IOPS: How much performance you need for SMB or Enterprise?

Overview
In this post, I’ll be talking about performance data from the DiskStations and RackStations which I have tested – and how to apply this data to the world.

IOPS Testing Objectives
As a follow up to my previous blog post, “IOPS: Performance Capacity Planning Explained”, I stated that I will be conducting tests of select DiskStations and RackStations to measure performance data and to determine how to apply this knowledge. My testing philosophy is based on pragmatism, using the simplest settings wherever possible. Just as I value my time, and your time is valuable as well, we’re not going to waste time attempting to squeeze out every last IOP beyond reasonable and reliable results.

Because of this objective, SSDs are not used, nor is 9000 MTU. The reasoning behind this is that while using SSDs and hyper-optimized network settings can show the highest potential performance, SSDs are cost prohibitive in terms of value per gigabyte purchased. Adjusting MTU values can bring about additional headaches when dealing with network compatibility issues such as determining the optimal frame size for your specific network environment. Also, two-disk redundant SHR is used where appropriate.

Storage Hardware

• DS1511+, DSM 3.1-1748
• RS2211RP+, DSM 3.1-1748
• RS3411RPxs, DSM 3.1-1746, with Intel X520-T2 10GbE
• HDD: ST32000641AS, with firmware CC13
• RAID Type: Synology Hybrid RAID, with one or two disk redundancy
• Jumbo Frames: 1500 MTU

Benching Computer

• CPU: Intel Core i7-2600
• RAM: 8GB DDR3-1333
• NIC: Intel X520-T2 10GbE

Procedure

• Have the DiskStation or RackStation directly connected to the computer
• Set IOMeter to control one target on the DiskStation or RackStation, using a single NIC
• Run IOMeter with 4KB IOP, concurrent mix write/read operations, 33% Write, 67% Read.^{Ref: 1}

Performance Results

Example Performance Requirements
Running Exchange Server 2010, with 1000 users sending 500 messages a day, requires 600^{Ref: 2} “application IOPS”. Applying the best practice of a 20% buffer; this equates to 720 “storage application IOPS”.

With 720 IOPS, 240 writing, 480 reading IOPS are concurrently needed per Exchange Server. This means that the RS2211RP+ on a single NIC can theoretically provide storage for upwards of eight Exchange Servers when using iSCSI! The DS1511+ is no slouch either, as it can aid Small Businesses with supporting up to three Exchange Servers. The RS3411RPxs, given it’s higher performing hardware, can support storage for twelve or thirteen Exchange Servers, again, assuming 1,000 users sending 500 messages a day.

Of course, this is theoretical, as I don’t have a production Exchange Server with 1000 users – it’s best to observe your current read/write IOPS usage to determine your performance needs. One advantage is that these IOPS numbers are attainable with regular mechanical drives, and not using SSDs.

Using this same logic, here’s a summary table denoting how many servers can be hosted by the DiskStation, or RackStation over iSCSI, assuming a single network link.

Estimated number of supported Application Servers for a Synology DiskStation or RackStation

Summary
I hope this data proves useful to help you plan your performance and storage capacities for your networks. As a reminder, it’s always best to observe your own network environments to know how much performance you currently are using before moving your storage to a Synology RackStation or DiskStation.

Preview
Stay tuned! My next blog post will be discussing the popular question, is a DIY NAS more affordable than a DiskStation. I will provide a few examples to illustrate the differences between the methods. This information will help both aspiring and veteran IT Professionals further understand what they are getting into.

Note: While some of the situations described in this article are theoretical, these situations are based on actual performance data gathered in the Synology labs. As always, performance numbers are a reflection of not only the storage hardware, but every other element of the network as a whole.

References
1. What Causes Exchange Disk I/O
2. Understanding Database and Log Performance Factors