Category Archives: Performance

The truth about storage benchmarking

Posted on by
Reply

Recently I was asked to review some performance testing done by an external party and my initial impression was the performance was well below what I expected.

So over the weekend I setup a block in my lab to reproduce the tests to see if the results were firstly repeatable, and if so, what performance would I get with and without tuning.

The only significant difference between my hardware and the hardware used by the 3rd party was that I used old dual socket Ivy Bridge E5-2670 2.6Ghz 8c processors and the 3rd party had a much newer dual Broadwell E5-2640 v4 2.4Ghz processors.

If we compare the two processors using CPUBoss.com we see the following:

CPUboss1

Not surprisingly the Broadwell E5-2640 v4 processor is faster, but possibly less than you would expect with a 16.28% better PassMark per core, and in my opinion, the per core value quite importaint especially when considering business critical applications.

None the less, a 16.28% performance improvement per core will be a significant factor for a benchmark with Nutanix as the Controller VM (CVM) is powered by the CPU of the host.

I thought I would whip up a quick post about performance benchmarking to show how different performance results can be on the same hardware depending on just a few factors and why storage benchmarking, especially competitive benchmarking, cannot and should not be trusted when making purchasing decisions.

This test was for a 10k user MS Exchange deployment and the hardware used for testing was performed on in both cases was 1 x 1.92TB SSD and 3 x 4TB SATA drives and they were both tested on the same GA Nutanix AOS build.

The required (or Target) IOPS was just 216 per MS Exchange instance (VM) as shown below by the Jetstress report.

TargetIO

This target is calculated by Jetstress when using the “Exchange Mailbox Profile” test scenario with the following configuration:

MailboxProfile2

The resulting SSD vs SATA ratio makes this test largely about the limitations of SATA performance as >87% of data is being read from the SATA tier.

TierUsage

Test 1: The Jetstress dataset was created and then the performance test was immediately ran for 2hrs with no pre-warming of the metadata or read cache.

Achieved Transactional I/O: 200.663
Avg Log Write Latency: 1.06ms
Avg DB Write Latency: 1.4ms
Avg DB Read Latency: 14ms

This result was 15.61% lower than the 3rd parties result and interestingly if we correct for CPU core performance, it’s less than 1% difference. As this was in line with my expectation knowing the importance of CPU clock speed, I would say for this testing that the baseline results were comparable.

Test 2: The Nutanix tiering was tuned to suit large working sets (which vastly exceed the SSD tier) and the Jetstress dataset was created and the performance test was immediately ran for 2hrs again with no pre-warming of the metadata or read cache.

Before we get to the results, I want to point out that Jetstress is in some ways is very good but in other ways a very unrealistic benchmarking tool as the entire dataset is “active” which is not the case in the real world. However, in one way this is a good thing because a passing Jetstress result in my experience means the production deployment performs very well from a storage perspective especially when using tiered storage which is built around the assumption not all data is active. As a result, a Jetstress test could be considered a “worse case scenario” style test for intelligent tiered storage.

Achieved Transactional I/O: 249.623
Avg Log Write Latency: 0.99ms
Avg DB Write Latency: 1.5ms
Avg DB Read Latency: 12ms

Test 3: I then setup Jetstress as per Nutanix MS Exchange best practices and ran the test again with no pre-warming of the metadata or read cache.

Achieved Transactional I/O: 389.753
Avg Log Write Latency: 0.95ms
Avg DB Write Latency: 2.0ms
Avg DB Read Latency: 17ms

Test 4: I then lowered the Jetstress thread count to the lowest value (roughly 33% lower) which I estimated would achieve the target IOPS (this is to simulate real world requirements) and ran the test again with no pre-warming of the metadata or read cache.

Achieved Transactional I/O: 300.254
Avg Log Write Latency: 0.94ms
Avg DB Write Latency: 1.5ms
Avg DB Read Latency: 12ms

Note: Test 4 achieved the highest I/O per thread.

Test 5: The same configuration as Test 4 but with pre-warming of the metadata cache.

