Home > Storage > ObjectScale and ECS > Product Documentation > ECS with F5 Deployment Reference Guide > F5 Overview
The F5 BIG-IP platform is a blend of software and hardware that forms the foundation of the current iteration of F5’s Application Delivery Controller (ADC) technology. On top of the BIG-IP platform are a suite of products that provide a wide range of application services. Two of the BIG-IP products often used in building an organization’s foundation for local and global traffic management include:
F5 offers the products in both physical and virtual form. This paper makes no recommendation between choosing physical or virtual F5 systems. We do recommend that F5 documentation and product specifications be consulted to properly size F5 systems. Sizing a BIG-IP system should be based on the cumulative current and projected workload traffic bandwidth (MB/s), quantity of operations (Ops/sec) or transactions per second (TPS), and concurrent client sessions. Properly sized BIG-IP systems should not add any significant transactional overhead or limitation to workflows using ECS. The next few paragraphs briefly describe considerations when sizing ECS and associated traffic managers.
There are two components to each transaction with ECS storage, one, metadata operations, and two, data movement. Metadata operations include both reading and writing or updating the metadata associated with an object and we refer to these operations as transactional overhead. Every transaction on the system will have some form of metadata overhead and the size of an object along with the transaction type determines the associated level of resource utilization. Data movement is required for each transaction also and refers to the receipt or delivery of client data. We refer to this component in terms of throughput or bandwidth, generally in megabytes per second (MB/s).
To put this into an equation, response time, or total time to complete a transaction, is the result of the time to perform the transaction's required metadata operations, or the transaction's transactional overheard, plus, the time it takes to move the data between client and through ECS, the transaction's data throughput.
Total response time = transactional overhead time + data movement time
For small objects transactional overhead is similar and is the largest, or limiting, factor to performance. Since every transaction has a metadata component, and the metadata operations have a minimum amount of time to complete, at objects with similar sizes that component, the transactional overhead, will be similar. On the other end of the spectrum, for large objects, the transactional overhead is minimal compared to the amount of time required to physically move the data between client and the ECS system. For larger objects the major factor in response time is the throughput which is why at certain large object sizes the throughput potential is similar.
Object size and thread counts, along with transaction type, are the primary factors which dictate performance on ECS. Because of this in order to size for workloads, object sizes and their transaction types along with related application thread counts should be cataloged for all workloads that will utilize the ECS system.
BIG-IP DNS and LTM each provide very different functions and can be used together or independently. When used together BIG-IP DNS can make decisions based on data received directly from LTMs. BIG-IP DNS is recommended if you want to either globally distribute application load, or to ensure seamless failover during site outage conditions where static failover processes are not feasible or desirable. BIG-IP LTM can manage client traffic based on results of monitoring both network and application layers and is largely mandatory where performance and client connectivity is required.
With ECS, monitoring application availability to the data services across all ECS nodes is necessary. This is done using application level queries to the actual data services that handle transactions as opposed to relying only on lower network or transport queries which only report IP and port availability.
A major difference between BIG-IP DNS and LTM is that traffic flows through an LTM whereas the BIG-IP DNS only informs a client which IP to route to and does not handle client application data traffic.