There has been a lot of buzz around EMC Replication Enabler for Exchange (REE) which is our free plug in on the CLARiiON platform that allows for EMC’s implementation of Microsoft’s Third Party Replication API for Exchange Server 2010. This plug in allows for management of the replication solution when used with MirrorView/S and RecoverPoint. Basically, Third Party Replication API was built by the Exchange Product Group to allow Third Party vendors to ‘plug-in’ to the Database Availability Group (DAG) framework and instead of leveraging the native Continuous Replication within the DAG, leverage other forms of data replication such as array based replication.
See below for how this works:
As part of the upcoming Microsoft Exchange Tested Exchange Solutions (ETS) program, EMC/Microsoft/Brocade/Dell have partnered together to release the 2nd whitepaper as a part of this program. The solution validation was done in partnership of EMC, Microsoft, Brocade, and Dell and was built out at the Microsoft Enterprise Engineering Center (EEC) labs in Redmond, WA This solution is titled “Zero Data Loss Disaster Recovery for Exchange 2010: Enabled by EMC Unified Storage, EMC Replication Enabler for Exchange 2010, Brocade, Dell and Microsoft Hyper-V” – Yes the title is a mouthfull!
Highlights of the Solution:
- Exchange Server 2010 on Microsoft Windows Server 2008 R2 Hyper-V, 10,000 users per site (20k total), 500MB mailboxes, 150 msgs per day
- 2 Exchange Database Avaliability Group (DAG) design, utilizing 3rd party replication API enabled by EMC REE
- EMC Replication Enabler for Exchange with Mirrorview/S for synchronous replication between 2 sites
- EMC CX4-480’s Storage (2 per site)
- EMC Replication Manager w/EMC Snapview for app consistent replicas
- Brocade ServerIron ADX application load balancers (1 per site)
- Brocade 300 SAN switching and Dual Port 8GB FC HBA
- Brocade FASTIron Ethernet Switches (1 per site)
- Brocade MLX8 Core Router (1 per site)
- Dell PowerEdge R910 servers (4 Hyper-V hosts) – Quad-Eight Core Intel Xeon 7560, 192GB RAM, 4 X 1GB Ethernet ports
The physical diagram of the solution looks like this where we have 4 Hyper-V hosts across two data centers hosting a total of 20k users:
In addition to showcasing the storage and replication aspect with EMC REE and MS Third Party Replication API, we were also able to show some best practices in terms of guest building block and placement with regards to Exchange roles across Hyper-V hosts as seen below we have 2 DAGs across the Hyper V hosts with CAS/Hub roles along with DC/GC servers virtualized:
The solutions teams also performed some extensive performance validation in terms of both storage performance with JetStress and overall validation with LoadGen in both local failure and site failure scenarios. Remember, a solution not only needs to ‘work’ but it also needs to perform under load.
In short, if you are looking at an Exchange 2010 design and considering HA/DR and virtualizing the infrastructure, you won’t want to miss this very detailed whitepaper. Get it here:
As many of us know, there have been tremendous strides made in reducing IO in Exchange versions. Exchange 2010 represents another ~70% reduction or so from Exchange 2007. And nearly double from Exchange 2003.
So, it makes people wonder if new technology such as EFD even has a place in Exchange storage design. For the most part, EFD is costly and just doesn’t justify itself for primary storage for Exchange. We instead move towards lower cost SATA and SAS as long as the requirements of the deployment justify it, of which most these days do gravitate towards.
What has become interesting with some storage vendors is to utility EFD technology to ‘accelerate’ performance of the 7.2 SATA drives to near 10k FC/SAS performance in terms of IOPS, acting as a cache. That idea is not necessarily new by any means as storage vendors have been using memory as a cache to accelerate performance for some time now.
Because the EFD as a cache is a new idea, EMC decided to perform some testing as part of our FLARE 30 launch on EMC CLARIION to determine if/how Exchange IOPS could be improved by using our flavor of this technology called ‘FAST Cache” in Exchange Server 2010.
The results were higher than we initially expected in that in a 40k user run with Jetstress we produced ~42% more achieved database IOPS in a Jetstress run with FAST Cache enabled vs. no FAST Cache:
This was on our CLARIION CX4-960 platform, with 1TB of FAST Cache (10 X 200GB).
These findings were based on preliminary testing where we did not do any LoadGen simulations, but only Jetstress however this initial findings were better than expected. One should note that we don’t say that every customer needs this but for customers with higher IOP profiles and requirements for SATA it can be an acceptable option especially if you are consolidating multiple workloads onto the array.
More testing is planned to come up with better guidance, but in the mean time you can read the full findings in the following whitepaper:
Until next time,