Achieved Transactional I/O: 334
Avg Log Write Latency: 0.98ms
Avg DB Write Latency: 1.9ms
Avg DB Read Latency: 12.4ms

Some of you might be asking, how did test 4 achieve higher transactional I/O and with lower read and write latency than Test 1 & 2 with less threads. Shouldn’t a higher thread count achieve higher IOPS?

The reasons is because the original thread count was pushing the SATA drives past their capabilities, leading to excessive latency. Lowering the thread count allowed the SATA drives to operate at somewhere around their most efficiency range leading to lower latency.

Test 6: The same configuration as Test 5 but with tuned extent cache (RAM read cache) and 100% medadata cached.

Achieved Transactional I/O: 362.729ms
Avg Log Write Latency: 0.92ms
Avg DB Write Latency: 1.7ms
Avg DB Read Latency: 12ms

As we can see from Test 1 through to Test 6, the performance differs by up to 81% depending on how the platform is configured.

Side note and future looking statement. Many of the optimisations I performed above wont be required for long as many of the areas these optimisations help improve are being addressed in upcoming code. In saying that, for a business critical application like Exchange, I don’t think it’s a problem doing some optimisation as long as 90% of the workloads run well by default and we’re only tuning for the 10% (vBCA) workloads.

But out of interest, what would happen if we enabled data reduction? How much of a performance hit would that take?

Test 7: The same configuration as Test 6 but with In-line compression enabled.

Achieved Transactional I/O: 751.275
Avg Log Write Latency: 0.97ms
Avg DB Write Latency: 3.4ms
Avg DB Read Latency: 5.9ms

That’s a 107.46% increase in transactional IO and with in-line compression! Log write latency remained sub ms and read latency has almost halved.

Note: As Jetstress data is highly compressible, (Nutanix achieves 8:1 or higher with non default settings), I tuned the compression slice size to give a more realistic data reduction ratio. The ratio for this test was 3.99:1 and the ratio of SSD to SATA was almost exactly 50% as shown below.

TierUsageAfterCompression

Why did performance improve so much with In-Line compression? Well there is two main reasons:

  1. More data is being served from the SSD tier as compression allows more effective SSD tier capacity.
  2. Reads from SATA are faster as less physical data needs to be read to service an I/O due to it being compressed. The higher the compression ratio, the more this can improve.

As we can see, the results varied significantly and had I wanted to optimise the test further, I could have achieved even higher performance but there was no need. The requirements for the solution were already achieved and in the case of Test 7, the requirements were exceeded by 247% meaning the solution had heaps of headroom.

Nutanix best practice is to enable In-line compression for MS Exchange and other databases such as Oracle and SQL as per my tweet below.

This testing was performed on Nutanix Acropolis Hypervisor (AHV) but was not using the upcoming Turbo mode, which will further improve performance and lower overheads.

This is a key point many people forget when benchmarking. If we assume the platforms in question are scalable (e.g.: Like Nutanix), it doesn’t matter if one platform does 100k IOPS and another does 200k IOPS if your requirements are 20k IOPS. Both platforms capabilities vastly exceed the requirement (10k IOPS) from a performance perspective, so performance is not longer a significant factor in your purchasing decision.

Question: Are the above performance results genuine?

All of the above results could be argued to be genuine results, at the same time none of the above represent the best performance that could be achieved, yet the results could be used to try and create FUD if they are improperly represented (which is almost always the case with competitive comparisons whether intentional or otherwise).

Let’s say this was your proof of concept, What should be the take away from benchmarking results like this?

Simple: The solution meets/exceeds your performance requirements.

Now for the point of this article: The truth about storage benchmarking is that there are so many variables that can affect the results that unless you’re truely experienced in benchmarking your applications AND an expert in the platforms you’re benchmarking, your results are unlikely to be indicative of the platforms capabilities and therefore of very little value.

If you’re benchmarking Vendor A vs Vendor B, it’s a waste of time doing “Like for Like” benchmarking because the Virtual machine and application settings which are optimal for one vendor, will likely be different for the other vendor. e.g.: SAN vs HCI.

On the other hand, a more valid test would be vendor A’s best practices vs vendor B best practices, but again if one vendor Jetstress achieves 500 and the other achieves 400, that 20% higher performance is all but irrelevant if your requirements are say, 216 like in this case.

