoring Report which can show you how the CPU and I/O resources are being utilized, then digging down a bit deeper, which processes or files are using them.

Under Reports on the Main Menu, the System Monitoring Report provides you with data on the system’s CPU usage and I/O Activity.

 

The CPU Usage report takes the average CPU usage from the past 7 days, then provides a graph of the hourly usage on an average day. You can then see at which times the CPU resources are being hit the most and by how much.

Digging down some more, you can then see which processes utilized the most CPU resources.

 

The Disk I/O Activity report takes the average disk I/O activity from the past 7 days, then provides a graph of the hourly activity on an average day. You can then determine at which times the I/O activity is the highest.

Digging down some more, you can then see which processes utilized the I/O resources the most, plus what processes are causing the most split (extra) I/Os.

 

You can also see which file types have the highest I/O utilization as well as those causing the most split (extra) I/Os.  This can help indicate what files and related processes are causing this type of extra I/O activity.

 

So, if you are trying to see how your system is being used, maybe for performance issues, this report gives you a quick and easy look on how the CPU and Disk I/O resources are being used on your system and what processes and file types are using them. This along with some other Microsoft Utilities, like Task Manager and Performance Monitor can help you tune your system for optimum performance.

Solving the IO Blender Effect with Software-Based Caching

by Spencer Allingham 5. July 2018 07:30

First, let me explain exactly what the IO Blender Effect is, and why it causes a problem in virtualized environments such as those from VMware or Microsoft’s Hyper-V.



This is typically what storage IO traffic would look like when everything is working well. You have the least number of storage IO packets, each carrying a large payload of data down to the storage. Because the data is arriving in large chunks at a time, the storage controller has the opportunity to create large stripes across its media, using the least number of storage-level operations before being able to acknowledge that the write has been successful.



Unfortunately, all too often the Windows Write Driver is forced to split data that it’s writing into many more, much smaller IO packets. These split IO situations cause data to be transferred far less efficiently, and this adds overhead to each write and subsequent read. Now that the storage controller is only receiving data in much smaller chunks at a time, it can only create much smaller stripes across its media, meaning many more storage operations are required to process each gigabyte of storage IO traffic.


This is not only true when writing data, but also if you need to read that data back at some later time.

But what does this really mean in real-world terms?

It means that an average gigabyte of storage IO traffic that should take perhaps 2,000 or 3,000 storage IO packets to complete, is now taking 30,000, or 40,000 storage IO packets instead. The data transfer has been split into many more, much smaller, fractured IO packets. Each storage IO operation that has to be generated takes a measurable amount of time and system resource to process, and so this is bad for performance! It will cause your workloads to run slower than they should, and this will worsen over time unless you perform some time and resource-costly maintenance.

So, what about the IO Blender Effect?

Well, the IO Blender Effect can amplify the performance penalty (or Windows IO Performance Tax) in a virtualized environment. Here’s how it works…

 

As the small, fractured IO traffic from several virtual machines passes through the physical host hypervisor (Hyper-V server or VMware ESX server), the hypervisor acts like a blender. It mixes these IO streams, which causes a randomization of the storage IO packets, before sending out what is now a chaotic mess of small, fractured and now very random IO streams out to the storage controller.

It doesn’t matter what type of storage you have on the back-end. It could be direct attached disks in the physical host machine, or a Storage Area Network (SAN), this type of storage IO profile couldn’t be less storage-friendly.

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 means many 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.

How can RAM caching alleviate the problem?

 

Firstly, to be truly effective the RAM caching needs to be done at the Windows operating system layer. This provides the shortest IO path for read IO requests that can be satisfied from server-side RAM, provisioned to each virtual machine. By satisfying as many “Hot Reads” from RAM as possible, you now have a situation where not only are those read requests being satisfied faster, but those requests are now not having to go out to storage. That means less storage IO packets for the hypervisor to blend.

Furthermore, the V-locity® caching software from Condusiv Technologies also employs a patented technology called IntelliWrite®. This intelligently helps the Windows Write Driver make better choices when writing data out to disk, which avoids many of the split IO situations that would then be made worse by the IO Blender Effect. You now get back to that ideal situation of healthy IO; large, sequential writes and reads.

Is RAM caching a disruptive solution?

 

No! Not at all, if done properly.

