Oracle Exadata Reviews

Performance with use of storage compute.

Single appliance with no additional tooling required for deployment reduces overall maintenance and cost of ownership.

Cost, need for additional licenses for core Oracle products.

Two years.

Yes, Cisco switch is not a standard within our company. Need to deploy with Oracle switch, after initial validation and setup, un-plug all network, replace the Cisco switch, and recable. This is time consuming and should not be required if Oracle would simply put in a different Cisco switch.

No.

Yes, memory limitations, 756gb limits the performance of databases.

Platinum support services supplied with the Exadata is worthless.

If I have an application down, we do not call oracle support as their response times to get a qualified engineer on the phone take 4+ hours. We solve problems ourselves.

ODA and standalone.

Scalability and consolidation.

See the switch issue mentioned above.

Implemented in-house with the assistance of Technologent.

Negative.

Total ROI should include the HW and all the associated SW licenses required. With an EXA, the cost to implement is prohibitive when also performing replication as you have to procure two of these very expensive boxes.

ODAs.

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.

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.

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.

I've been using Exadata since V2 in 2009.

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.

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.

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.

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.

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.

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.

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.

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.

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

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.

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.

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.

Oracle Exadata is a flash disk, so my company's primary use case is achieving target performance on the database level.

What I like most about Oracle Exadata is its smart scan feature.

I also like that it supports higher capacities and it's high-performing, so my company can use Oracle Exadata for massive databases.

Oracle Exadata is one of the best products of Oracle, with the best engineering.

One small area for improvement in Oracle Exadata is integration, particularly at the consolidated application level.

Additional features I would like to see in Oracle Exadata in its next version include a cloud interface and GUI.

I have around one year of experience with Oracle Exadata, and I'm still working with it.

Oracle Exadata is a nine on a scale of one to ten, stability-wise.

Oracle Exadata scalability is a nine out of ten.

The initial setup for Oracle Exadata is straightforward and user-friendly because it has a great feature called OneCommand, which makes installation automated.

Both in-house and vendor teams implemented Oracle Exadata.

I'm giving Oracle Exadata five out of ten in terms of pricing. Its price could be lower.

Compared to other solutions, Oracle Exadata is more valuable because it has excellent features and it's a one-box solution.

I've worked with different versions of Oracle Exadata, such as X7, X8, and X9.

My customers use Oracle PCA (Private Cloud Appliance) for Oracle Exadata deployment. Banks use a private cloud, not a public cloud.

My company is a platinum partner of Oracle, with sixty to seventy customers using Oracle Exadata. In terms of maintenance, forty to fifty technical staff handle the maintenance for all customers.

My company is an Oracle vendor working with customers on implementation and deployment projects. The company is a reseller of Oracle Exadata.

I'm rating Oracle Exadata as eight out of ten.

My primary use case is for EIMS (Enterprise Information Management Systems). We run this in the night time, and every day in the morning. Then the reports are readily available, so there is zero failure.

This improves the way our organization functions because it is on X86 machines. The X86 machines and also because the internal fiber channels are internally integrated, but while you want to have it in the network because integrating with the network, that is the external network(the tape drives and all). Those things are really cumbersome.

The reports are always readily available. The backup solution that is the data file system is available for a 30 day insert. There is really no other software solution that gives us the same quality backup and restoration like this.

We have had issues with system restoration. If there is a system fail, the internal drives are useless. We need Exadata to integrate with that one, and to restore it back. But there should be something in place in case of system failure. One of the systems should have the facilities to troubleshoot from the other nodes to this note remotely. Or, even if not remotely, just to troubleshoot at least among the thrusters. 

There is no system collectively. Let us say this system failed; do we have any disks? Total disks, let's just say 100 disks. Five disks are failed. 100 disk measuring is not data top. If mirroring is that, like we can HRCMR. Let's say IBM CRM are CTC shadow image, REMC like that, the storage level they are giving the facilities inserted data mirroring, so immediately we can bring the DR. But internally, the Exadata doesn't have such features. Maybe they can think through that mirroring of collective system. Insert automatic measuring. If we have a collective recovery, let's say system-to-system, that DR solution, though they are giving a data guard, we have to set it up into the second system and the Data guard is a software solution. There is a really high skill required for a DBA to bring up the database. So a non-skilled person, like the storage engineer, can bring up for shadow images. HRCMI, HRCMR, like that. But in the case of Oracle, you must be really specialized to to bring up the data guard.

In addition, they should insert automatic measuring. AM is, let's say, one table fail. Okay, you can recover it. Or one disk failed, then they can troubleshoot it. But automatic recovery, fine, but if we have a collective recovery, let's say system-to-system, that DR solution, though they are giving a data guard. We have to set it up into the second system and the Data guard is a software solution. There is a really high skill required for a DBA to bring up the database. For a non-skilled person, like the storage engineer can bring up for shadow images. HRCMI, HRCMR, like that. But in case of Oracle, you need to be really specialized to  to bring up the data guard.