A very good example of invalid “like for like” benchmarking would be to size the active working set (i.e.: The capacity of the data you plan to benchmark against) to fit within the cache/SSD tier of one platform, but exceed the cache/SSD capacity of the other platform. The results will be vastly different and will not be indicative of real world performance. This is what vendors do when competitive benchmarking and it’s likely one of the main reasons we see End User License Agreements (EULA) from most if not all storage vendors preventing publishing benchmark results without written agreement.

So the (unpopular) truth about storage benchmarking is it’s not as easy and building a VM and running I/O meter with the same profile on multiple system like some vendors and even 3rd party storage analysts would have you believe. The vast majority of people (customers, analysts and even vendors) doing benchmarking don’t have the skill/experience to produce repeatable or meaningful results, especially on multiple platforms.

In fact it’s unrealistic/unreasonable to expect a person (customer, vendor, consultant) to be an expert in multiple platforms, and very few people are!

Related Articles:

  1. Peak Performance vs Real World Performance
  2. The Key to performance is Consistency

It’s 2017, let’s review Thick vs Thin Provisioning

Posted on by
Reply

For a long time, it has been widely considered that thick provisioning is required to achieve maximum storage performance and for many years this was a good rule of thumb.

Before we get into details, what are Thick and Thin provisioning?

Thick provisioning is where storage allocated to a LUN, NFS mount or Virtual Disk (such as a VMDK in ESXi, VHDX in Hyper-V or vDisk in AHV) is zeroed out and/or fully reserved regardless of how much capacity is actually used.

Thick provisioning avoids a storage subsystem from having to zero out a block before writing new data which is one of the reasons higher performance could be achieved on many storage platforms.

Thin provisioning on the other hand is where storage allocated to a LUN or Virtual Disk is zeroed as data is written and allows physical capacity to be overcommitted.

The advantages of Thick provisioning included easier capacity management, or simply put a “What you see is what you get” as well as maximum performance on most platforms. But by maximum performance, even on older storage platforms the advantage was rarely significant as people would claim.

VMware conducted a Performance Study of VMware vStorage Thin Provisioning back in the ESXi 4.0 days (~2009) which I will briefly summarise.

On page 6 of the performance study the following graph shows the different in performance between Thin and Thick VMDKs during zeroing and post-zeroing.

As you can see the performance is almost identical.

The disadvantages though were and remain significant to this day which include an inability to overcommit storage, meaning physical free space has to be maintained at multiple layers such as RAID group, LUN, Virtual Disk layers, leading to inefficiency.

The advantages of Thin provisioning include the ability to overcommit storage which results in more flexibility when sizing LUNs & Virtual Disks and less wasted space. The only real downsides were potentially increased capacity management complexity and lower performance.

I have previously written two example architectural decisions regarding using “Thin on Thin“, meaning thin provisioned virtual disks on a thin provisioned LUN or NFS mount as well as “Thin on Thick” meaning thin provisioned virtual disks on a thick provisioned LUN or NFS mount. These two examples cover off many of the traditional pros and cons between thick and think, so I won’t repeat myself here.

I never wrote an example design decision for Thick on Thick, but this was common practice when provisioning storage was time consuming, difficult and involved lengthly delays to engage subject matter experts.

In early 2015, I wrote a two part blog series where I explained it’s not as simple as you might think to calculate usable capacity where I compared SAN/NAS verses Nutanix. In part 1, I highlight that the LUN Provisioning Type is one area which can greatly impact the usable capacity of a traditional storage platform.

But fast forward into the era of hyper-converged platforms like Nutanix and some modern storage arrays and the major downsides of thin provisioning, being complexity of capacity management and reduced performance have not only been reduced, but at least in the case of Nutanix, have been eliminated all together.

Let’s address Capacity management w/ Nutanix:

Storage utilisation only needs to be monitored in ONE place, the storage summary which lives on the home screen of the Nutanix HTML 5 UI.

NutanixStorageSummary

No matter how many nodes in your cluster, number of containers (which translate to datastores in a VMware environment), virtual machines & virtual disks or physical servers connecting via ABS, this is the only place you need to monitor capacity.