Condusiv’s V-locity software for virtualised environments is completely non-disruptive to live, running workloads such as SQL Servers, Microsoft Dynamics, Business Information (BI) solutions such as IBM Cognos, or other important workloads such as SAP, Oracle and the such.

In fact, all you need to do to test this for yourself is download a free trialware copy from:

www.condusiv.com/try

Just install it! There are no reboots required, and it will start working in just a couple of minutes. If you decide that it isn’t for you, then uninstall it just as easily. No reboots, no disruption!


How to Improve Application Performance by Decreasing Disk Latency like an IT Engineer

by Spencer Allingham 13. June 2018 06:49

You might be responsible for a busy SQL server, for example, or a Web Server; perhaps a busy file and print server, the Finance Department's systems, documentation management, CRM, BI, or something else entirely.

Now, think about WHY these are the workloads that you care about the most?

 

Were YOU responsible for installing the application running the workload for your company? Is the workload being run business critical, or considered TOO BIG TO FAIL?

Or is it simply because users, or even worse, customers, complain about performance?

 

If the last question made you wince, because you know that YOU are responsible for some of the workloads running in your organisation that would benefit from additional performance, please read on. This article is just for you, even if you don't consider yourself a "Techie".

Before we get started, you should know that there are many variables that can affect the performance of the applications that you care about the most. The slowest, most restrictive of these is referred to as the "Bottleneck". Think of water being poured from a bottle. The water can only flow as fast as the neck of the bottle, the 'slowest' part of the bottle.

Don't worry though, in a computer the bottleneck will pretty much always fit into one of the following categories:

•           CPU

•           DISK

•           MEMORY

•           NETWORK

The good news is that if you're running Windows, it is usually very easy to find out which one the bottleneck is in, and here is how to do it (like an IT Engineer):

 •          Open Resource Monitor by clicking the Start menu, typing "resource monitor", and pressing Enter. Microsoft includes this as part of the Windows operating system and it is already installed.

 •          Do you see the graphs in the right-hand pane? When your computer is running at peak load, or users are complaining about performance, which of the graphs are 'maxing out'?

This is a great indicator of where your workload's bottleneck is to be found.         

 

SO, now you have identified the slowest part of your 'compute environment' (continue reading for more details), what can you do to improve it?

The traditional approach to solving computer performance issues has been to throw hardware at the solution. This could be treating yourself to a new laptop, or putting more RAM into your workstation, or on the more extreme end, buying new servers or expensive storage solutions.

BUT, how do you know when it is appropriate to spend money on new or additional hardware, and when it isn't. Well the answer is; 'when you can get the performance that you need', with the existing hardware infrastructure that you have already bought and paid for. You wouldn't replace your car, just because it needed a service, would you?

Let's take disk speed as an example.  Let’s take a look at the response time column in Resource Monitor. Make sure you open the monitor to full screen or large enough to see the data.  Then open the Disk Activity section so you can see the Response Time column.  Do it now on the computer you're using to read this. (You didn't close Resource Monitor yet, did you?) This is showing the Disk Response Time, or put another way, how long is the storage taking to read and write data? Of course, slower disk speed = slower performance, but what is considered good disk speed and bad?

To answer that question, I will refer to a great blog post by Scott Lowe, that you can read here...

https://www.techrepublic.com/blog/the-enterprise-cloud/use-resource-monitor-to-monitor-storage-performance/

In it, the author perfectly describes what to expect from faster and slower Disk Response Times:

"Response Time (ms). Disk response time in milliseconds. For this metric, a lower number is definitely better; in general, anything less than 10 ms is considered good performance. If you occasionally go beyond 10 ms, you should be okay, but if the system is consistently waiting more than 20 ms for response from the storage, then you may have a problem that needs attention, and it's likely that users will notice performance degradation. At 50 ms and greater, the problem is serious."

Hopefully when you checked on your computer, the Disk Response Time is below 20 milliseconds. BUT, what about those other workloads that you were thinking about earlier. What's the Disk Response Times on that busy SQL server, the CRM or BI platform, or those Windows servers that the users complain about?

If the Disk Response Times are often higher than 20 milliseconds, and you need to improve the performance, then it's choice time and there are basically two options:

           In my opinion as an IT Engineer, the most sensible option is to use storage workload reduction software like Diskeeper for physical Windows computers, or V-locity for virtualised Windows computers. These will reduce Disk Storage Times by allowing a good percentage of the data that your applications need to read, to come from a RAM cache, rather than slower disk storage. This works because RAM is much faster than the media in your disk storage. Best of all, the only thing you need to do to try it, is download a free copy of the 30 day trial. You don't even have to reboot the computer; just check and see if it is able to bring the Disk Response Times down for the workloads that you care about the most.

