DSS/OLTP

Hi Tom,

Actually, OLTP and DSS typically refer to a type of workload, not a type of database.

OLTP (online transaction processing) is often categorized as ~70/30 read/write mix with a lot of random read activity. Its measuring criteria is likely to be IOPS and IO response time.

DSS (decision support system) is often categorized as read-only (or mostly) workload, with large sequential reads or writes ('batch' and ETL). It's measuring criteria is likely to be GB/sec (bandwidth).

The databases in your example above (Oracle and Sybase) can both run either OLTP type workload, or DSS (though in Sybase/SAP case it will be different database engines: ASE, IQ, or HANA).

Back to VMAX: it does fantastic for both type of workloads (and others). You can find an Oracle example for VMAX AF with OLTP and DSS type workloads in the following white paper:

http://www.emc.com/collateral/white-papers/h14557-vmax3-all-flash-storage-for-oracle-database-wp.pdf

Thanks,

Yaron

HI Yaron,

Thanks for the good reply.  I might have worded my inquiry somewhat incorrectly.  I was trying to identify examples of host apps/services that fall under the workload type of DSS.

From what I gather from your reply, a database app could fall under either workload type, based on the monitored IOPS and response times.  Am guessing a nighty backup (ie.. Netbackup) could be considered DSS due to the Read only nature on source side frame?

Also, in the example of an ESXi cluster, where an oltp workload "might" be virtualized, I guess it would be safe to say if the random IO of OLTP dataset was properly placed in storage groups, OLTP would be the workload for the DB and trans logs as DSS, each isolated by SG and workload configured?  And given this, the fact the app was running on an ESX cluster wouldn't really matter?

Just a "random" thought while I was typing ...would be the implementation of ENas on vmax3 with it's random IO workload, lend itself to that of an OLTP workload? 

Thanks,

Tom

Hi Tom,

I guess we could say that the problem with the database and application world is that like most worlds, it isn’t perfect. As such, you are correct by saying that an application that during the day behaves as an OLTP type, at night shifts to a batch/DSS like behavior. This happens due to natural business cycles where the operational systems’ data need to produce summary reports, run backups, or the data extracted, cleansed and sent to other systems such as Data Warehouses (DW), Business Intelligence (BI), OLAP, etc. The result is that the border between OLTP and DSS softens. When today I ask DBAs if their database is an OLTP or DSS the answer is most often – both!

There is another trend that soften the boundary, which is in-memory databases (IMDB). Being that IMDBs are very fast, they are often positioned as a single repository of data for both OLTP and DSS alike. I’m not a great believer in this trend since a true BI or DW indeed shows a DSS workload behavior (large data scans, sequential nature and so on), however, their *content* is not the same as the operational systems. It is often an aggregation and summary of data from multiple operational systems and requires some cleansing first (often data in different operational systems isn’t inserted in the same way, may include missing records, too detailed, etc.).

Regarding your question of treating transaction logs as DSS my take is that although like you said the logs are typically sequential write (and reads for archiving), we tend to use the terms DSS or OLTP for a whole database or application, and more often to the nature of how the data is being accessed, not so much the transaction logs. Still, you’re right if you only consider the access pattern per storage group.

Regarding eNAS on VMAX3/AF, we don’t necessarily distinguish between FC, iSCSI (block storage), or eNAS (file storage) as far as application type and leave it to the customer to decide how they want to access their data. You can find a white paper on VMAX3 with SQL Server and eNAS here.

To actually go back to what you asked about – “I was trying to identify examples of host apps/services that fall under the workload type of DSS.” I’m not an expert in this space to provide a list, however, if you’re interested, one approach is looking at Gartner’s “Magic Quadrant” (example here, just click on the image to magnify. The full report goes into more details though cost money). Or google the subject (here is one result I found). Since I didn’t read closely either of these two links, I just bring them as examples and not advocating their content However, note how they usually point to an application and no a database. If you dig deeper, some of these application use the same databases OLTP systems may use, and other have their own.

Thanks,
Yaron