There are no RAID groups, Disk Groups, Aggregates, LUNs etc where capacity needs to be managed. All nodes in a cluster contributed to the capacity of the cluster and even when one or more virtual machines use more capacity than a the node they run on, Nutanix Acropolis Distributed Storage Fabric (ADSF) takes care of it.

So issue #1, Capacity management, is solved. Now it’s onto the issue of performance.

Thin Provisioning Performance w/ Nutanix:

When running ESXi, Nutanix runs NFS datastores and supports thick provisioning via the VAAI-NAS Space reservation primitive as discussed in this post. This allows the creation of thick provisioned (Eager Zero or Lazy Zero Thick) VMDKs when traditionally NFS datastores did not support it.

However this was only required for Oracle RAC and VMware Fault Tolerance and was not a performance requirement.

However from a performance perspective, Thin provisioning actually outperforms thick on intelligent storage such as Nutanix. In the specific case of Nutanix, random write I/O is serviced by the fastest tier available (e.g.: SSD) and via the operations log (OPLOG) which takes the random writes commits them to persistent media, and then coalesces them into sequential IO to then commit to SSD before tiering it off to lower cost storage in the case of hybrid nodes.

This means the write penalty for overwriting or zeroing blocks before writing new I/O is eliminated.

In fact if you configure thick provisioned virtual disks, as the zeros (or whitespace) is being written by the hypervisor, the Nutanix storage fabric acknowledges every I/O and discards the zeros in favour of storing metadata and simply reserving the capacity. In simple terms, this just means Nutanix has to acknowledge a whole bunch of nothing and the thick provisioning is achieve with a simple reservation as opposed to zeroing out many GBs or TBs of storage.

This means thick provisioning is actually lower performance than thin provisioning on Nutanix.

With modern, intelligent storage, there is limited if any benefits to using thick provisioning, the only example I can think of is to artificially inflate the deduplication ratio as thick provisioned virtual disks tend to have a lot of zeros all of which dedupe. I wrote an article titled: “Deduplication ratios – What should be included in the reported ratio?” which covers off this point in detail but in short, don’t create unnessasary data (in this case, zeros) just to inflate your dedupe ratio, it just wastes storage controller resources and achieves no additional benefits.

The following is a comprehensive list of the real world advantages of using thick provisioning on Nutanix.

This space is intentionally left blank

Summary:

For the best efficiency and performance when deploying virtual machines or storage for physical servers via ABS on Nutanix, use thin provisioning!

SQL & Exchange performance in a Virtual Machine

Posted on by
2

The below is something I see far to often: An SQL or Exchange virtual machine using a single LSI Logic SAS virtual SCSI controller.

LSIlogic

What is even worse is a virtual machine using a single LSI controller and a single virtual disk for one or more databases and logs (as shown above).

Why is this so common?

Probably because the LSI Logic SAS controller is the default for Windows 2008/2012 virtual machines and additional SCSI controllers are not automatically added until you have more than 16 virtual disks for a single VM.

Why is this a problem?

The LSI controller has a queue depth limit of 128, compared to the default limit for PVSCSI which is 256, however it can be tuned to 1024 for higher performance requirements.

As a result, the a configuration with a single LSI controller and/or a limited number of virtual disks can artificially significantly constrain the underlying storage from delivering the performance it is capable of.

Another problem with the LSI controller is the amount of CPU it uses is higher than the PVSCSI controller for the same IO levels. This means you’re wasting virtual machine (and the underlying hosts) CPU resources unnecessarily.

Using more CPU could lead to other problems such as CPU Ready which can also lead to reduced performance.

A colleague and friend of mine, Michael Webster wrote a great post titled: Performance Issues Due To Virtual SCSI Device Queue Depths where he shows the performance difference between SATA, LSI and PVSCSI controllers. I highly recommend having a read of this post.

What is the solution?

Using multiple Paravirtual (PVSCSI) adapters with virtual disks evenly spread over the four controllers for Windows virtual machines is a no brainer!

This results in:

  1. Higher default queue depth
  2. Lower CPU overheads
  3. Higher potential performance

How do I configure this?

It’s fairly straight forward, but don’t just change the LSI Controller too PVSCSI as the Guest OS may not have the driver installed which will result in the VM failing to boot.