           If you have tried the Diskeeper or V-locity software, and you STILL need faster disk access, then, I'm afraid, it's time to start getting quotations for new hardware. It does make sense though, to take a couple of minutes to install Diskeeper or V-locity first, to see if this step can be avoided. The software solution to remove storage inefficiencies is typically a much more cost-effective solution than having to buy hardware!

Visit www.condusiv.com/try to download Diskeeper and V-locity now, for your free trial.

 

Windows is still Windows Whether in the Cloud, on Hyperconverged or All-flash

by Brian Morin 5. June 2018 04:43

Let me start by stating two facts – facts that I will substantiate if you continue to the end.

Fact #1 - Windows suffers from severe write inefficiencies that dampen overall performance. The holy grail question as to how severe is answered below.

Fact #2, Windows is still Windows whether running in the cloud, on hyperconverged systems, all-flash storage, or all three. Before you jump to the real-world examples below, let me first explain why.

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. Files are always broken down to be excessively smaller than they need to be. Since each piece means a dedicated I/O operation to process as a write or read, this means an enormous amount of noisy, unnecessary I/O traffic is chewing up precious IOPS, eroding throughput, and causing everything to run slower despite how many IOPS are at your disposal.

How much slower?

Now that the latest version of our I/O reduction software is being run across tens of thousands of servers and hundreds of thousands of PCs, we can empirically point out that no matter what kind of environment Windows is running on, there is always 30-40% of I/O traffic that is nothing but mere noise stealing resources and robbing optimal performance.

Yes, there are edge cases in which the inefficiency is as little as 10% but also other edge cases where the inefficiency is upwards of 70%. That being said, the median range is solidly in the 30-40% range and it has absolutely nothing to do with the backend media whether spindle, flash, hybrid, hyperconverged, cloud, or local storage.

Even if running Windows on an all-flash hyperconverged system, SAN or cloud environment with low latency and high IOPS, if the I/O profile isn’t addressed by our I/O reduction software to ensure large, clean, contiguous writes and reads, then 30-40% more IOPS will always be required for any given workload, which adds up to unnecessarily giving away 30-40% of the IOPS you paid for while slowing the completion of every job and query by the same amount.

So what’s going on here? Why is this happening and how?

First of all, the behavior of Windows when it comes to processing write and read input/output (I/O) operations is identical despite the storage backend whether local or network or media despite spindles or flash. This is because Windows only ever sees a virtual disk - the logical disk within the file system itself. The OS is abstracted from the physical layer entirely. Windows doesn’t know and doesn’t care if the underlying storage is a local disk or SSD, an array full of SSDs, hyperconverged, or cloud. In the mind of the OS, the logical disk IS the physical disk when, in fact, it’s just a reference architecture. In the case of enterprise storage, the underlying storage controllers manage where the data physically lives. However, no storage device can dictate to Windows how to write (and subsequently read) in the most efficient manner possible.

This is why many enterprise storage controllers have their own proprietary algorithms to “clean up” the mess Windows gives it by either buffering or coalescing files on a dedicated SSD or NVRAM tier or physically move pieces of the same file to line up sequentially, which does nothing for the first penalized write nor several penalized reads after as the algorithm first needs to identify a continued pattern before moving blocks. As much as storage controller optimization helps, it’s a far cry from an actual solution because it doesn’t solve the source of the larger root cause problem - even with backend storage controller optimizations, Windows will still make the underlying server to storage architecture execute many more I/O operations than are required to write and subsequently read a file, and every extra I/O required takes a measure of time in the same way that four partially loaded dump trucks will take longer to deliver the full load versus one fully loaded dump truck. It bears repeating - no storage device can dictate to Windows how to best write and read files for the healthiest I/O profile that delivers optimum performance because only Windows controls how files are written to the logical disk. And that singular action is what determines the I/O density (or lack of) from server to storage.

The reason this is occurring is because there are no APIs that exist between the Windows OS and underlying storage system whereby free space at the logical layer can be intelligently synced and consolidated with the physical layer without change block movement that would otherwise wear out SSDs and trigger copy-on-write activity that would blow up storage services like replication, thin provisioning, and more.