It is really super-stable.

The scale is somewhat limited. The existing axis can only have four blades.  So, there are limitations. The storage is around 200TB to 400TB, which is not infinite storage.

The technical support team are real professionals. I admire their technical skills and supports. Their supports are really admirable.

We had very complex situation, actually, with our Exadata migrations. Due to the shortage of our manpower, one of our engineers and one and also Oracle did migrations. During the migration phase, they have a lot of discussion and talks and deviation. But, they had someone help us migrate excellently. Because we have a data that's set up and also because it is not one-to-one relationships. End-to-end relationships, we had it over there. Database level, not table level. Database level, we have physical data within some other machine. The secondary data will be here. The data guard will be inside the Exadata, and the Exadata itself have the data guards and it's really nurtured relationships further, but some start up guys also have come back to the preparatory work center. 

All in all, in Oracle, you need a skill set for implementation. I don't say it is difficult, but you need a skill set for it.

When considering a product, I make sure that I have:

• Manpower resource available.
• The number of DBAs available in the market

The price is very high. But, it is worth it.

The system is engineered to run both databases and enterprise applications unlike other engineered systems, which are either tailored only for databases or only for applications. The ability to consolidate all of the databases is a big plus.

The SuperCluster provides best of both worlds – Oracle Exadata functionality along with virtualization at the firmware and the OS kernel layers.

On the Exadata side the features most valuable are – Hybrid Columnar Compression (HCC) for both Data Warehouse and OLTP workloads, Storage Indexes for Smart Scan, ability to use Flash Cache for DB storage, and more.

With SuperCluster and Exadata, all of the servers and storage are integrated within the same rack. This reduces the configuration and setup time, increases performance, and makes maintenance and patching easy. We are able to consolidate all of the databases and application stack on a single SuperCluster with Exadata.

In the area of Solaris zone-level virtualization, it would be good to have memory capping as a tool for memory management. Currently for 11g databases running on Exadata with smart scan against hybrid columnar compressed (HCC), tables could result in errors. It would be nice to have a patch rather than the current solution of upgrading the databases to 12c.

Currently, we have a Half-Rack SuperCluster with 2 SPARC T5-8 Compute Nodes and 4 Exadata Storage Servers. Current version of Exadata is 12.1.2.1.3. The compute nodes are running Solaris 11.2 with Oracle11g 11.2.0.4 databases. We've been using it for close to three years.

We had few issues while deploying DB zones. This virtualization has to be carried out differently when compared to DB LDOM virtualization.

There have been no major problems so far with stability.

There have been no major problems so far with scalability.

Overall it has been good so far.

Overall it has been good so far. In the case of engineered systems like SuperCluster/Exadata, the patching has to go through the support team and there is definitely room for improvements in this area.

Previously, we had multiple servers both with and without physical partitioning. The storage for all of the servers had to be zoned to a SAN. The servers and the storage previously were from different vendors and we had to integrate those. Other product offerings were evaluated, but with all of those we had to explicitly integrate the compute, storage, and networking components. In addition, we could not get the benefits of database optimizations with Exadata and have to pay the penalty for virtualization overhead and network traffic between compute and storage layers.

There is a certain degree of complexity with respect to the initial design of the Exadata storage cell disks and grid disks to meet the customer’s application needs. This especially true when migrating from an existing setup. Care has to be taken with regard to the initial domain configurations since this will determine the LDOM and zone-level virtualization. The Exadata disks have to be exposed to both the DB LDOMS and DB Zones.

The initial setup was from the vendor team (for any engineered systems from Oracle). But later we had to continue the setup to cater to our application and business needs. Prior to the implementation, all of the IP allocations are to be completed for the three layers – Client/Public network, InfiniBand private network, and the Management network.

At this time, I do not have the actual numbers but would rate the ROI is pretty good.

In the long run one can consolidate the various DB related licenses. The number of cores required to run the DBs and applications is much lower on a SuperCluster with Exadata, thereby leading to fewer licenses and reduced cost. Since the entire stack is owned by one vendor (in this case Oracle) the core factor for licensing is 0.5 leading to fewer licenses for the software components.

The SuperCluster in addition to Exadata storage also comes with ZFS storage cluster. Since the compute nodes, storage and networking components are fully integrated with InfiniBand I/O fabric this provides very high performance between various components. Also, it has built-in hardware encryption to provide data security.

There are so many it’s hard to just name a few but -

• Cell off-loading
• Reliability
• ExaCheck
• InfiniBand

Many diverse applications were migrated to the Exadata platform that had either performance or stability issues. All have had very reliable and performance since.

Their hardware engineers can’t be beat. They are the best people I have ever worked with and parts are installed very quickly and efficiently. They’re very easy to work with, schedule jobs with, and get things done. Also, they are very proactive as well.

