What is most valuable?
My top 4 most important features of Exadata are:
1. Smart Scan, the ability to offload intensive SQL workloads to the storage servers. Queries are offloaded to the storage layer and only the result sets of relevant data are returned to the database server thus significantly improving performance.
2. Exadata Hybrid Columnar Compression, where we can compress data from 10x to 50x. Deploying databases on the Exadata can significantly reduce the amount of storage that is needed.
3. Exadata Smart Flash Cache automatically moves data between DRAM, flash and spinning SAS or SATA disks to provide best performance.
4. Virtualization is a new feature introduced to the X5-2 family. Now Oracle’s Database Machine can be catered to large enterprise mission critical databases and can house smaller databases that need isolation and now even application servers together with the database. We can connect data intensive applications to the database over low-latency, high throughput infiniband.
How has it helped my organization?
We are able to provide a complete solution to our customers from data center installation/configuration, firmware upgrades, ILOM upgrades, OS patching/upgrades, cluster patching/upgrades and database patching/upgrades. We can upgrade the entire stack in a single evening with minimal outage. Depending on the customer’s tolerance for downtime, we can either perform the patching or upgrade in a rolling fashion.
Patching and upgrade services have proven to be a value-add differentiator for SMB and mid-market customers where resources and budgets are often limited. A single consulting resource can patch/upgrade the entire Exadata stack because patching and maintenance has become reliable and simplified.
What needs improvement?
For the initial instantiation process, the Oracle Exadata Deployment Assistant (OEDA) could have better error checking and pre-check validation as you navigate through the tool. When executing the OneCommand utility, generated logs are decent but the logs are not detailed enough to pinpoint to where the error occurred in the stack. Oracle can do a better job with error isolation. After the OneCommand, other one-off commands have to happen (i.e. we have to login to the infiniband and cisco switch). Ideally, OEM could be leveraged to configure the remaining components of the Exadata after the OneCommand. The idea would be to reduce the number of people required to support the stack. If we leverage OEM, we can leverage a single resource that minimally understands the stack to support the workflow.
For how long have I used the solution?
I've been using Exadata since V2 in 2009.
What was my experience with deployment of the solution?
For new DBA's trying to understand the Exadata infrastructure, they struggle initially trying to understand all the components between ILOM, ASR, and OEM (what they manage, what they support, and what problem area that they detect) and which does what since they are overlapping. Consolidation management across the Exadata is another common issue on the Exadata. OEM is good at managing individual targets but not so great yet on identifying culprits across environments with heavily consolidated databases. When you look at the compute wholistically, it is difficult to identify what database is consuming most CPU and most I/O resources. Doing any kind of showback is difficult to do across instances.
What do I think about the stability of the solution?
I only see stability issues with Exadata when I see too much consolidation, and/or the Exadata is not sized properly. Often the customer tries to throw everything and anything on the Exadata, by over-parallelizing OLTP/batch processes without any resource management across any databases.
What do I think about the scalability of the solution?
With the Exadata, if I see an issue with scalability, it is typically goes back to being a sizing issue. The real question I have to ask is: did you get the right Exadata configuration for your database(s). If your Exadata configuration is sized properly, you should not have scalability issues.
If you let every database see every CPU on the Exadata compute node, you can potentially run into scalability issues. Customers who do not take advantage of database resource manager or IO Resource Manager (IORM) often run into performance issues in a consolidated environment. Likewise, if a customer tries to over-parallelize their application code, it can cause scalability issues. We tend to see more issues with improper management of parallel execution on the Exadata because it is perceived as something you can throw anything at.
How are customer service and technical support?
The Exadata stack is well known and has become a standard platform with Oracle customers; thus, triage to resolution has become much more streamlined. When you create a service ticket with Oracle Support, time to resolution is significantly reduced. On another note, Oracle Field Support engineers are remarkable. They have been quick to respond, flexible, knowledgeable, and willing to work with our schedule. Technical Support
Oracle Support Engineers are outstanding but you have to know how to navigate the system. Oracle Support provides all the technical support for the Exadata including the database software, OS, and hardware. Not knowing how to navigate through Oracle’s support structure and escalation policies can leave you feeling stranded by the vendor. Logging a support ticket with any component of the Exadata is no different than logging a support ticket for a database issue. You need to know how to raise a severity for a service ticket and how to escalate a support issue with the duty manager when production issues occur. Having a seasoned DMA (database machine administrator) is crucial to a successful Exadata deployment.
Having said that, often with Exadata customers, they can create a service ticket with the wrong Oracle Support group. This can cause confusion and elongated response times at early stages of the Exadata deployment as the service ticket gets routed to different teams within Oracle Support.
Which solution did I use previously and why did I switch?
Lot of our customers come from best of breed technologies (UCS/Dell/HP, EMC/Hitachi) to choosing Exadata. Typically, our customers choose Exadata for pure performance in IOPs, throughput, and low latency for their database workloads; however, we have seen a trend of customers choose the Exadata platform because they are short staffed, have high rates of attrition, and thus, have inability to support the hardware and software technologies.
How was the initial setup?
The initial setup is straightforward with the Exadata X5-2. We have to re-image the factory Exadata and leverage OneCommand for configuring the compute and storage nodes.
What about the implementation team?
A lot of the initial setup is configured by the Oracle ACS (Advanced Customer Support) organization; however, Viscosity is a certified Exadata implementation specialist and often perform the initial installation and configuration at the customer data center. By the time the customer receives a fully configured Exadata, they are ready to deploy databases as a RAC or non-RAC database.
What was our ROI?
For our customers, we are able to significantly reduce both CapEx and OpEx for customers for 3-5 years TCO. We are able to:
- Accelerate implementation to meet the functionality demand from the business users
- Deliver quick implementations to meet the market demands
- Lower implementation costs
What's my experience with pricing, setup cost, and licensing?
Setup costs for the Exadata varies from customer to customer and depends on database size, number of databases, and number of applications. For our most recent customer in Dallas Texas, the cost rolled up to approximately 800k for 2 X Quarter racks, which include storage cell software, and 25k for setup services. To determine day-to-day cost, we estimated the cost to average about 7k per month for both QTR racks.
Which other solutions did I evaluate?
A lot of our customers typically look at two other options. We have seen customers perform side-by-side evaluations by building their own high performance system with EMC/Pure Storage/Violin All Flash Array and UCS/HP/Dell blades or perform comparisons with a converged system such as VCE‘s Vblock. Customers typically ask for a proof-of-concept demo and run performance benchmarks with their own database and application to see the immediate impact and value-add for their organization. Our last 3 customers have compared Exadata with the Vblock.
What other advice do I have?
For new customers who are about to embark on the Exadata journey, they should consult with a vendor specializing in Exadata implementations for the first set of database migrations and technical direction. Customers should also do the proper sizing exercises either with Oracle or with the Exadata Specialty niche vendor to buy the suitable Exadata configuration what will suit their business needs for the next several years.
When purchasing Exadata, they should also look into purchasing either the ZFS Storage Appliance (ZFSSA) or the Zero Data Loss Recovery Appliance (ZDLRA) to offload their backups to leveraging Infiniband technology for maximum throughput.
OEM CC 12c provides a comprehensive monitoring and management of the Exadata platform. Not only can OEM monitor and maintain at the hardware level for compute, storage and network but also at the OS, cluster and database level. OEM Cloud Control can monitor all components of the Exadata.