Ask the Expert: Multicore Optimization (MCx) in the New VNX

Jim,

As you saw in the Mega-launch, the new VNX-8000 with MCx now scales to 5x over the older VNX in NAS SPECsfs ops. In the previous VNX generation, the Flare block backend would limit the overall NAS file performance. With MCx, the backend now scales to handle multiple Data Movers throughout the family. The largest increase is seen in the VNX-8000 which can deliver as much as 580K SPECsfs ops (see our publication on spec.org). Of course, as you move down through the new VNX models, the number of active Data Movers needed to saturate the block backend decreases. However, the performance is higher across the board for all Unified NAS platforms.

For example, the VNX-5800 can deliver as much as 40% more NFS ops than a VNX-5700 with 2 active Data Movers.

Thanks,

Phil

Thanks for your question. I am researching this topic for you and hope to post an answer soon.

Thanks,

Phil

Sorry for interruption, Can anybody tell me what was the change to CS in NG VNX ?

I found guide says dual CS is default , would this be it ?

Hi there ...

I would like to get your opnion about Symnetric Active-Active..

This is the story ..

Current VNX >

Load balancing by using ONE HBA connecting SPB/SPA as it was Active-Passive access .

see below ..

NewGeneration VNX >

Using symmetric active/active access ,  it will be possible to configure VNX like a VMAX , or Use ONE HBA per One SP .

( see below )

Which would be the best practice for config VNX with New Generation VNX ? ( one HBA with 2sp ? or One HBA per SP ? )

I think VNX2 users will often put multiple hotspares in an enclosure because of the new hot spare policies.
For example, a user chooses "No Hot Spares" as his/her hotspare policy and puts several blank disks into a DAE for future use, all the blank disks will be able to work as hotspare.

I believe CX and VNX uses the lowest numbered slot's disk as hotspare if all the hotspare-disk conditions are the same except the slot number in an enclosure. (If I am wrong, please correct me!)
How about VNX2? Is it still the same?

[Let me clarify the question using an example]
Suppose a DAE has three hotspares on its slot0, 5 and 14. All the disks in the DAE are the same (type, capacity, rpm, etc..). When the disk in slot6 fails, which disk, slot0 or 5 or 14's, is kicked in as hotspare? Is it slot0's disk?

Disk drives are ordered based on enumeration.. that is when the drive was first discovered by MCR it is assigned a logical identifier.  This logical ID stays with the drive for it's lifetime. This is not slot based.

MCR supports the concept of a portable drive.  You can literally pull a drive and put it in a different slot.  It will come online with the same logical ID.

For sparing MCR will choose the same type of drive first but will prefer the same enclosure.  If an exact spare type is in another enclosure it will choose that.  If there are multiple exact spare matches in the same enclosure then MCR will assign the lowest numbered logical ID. 

can logical ID be viewed from navicli ?

Symmetric Active/Active access is a feature of FLUs.  Once one of the Advanced Data Services (such as SnapView) are used, the LUN is changed to Asymmetric Active/Active (ALUA).  You would only change your environment if you were only using FLUs without any of the Advanced Data Services.

No it cannot.  This logical ID is an internal ID.

Can anybody tell me whether Unisphere /Navisphere CLI use MCx technology ?

I believe old Unisphere/Navisphere CLI was designed for Single process , that is why we have had communication errors  when multiple user used  Unisphere/Navisphere CLI  at the same time ... .

Would NG VNX got over above points ? if it is not yet ,,when will be ?

thanks !

Thank you Dan-san,

I didn't know there was a new logic to pick up a hotspare. I thought the logic had not been changed since CX products were released!
BTW, is it a new function of R33? Or even R31 and/or 32 (VNX) has the same logic with logical ID of each disk?
If even R31 and/or 32 has the logic, I have to revice my knowledge...

This is a new feature of MCx OE see support.emc.com/docu48786_VNX-MCx-Multicore-Everything.pdf P37

Thanks

Gearoid

I see. Thank you Gearoid-san.

MCx is exclusively a data path improvement however with that said MCx does monitor each core and will yield processor time to Navi.  MCx has mechanisms to detect and give a small percentage of the processor time to non data path tasks when under load.  Navi can run on any core, as best chosen by the operating system.  The issue that you mention is a restriction that Navi is single user not that it is multi-threaded or multi-core.  I will follow up with the Navi team to get the correct answer here.

I should also mention that there has been some enhancements to Navi that comes with MCx.  Improvements have been made to reduce the amount of time it takes to poll the system and collect inventory.  With MCx the system information is cached and asynchronous notifications are used to provide the updates.  The net result is less time is spent polling for changes in the system topology.

The next generation VNX has a new control station with more CPU and memory compared to the previous generation