Replication between remote locations can be challenging due to a number of factors. These include bandwidth/latency, security, costs and long data transfer times, just to name a few. For EqualLogic customers, WAN acceleration and optimization products, such as the BIG-IP® product family from F5 Networks, can encrypt traffic and help dramatically reduce transfer times and improve bandwidth utilization, and allow for replication across greater distances. F5 was on the Dell campus for work on a Dell Power Solutions article and while here they dropped off a couple of BIG-IP devices and a WAN emulator in the Dell TechCenter lab, and allowed us to run a few WAN EqualLogic replication tests of our own.
Updated: Read the CSG Systems Direct case study to learn how Dell and F5 helped CSG better meet Disaster Recovery SLAs and significantly decrease EqualLogic data replication times between data centers.
F5's BIG-IP® product family offers the WAN Optimization Module (WOM), which allows for the following, according to the WOM datasheet:Encrypt and accelerate data between high availability BIG‑IP devicesOptimize, encrypt, and tunnel data transfers between two or more BIG‑IP LTM devices.Reduce server and bandwidth usageCreate TCP optimization and symmetric adaptive compression between BIG‑IP devices.Improve user performance and scalabilityControl traffic and prioritize bandwidth to ensure that users accessing critical applications across the WAN to always get the fastest performance.Accelerate applications across the WANAccelerate data transfers over the WAN for faster file transfers, email, client-server applications, data replication, and more. Features include iSessions tunnels, SSL encryption, adaptive compression, symmetric deduplication and TCP optimizations.Consolidate WAN optimization and application deliverySave on hardware costs, rack space, and energy consumption as well as management resources with consolidated services.For more information on WOM acceleration, check out this white paper by Lori Mac Vittie (F5) entitled "Understanding Advanced Data Compression".
We put two EqualLogic Groups on either side of the BIG-IP 3600's with the WOM Module, per this diagram. This lab is isolated and has a simulated WAN link at 45Mbps, 100ms latency and 1% packet loss (10,000 packets dropped per million). Traffic Flow Description: The EqualLogic iSCSI replication traffic flows from the Central Office array to the central BIG-IP, the BIG-IP's work together to encrypt/compress/deduplicate/TCP optimize the traffic as it goes across the simulated WAN link. The Branch Office BIG-IP decrypts the traffic and forwards it to the branch array. All tests were run with SSL encryption enabled.Datasets: We created three EqualLogic volumes and connected a Dell PowerEdge Server running Windows to configure and populate the volumes. The file systems were formatted with NTFS and the file datasets were copied to each drive.Volumes and File Datasets
We used small 2GB test volumes because we did not have the time to spare for long duration test runs. This makes the results a little volatile from test to test but gives you the basic idea of how this set up works.Deduplication Store: In the configuration video, you'll notice that we set the deduplication cache store to Memory Mode (vs. Disk Mode), which provides fast access to the de-dupe cache but limits the size of that store. This setting often works well for data replication applications. Depending on the dataset(s), the disk store may provide better performance. During our testing we wanted to experiment a little so we tried some different datasets and adjusted the de-dupe cache store mode to see the effects. As expected, we found that the type of dataset replicated has an impact on the throughput as well as the optimal setting for the de-dupe store. The majority of our tests were run using Disk Mode.
Note: These results are for demonstration purposes only. They should not be considered benchmarks of any kind.
Notice the high throughput numbers for the f5vol4 volume. This dataset is 99% compressible and the files are all duplicates of one another. This is not a likely/realistic dataset but we used it in our POC to demonstrate how the file content can effect the accelerated throughput.
Running concurrent volume replications (3 volumes simultaneously) reduces the individual volume throughput numbers but greatly increases the overall WOM effective throughput. In both cases (single and concurrent replications), the simulated WAN link bandwidth showed fully utilized as we monitored the Central Office BIG-IP server-side throughput performance statistics. It should be noted that this POC lab has the full WAN link bandwidth at our disposal for the tests but in a real world scenario it is likely that there will be other traffic and applications using the same WAN link. Anyone planning to deploy WAN optimization should thoroughly research their production networks to determine WAN conditions including available bandwidth, latency, packet loss and other characteristics. BIG-IP supports L7 QoS Rate Shaping which allows for the enforcement of bandwidth minimums and maximums per application, including burst control and support for terms of service (ToS) and Differentiated Services Code Point (DSCP).The benefits of WAN optimization can be monitored using the WOM Performance Dashboard. Here are dashboard screen shots from the POC lab during the single volume replication of f5vol2, second pass w/cache.Bandwidth Gains - Symmetric DeduplicationOptimizations - Compression
This two minute video shows the steps taken to configure the Central Office BIG-IP for the DellTechCenter lab set up.
F5 BIG-IP LTM VE TrialThe leading Application Delivery Controller from F5 Networks goes virtual. BIG-IP® Local Traffic Manager™ (LTM) Virtual Edition (VE) is now available in a 90-day trial version, allowing you to test, demo, and develop in a trial environment. Production versions will be available by summer of 2010.