This means Windows has no choice but to choose the next available allocation at the logical disk layer within the file systems itself instead of choosing the BEST allocation to write and subsequently read a file.

The problem is that the next available allocation is only ever the right size on day 1 on a freshly formatted NTFS volume. But as time goes on and files are written and erased and re-written and extended and many temporary files are quickly created and erased, that means the next available space is never the right size. So, when Windows is trying to write a 1MB file but the next available allocation at the logical disk layer is 4K, it will fill that 4K, split the file, generate another I/O operation, look for the next available allocation, fill, split, and rinse and repeat until the file is fully written, and your I/O profile is cluttered with split I/Os. The result is an I/O degradation of excessively small writes and reads that penalizes performance with a “death by a thousand cuts” scenario.

It’s for this reason, over 2,500 small, midsized, and large enterprises have deployed our I/O reduction software to eliminate all that noisy I/O robbing performance by addressing the root cause problem. Since Condusiv software sits at the storage driver level, our purview is able to supply patented intelligence to the Windows OS, enabling it to choose the BEST allocation for any file instead of the next available, which is never the right size. This ensures the healthiest I/O profile possible for maximum storage performance on every write and read. Above and beyond that benefit, our DRAM read caching engine (the same engine OEM’d by 9 of the top 10 PC manufacturers), eliminates hot reads from traversing the full stack from storage by serving it straight from idle, available DRAM. Customers who add anywhere to 4GB-16GB of memory to key systems with a read bias to get more from that engine, will offload 50-80% of all reads from storage, saving even more precious storage IOPS while serving from DRAM which is 15X faster than SSD. Those who need the most performance possible or simply need to free up more storage IOPS will max our 128GB threshold and offload 90-99% of reads from storage.

Let’s look at some real-world examples from customers.

Here is VDI in AWS shared by Curt Hapner (CIO, Altenloh Brinck & Co.). 63% of read traffic is being offloaded from underlying storage and 33% of write I/O operations. He was getting sluggish VDI performance, so he bumped up memory slightly on all instances to get more power from our software and the sluggishness disappeared.

Here is an Epicor ERP with SQL backend in AWS from Altenloh Brinck & Co. 39% of reads are being eliminated along with 44% of writes to boost the performance and efficiency of their most mission critical system.

 

Here’s from one of the largest federal branches in Washington running Windows servers on an all-flash Nutanix. 45% of reads are being offloaded and 38% of write traffic.

 

Here is a spreadsheet compilation of different systems from one of the largest hospitality and event companies in Europe who run their workloads in Azure. The extraction of the dashboard data into the CSV shows not just the percentage of read and write traffic offloaded from storage but how much I/O capacity our software is handing back to their Azure instances.

 

To illustrate we use the software here at Condusiv on our own systems, this dashboard screenshot is from our own Chief Architect (Rick Cadruvi), who uses Diskeeper on his SSD-powered PC. You can see him share his own production data in the recent “live demo” webinar on V-locity 7.0 - https://youtu.be/Zn2QGxBHUzs

As you can see, 50% of reads are offloaded from his local SSD while 42% of writes operations have been saved by displacing small, fractured files with large, clean contiguous files. Not only is that extending the life of his SSD by reducing write amplification, but he has saved over 6 days of I/O time in the last month.

 

Finally, regarding all-flash SAN storage systems, the full data is in this case study with the University of Illinois who used Condusiv I/O reduction software to more than double the performance of SQL and Oracle sitting on their all-flash arrays: http://learn.condusiv.com/rs/246-QKS-770/images/CS_University-Illinois.pdf?utm_campaign=CS_UnivIll_Case_Study

For a free trial, visit http://learn.condusiv.com/Try-V-locity.html. For best results, bump up memory on key systems if you can and make sure to install the software on all the VMs on the same host. If you have more than 10 VMs, you may want to Contact Us for SE assistance in spinning up our centralized management console to push everything at once – a 20-min exercise and no reboot required.

Please visit www.condusiv.com/v-locity for more than 20 case studies on how our I/O reduction software doubled the performance of mission critical applications like MS-SQL for customers of various environments.

RecentComments

Comment RSS

Month List

Calendar

<<  October 2018  >>
MoTuWeThFrSaSu
24252627282930
1234567
891011121314
15161718192021
22232425262728
2930311234

View posts in large calendar