Cisco UCS X-Collection, Resolution Check Drives



Please welcome visitor blogger, Tushar Patel, Principal Engineer, Oracle Options 

Confirmed 37 Million IOPS, now augmented with 100G Networking and Built-in GPU – Oh My!

We dwell in a world of change.  Many people are sufficiently old to recollect a time earlier than cable tv, the web, and cell telephones have been commonplace.  Motion pictures made within the 1980’s gives a view of the previous.  But know-how continued to advance fed be a want to deliver, for instance, broadcast programming sooner and in a kind to save lots of the buyer complexity and value.  Tv reveals transitioned from a giant picket field within the household room to VCR tapes, to DVDs, to streaming on Netflix considered in your newest smartphone.  All of this modification within the span of a single technology!

Server structure has developed over the previous decade as effectively.  When Cisco launched the Unified Computing System (UCS) again in 2009, the Cisco UCS B-Collection 200 blade solely had capability for 2 disk drives and these housed the working system on one drive whereas the opposite served as a back-up drive.  All utility or database information have been positioned on exterior storage arrays akin to NetApp. This result in the event of Converged Infrastructures akin to FlexPod (Cisco UCS & NetApp) and FlashStack (Cisco UCS & Pure Storage) which grew to become a extensively accepted infrastructure to host massive database, ERP, and different mission crucial workloads.  The biggest of Oracle databases are hosted on Converged Infrastructures and there’ll proceed to be an ample marketplace for this method design.

Expertise continues to advance, and Cisco has captured many of those advances within the new Cisco UCS X-Collection chassis which homes as much as eight UCS X210c M6 compute nodes.  Consider these nodes as actually a rack optimized server that’s sled vertically into the brand new Cisco X-Collection Modular System (Cisco UCS 9508 chassis).   This new design is managed by Cisco Intersight, a cloud primarily based system administration functionality that leverages key options of the previous UCS Supervisor which managed the earlier technology methods.   Whereas there are lots of new capabilities we might elaborate on, for immediately, lets concentrate on the truth that the X210c has six NVMe PCI Gen 4 stable state disk drives (SSD) along with the 2 drives which home the working system.

Cisco UCS X9508 Server Chassis (entrance view)

This month the choices develop!  Database Directors (DBAs) take a look at each system design level to maximise efficiency.  Community bandwidth, historically the community admin area, was thought of out of bounds for the DBA. However, Cisco UCS X-Collection now helps 100G networking offering a capability for these groups to design new options delivering new efficiency thresholds.  Simply consider the probabilities!  This technique can simply drive your database, VDI, AI/ML, and different enterprise purposes.

With this new design, clients who’ve single occasion databases akin to Oracle or Microsoft SQL Server now have as much as 90TB of onboard cupboard space per X210c compute node.  System architects can use these drives to accommodate a single occasion of a database and keep away from the price and complexity of investing in exterior storage.  This additionally saves house and energy inside the datacenter with out sacrificing efficiency.

Servers supporting Digital Desktops can drive extra complicated purposes than ever earlier than.  Graphic intensive workloads require Graphical Processing Items (GPUs) to maximise efficiency.  X-Collection gives two choices.  The primary is the flexibility configure 1-2 Nvidia T4 GPU per X210c compute node.  The second is a devoted GPU compute node which is right for a big selection of workloads akin to VDI.

You may recall again in July 2021, John printed a  weblog that detailed Digital Desktop Infrastructure (VDI) testing that was carried out on a single X210c M6.  The outcomes proved that this compute node might ship wonderful efficiency.  Database directors studying it will level out {that a} database is much completely different for any variety of causes.  Whereas they’ve some extent, we wished to first decide a baseline efficiency for these drives.  It’s a typical observe to judge the system baseline efficiency earlier than deploying any database utility. System baseline efficiency testing is carried out utilizing frequent I/O calibration instruments akin to Linux FIO. These instruments can generate I/O patterns that mimic the kind of I/O operations carried out by Oracle databases.

In our exams, we used the FIO load generator to train 6 native Intel 5600 NVMe disks with 3.2 TB every. Although it is a single occasion oracle database operating on single blade server, we determined to run efficiency exams on all 8 absolutely populated X210c blades concurrently so we are able to be certain that every blade is maximizing its designed efficiency inside the X9108 chassis.  From this testing we might conclude:

  • No affect on a person blade server efficiency as a consequence of energy and cooling
  • No slot dependency. All blades obtain equivalent efficiency no matter slot contained in the chassis.

Check Standards

  • Precondition all disks by writing random knowledge patterns
  • Observe and validate sustained IOPs and efficiency over 30 minutes, 1-hour and 4-hours with common and 90th percentile sub-millisecond latency for numerous learn/write ratios. A number of runs for every iteration have been performed and cautious evaluation was carried out to confirm the info consistency
  • Generate reproducible outcomes with minimal parameter adjustments whereas mimicking utility habits. In different phrases, we might generate barely larger numbers however the objective is to not generate peak efficiency benchmark numbers. For every take a look at, acceptable FIO iodepth and variety of jobs have been chosen together with async IO as parameter
Efficiency Outcomes
  • Every blade server generated roughly 4.7 million sustained IOPs at 100% reads with complete round 37+ Million IOPs at almost 500 Microseconds latency with 4K knowledge blocks.
  • Close to An identical IO habits on all 8-blade servers

That is wonderful efficiency, higher than you may need imagined! Now with this baseline in place it’s time to load up a single occasion of the Oracle database 19c and see how the X210c M6 performs – which would be the matter of our subsequent weblog.  Keep tuned!