christianZ
Champion
Champion

Warning: Comparison Equallogic, NetApp, EMC

If anybody see that:

http://www.sct.dk/documents/00089.pdf

It is not quite true.

NetApp's 9000 IOPS were reached with one head (3020) and 28 disks -

the hole sysem (2 heads 112 disks reached ~ 15000 iops)

EMC's 1700 IOPS were reached with one sp and 8 disks -

the hole sysem reached ~14000 iops.

All those could be red on veritest site:

http://www.lionbridge.com/lionbridge/en-US/services/outsourced-testing/competitive-analysis/Netapp.h...

Just my correction - there is no magic here !

One should remember:

\- Equallogic is working with active/standby controllers but they can cluster 2,3,4,... systems together so that one lun will be served over all controllers

\- the other both with active/active (NetApp - 2 heads (clustered - it is possible to buy only one head - rational ?? EMC - one head with 2 SPs)

i.e. one lun can be served only per one controller but the perfomance of the hole system could be higher, when enough disks are used and both controllers are configured properly.

0 Kudos
18 Replies
christianZ
Champion
Champion

In addition this statistic was showed in this session:

http://download3.vmware.com/vmworld/2006/tac9884.pdf

0 Kudos
epping
Expert
Expert

it has to be remembered though that currently esx can only access one path at a time so having a multiple front end ports and active/active currently has little performance benefits.

0 Kudos
christianZ
Champion
Champion

...for one ESX host, but when you have 2 or more...

0 Kudos
BUGCHK
Commander
Commander

esx can only access one path at a time

It is my understanding that this limitation is per disk, not per entire ESX server or did I misunderstand you?

0 Kudos
gogogo5
Hot Shot
Hot Shot

Please can someone clarify something for me. Based on ChristianZ comment on EqualLogic working in active/standby mode, when looking at the Storage / SAN Compatbility Guide PDF it shows in the SAN Array Model Reference table:

EqualLogic --> PS Series --> Active/Active --> Fixed

So can someone please explain the Active/Active in the table versus the Active/Standby comment in the thread. Which one is it?

0 Kudos
admin
Immortal
Immortal

I'm pretty sure the Equallogic boxes are supposed to be using in Active/Passive for ESX, well that's what the Equallogic guys told us.

On that note we've been looking at one of their boxes today, gotta say I'm impressed!

0 Kudos
gogogo5
Hot Shot
Hot Shot

Yeah, I had a demo last week and was very impressed too. Adding arrays literally doubles, triples etc your throughput since each unit has control modules. As opposed to have a control module with multiple disk shelves all connected to your SPs.

But I would really like to get to the bottom of the correct setup for the PS series. Does this look like a documentation error in VMware guide?

Anyone got a PS series configured in production? What's your settings and path policy?

0 Kudos
christianZ
Champion
Champion

My configuration shows "fixed" althought they are working als active/standby.

I have tested the controller failover many times and it was always working w/o problems.

Maybe wohharow could correct it - when he sees that (would be nice).

For performance values check this:

http://www.vmware.com/community/thread.jspa?threadID=73745

0 Kudos
gogogo5
Hot Shot
Hot Shot

ok, so only one control module is active. Would I be correct in saying that both HBAs in the ESX host would be actively serving I/O but ultimately arrive at the one active control module (which could have 1,2 or 3 network connections)?

0 Kudos
femialpha
Enthusiast
Enthusiast

I have 2 PS300's and they are indeed active/passive. Path policy is fixed as equallogic will handle the redirection/loadbalancing on its on end. They use trunking so the client only sees one target.

0 Kudos
woharrow
Enthusiast
Enthusiast

me again.... I think there is some confusion in the thread about what VMWare is describing as Act/Act. My impression is they are talking about using HBAs on the ESX host side in an Act/Act config. Since ESX shouldn't be able to tell if an array is running Act/Act or Act/Pass at the controller level. (if the array is built correctly.)

My conclusion is from knowing that I can run Multiple Qlogic 405x cards in a server all Actively working to connect to volumes on an EQLX array, while the arrays are Act/Pass on the controller side. (I have this running in my lab)

**I'll take this time to remind people not to confuse availability with uptime, if you are running Act/Act at the controller level and lose a controller then you get a 50% degridation in performance with a failure. While in an Act/Pass controller setup you not only stay online, but at 100% of performance. **

I hope that I've cleared up some of the confusion and not added to it. I'm also glad to see everyone in the thread enjoyed their demo!

-also- Thanks Femi !

0 Kudos
christianZ
Champion
Champion

Wade,

thanks for this clearing.

0 Kudos
BUGCHK
Commander
Commander

EqualLogic --> PS Series --> Active/Active --> Fixed

It is my understanding that the second controller is really 'passive' - it does not present any LUNs while being passive. After failover, it takes over the IP addresses of the former active controller - otherwise you'd deal with 7 IP addesses in a single-member group, no?

An Active/Passive Fibre Channel array works different. The 'passive' controller does present the same set of LUNs, but it answers with a "NOT READY" response to I/O requests. It must present the LUNs, so that the multipath solution on the host 'knows' that there are standby paths it can failover to.

In the EqualLogic world, the host is not aware that there is a second controller and Microsoft's MPIO can talk to multiple ports of the same[/b] controller whose ports are all active.

0 Kudos
BUGCHK
Commander
Commander

Since ESX shouldn't be able to tell if an array is running Act/Act or

Act/Pass at the controller level.

A/A or A/P refers to the behaviour of the storage array.

ESX MUST[/b] know the difference. Doing boot it probes all SCSI LUNs to find out if it is working on an active or a standby path.

0 Kudos
gogogo5
Hot Shot
Hot Shot

But depending on the storage used it also dictates whether the HBA cards in an ESX host are both serving I/O or just one of them.

We use ESX with an EMC fibre SAN and only one HBA is serving I/O, then if the card fails the second card will take over. It is my understanding from this thread that when using ESX with EqualLogic storage both HBAs can be configured to spread the I/O between them, and that this I/O will arrive at the single active controller.

Would someone confirm or refute my statement just I know for myself. Thanks.

0 Kudos
BUGCHK
Commander
Commander

I am not familiar with your EMC box - maybe this is a low-end FC array like an HP MSA where each controller has only one host port and a strict A/P controller software.

On a midrange array, where a controller has multiple host ports, they go to different fabrics. In that case, you can access different disks over different Fibre Channel adapters.

iSCSI is a special case if you use the ESX software initiator, because it currently supports only one path. On the other hand, it is my understanding that you can access different disks through different iSCSI HBAs to the same EqualLogic PS controller which has 3 active ports.

Should this controller fail, the other one will take up its identity (IP addresses). This should be mostly transparent for ESX (maybe and I/O retry) and ESX does not initiate a failover like in the FC world, because there are no standby paths here.

0 Kudos
gogogo5
Hot Shot
Hot Shot

We dont't have a low end FC array - its and EMC CX700.

I cannot force I/O paths through vmhba2 - only vmhba1 is able to carry I/O for all LUNs.

0 Kudos
christianZ
Champion
Champion

It is correct - with EQL and 2 iscsi hbas you can use both hbas to access the luns.

0 Kudos