Most of my suggestions for Exadata itself have been addressed in X5. Oracle keeps improving the reliability and adding more hot swappable parts for the hardware and software.

The only improvements left are not with the product itself but with MOS support, ASR and patching. For MOS software support, there are some very talented people there, but not enough of them. They need more software engineers with in depth training on how to use the ASR system and the jump gate.

When the jump gate is updated or patched, it seems to take a bit to get things working properly again so patching and ASRs can be done. Platinum patching continues to improve but it is not a hands off solution.There is quite a bit of manual effort involved with each patch cycle to co-ordinate the patching. With some of the patching there, is downtime, and most of it is rolling but there are a few cases where you do need have downtime.

Since v1, almost six years.

Just the normal software bugs, nothing major. In fact considering the complexity, there are very few bugs at hardware/software level. The ASR system with the jump gate was more problematic than the Exadata setup.

No, very stable and reliable. On one quarter-rack we have 27 databases running, which astounds me every time I patch and do maintenance.

We have added more memory and additional storage arrays, and adding more memory is very easy, while adding a storage array takes some time, but it is not as complicated as I expected. We will be expanding our local file system soon not sure how complex this will be. Adding ZFS is fairly straightforward and has become easier since I first did it three years ago.

My hardware support is beyond compare. The MOS software support is good but there are just too few experienced people.

AIX, and we switched partly due to a hardware refresh, and partly due to performance, and the growth of some of the applications.

Documents are very clear, and there were many software and hardware engineers available to provide assistance and guidance along the way. The most difficult piece was the ASR system set up with the jump gate, password vault and set up in OEM. The other difficulty was the network, and some of that was due to my lack of experience in that area.

We had one application that went down constantly and since we moved to the Exadata there have been few issues outages and incidents.Huge performance gains as well.

Really evaluate your needs from a hardware and software end. Buy what you will need for between three and five years down the line, in terms of storage and memory. Buy products either before Oracle’s fiscal year end (May) or Open World. You can bundle all of your Oracle products into one agreement to get the best discounts. Try to get training and some consulting time into the agreement.Some companies leverage the partner or beta programs. I did one such beta test and was glad I did because I made improvements.

We had tried other storage and hardware options but nothing really got us performance we needed for some of our applications. We migrated applications from AIX to Linux VM and Exadata. The most critical or resource intensive applications go to Exadata.

There is a learning curve to this product and it is complex at the hardware, network and software levels. Train all your staff, not just the database administrations as the network and UNIX admns have to get familiar with the components. Exadata classes are available and the DBAs should know RAC and services well.

The starter databases are a template for new databases. Run ExaChks weekly, and set up OEM to monitor Exadata, but spend time to make sure you do it correctly. If you have a large Exadata foot print look at ZFS for backups and evaluate if you have the staff to support it. If not make sure you partner with the correct consultants.

Oracle does provide Platinum patching services but that has to be managed. Read about the product and really evaluate how to use it properly. Talk to other customers and join Exadata groups. Do a proof of concept, so see if there is a sandbox to try out. And probably the best hint is that MOS has an Exadata best practice guide - follow it as best you can. Granted there are some things you have to do for vendor products. But the more you can get out of these the better off you will be for patching, MOS etc.

We used to work on business intelligence projects, so we really enjoyed the service-like performance.  It's been great.

The business intelligence is very good.

We have found that the analytics environment is excellent.

We've discovered that the solution is quite a complex product, which can make it difficult to sell. Snowflake, for example, is simpler and therefore an easy sell.

The customization can sometimes be difficult to achieve.

We've been working with the solution for a very long time. It's been more than 15 years personally and at least 15 years at this company, so it's been well over a decade at this point.

We don't deal with bugs or glitches and the product doesn't crash or freeze. The only real issues we come across are related to customization. It's quite stable. We haven't had problems.

We have about 100 people using the product currently.

Our clients are quite sizeable companies.

I've never used technical support and therefore can't really comment on their responsiveness or level of knowledge. I have an internal team that assists with troubleshooting as necessary.

As a consultancy, we work with a variety of products. This includes Oracle and Snowflake, for example.

We are an integrator of the solution. Our clients are sizeable companies, and there are several providers involved in different projects with Exadata. We're not the only administrators. 

I personally haven't handled an installation. We're mostly focused on projects related to BI and analytics. Someone else handles the setup and we just use the product or have our clients use it. I don't have a sense of whether the implementation is difficult or complex or how long deployments take.

We are a consultancy. We are partners with Oracle, with Snowflake, and with other vendors, software vendors.

We work in different deployment models with our customers. We try to fit our customers' needs. It depends on the customer and the project in terms of which deployment model we'll recommend or use.

Overall, I would rate the solution a nine out of ten. It's very good. We've had a great experience using it over the years.