Too avoid this, simply edit the virtual machine and add a new Virtual Disk of any size and for the virtual device node, select SCSI (1:0) and follow the prompts.

VirtualDiskSCSI10

Once the new virtual disk is added you should see a new LSI Logic SAS SCSI controller is added as shown below.

NewLSIController

Next highlight the adapter and select “Change Type” in the top right hand corner of the window and select Paravirtual. Once this is complete you should see similar to the below:

AddPVSCSIController

Next hit “Ok” and the new Controller and virtual disk will be added to the VM.

Now we open the console of the VM and open Compute Management and goto Device Manager. Under Storage Controllers you should now see VMware PVSCSI Controller as shown below.

DeviceManagerPVSCSI

Now we are safe to Shutdown the VM.

Once the VM is shutdown, Edit the VM setting and highlight the SCSI Controller 0 and select Change Type as we did earlier and select Paravirtual. Once this is done you will see the original controller is replaced with a new controller.

ChangeLSItoPVSCSI

Now that we have the boot drive change to PVSCSI, we can now balance the data drives across up to four PVSCSI controllers for maximum performance.

To do this, simply highlight a Virtual Disk and drop down the Virtual Device Node and select SCSI (1:0) or any other available slot on the SCSI (1:x) controller.

ChangeControllerID

After doing this you will see new SCSI controllers appear and you need to change these to Paravirtual as we have done to the first controller.

ChangeControllerIDMultipleVdisks

For each of the virtual disks, ensure they are placed evenly across the PVSCSI controllers. For example, if you have a VM with eight virtual disks plus the OS disk, it should look like this:

Virtual Disk 1 (OS) : SCSI (0:0)
Virtual Disk 2 (OS) : SCSI (0:1)
Virtual Disk 3 (OS) : SCSI (1:0)
Virtual Disk 4 (OS) : SCSI (1:1)
Virtual Disk 5 (OS) : SCSI (2:0)
Virtual Disk 6 (OS) : SCSI (2:1)
Virtual Disk 7 (OS) : SCSI (3:0)
Virtual Disk 8 (OS) : SCSI (3:1)
Virtual Disk 9 (OS) : SCSI (0:2)

This results in two data virtual disks per PVSCSI controller which evenly distributes IO across all controllers with the exception being first controller (SCSI 0) also hosting the OS drive.

What if I have problems?

On occasions I have seen problems with this process which has resulted in VMs not booting, however these issues are easy to fix.

If your VM fails to boot with a message like “Operating System not found”, I suggest you panic! Just kidding, this is typically just the boot order of the Virtual machine has been screwed up. Just go into the bios and check the boot order has the PVSCSI controller showing and the correct virtual disk in first priority.

If the VM boots and BSOD or crashes and goes into a continuous reboot loop then power off the VM and set the first SCSI controller where the boot disk is running back to LSI. Then reboot the VM and make sure the PVSCSI driver is showing up (if its not you didn’t follow the above instructions) so go back and follow them so the PVSCSI driver is loaded and working, then shutdown and change the SCSI controller back to PVSCSI and you should be fine.

If the VM boots and one or more drives do not show up in my computer, go into Disk Manager and you may see the drives are marked as offline. Simply right click the drive and mark it as online and reboot and you’re good to go.

Summary:

If you have made the intelligent move to virtualize your business critical applications, firstly congratulations! However as with physical hardware, Virtual machines also have optimal configurations so make sure you use PVSCSI controllers with multiple virtual disks and have your DBA span the database across multiple virtual disks for maximum performance.

The following post shows how to do this in detail:

Splitting SQL datafiles across multiple VMDKs for optimal VM performance

If the DBA is not confident doing this, you can also just add multiple virtual disks (connected via multiple PVSCSI controllers) and create a stripe in guest (via Disk Manager) and this will also give you the benefit of multiple vdisks.

Related Articles:

1. Peak Performance vs Real World Performance

2. Enterprise Architecture & Avoiding tunnel vision

3. Microsoft Exchange 2013/2016 Jetstress Performance Testing on Nutanix Acropolis Hypervisor (AHV)