Let me start with a few lines on the importance of storage in an IT setup. If I ask you what is 'THE MOST' critical part in your organization, the answer would probably be 'INFORMATION'. This information resides as data in your storage, making storage the backbone of your business. Depending upon the type of organization, criticality of this data can vary. But one thing is clear, if you are running servers, you need databases and storage space where this data would reside. You also need the means to safeguard this data from corruption, plus you also need to take regular backup for safety and more importantly to adhere to your organization's compliance policy. An ideal storage solution for your organization would be one that can grow infinitely and that too without any downtime so that you don't have to invest upfront. It should also be simple to manage and setup. Finally it should come bundled with all features needed for regular maintenance. We reviewed Dell EquaLogic IP SAN, and found that it has all these features.
Conventional SAN setups are also known as frame based systems where a frame defines the boundaries of your storage.One can't go beyond maximum capacity of a frame and there is often a performance degradation as the capacity increases. This is because such frame based systems only have one set of storage controllers per frame (A set includes two controllers, out of which one is active and the other is for redundancy). There are a number of issues in this type of a setup like. For one, you need to define capacity in advance for future usage (which most of the time turns out to be incorrect). Even if it is correct, this approach is not optimized. You pay upfront for something that you'll use in the future, thereby ending up with a large storage capacity that lies unused and unallocated. Once you've consumed the unused capacity, and need to add more, then the only way to increase capacity is to replace your current frame with an even bigger frame (which can accommodate more). This activity requires downtime, which is not recommended for a business. These shortcomings of a framed based architecture can be overcome by using a frameless architecture. A frameless architecture adds an abstraction layer on top of your storage arrays, making it look as one pool of storage. This concept can be mapped to server virtualization wherein a virtualization layer (like hypervisor) sits on top of a pool of hardware resources (memory, CPU) and presents it to virtual machines running on top of it. Because of this abstraction layer, you can add new capacity in the form of new storage arrays without incurring any kind of downtime, which in turn means you can always start with lower capacity and grow according to your needs. The PS series of iSCSI storage arrays from Dell are based on a frameless architecture and can scale up to 16 physical storage arrays.
Setup plus few important concept
We received two PS600 Series iSCSI storage arrays--one with SATA drives and another with SAS drives. Plus, we received a Dell PowerConnect switch which is optimized for iSCSI traffic. There are different flavors of PS600 series with different type of drives like SATA, SAS, and SSD. You can also use a mixture of different drive types in the same array.
What EquiLogic boasts of is quick setup and we were not disappointed. We were able to set it up in minutes. Just connect PS6000 to a server (which could be Windows Server 2003 SP1/Server 2003 R2/Server 2008) using a standard Ethernet cable through the iSCSI optimized switch. Next we installed Host Integration tools and then run a 'Remote Setup Wizard'. This automatically discovered uninitialized PS series arrays. One word of caution here is to ensure that you use the right types of Gigabit Ethernet cables, else the discovery may not happen. Plus, it's also important that the Gbe switch that's used in between is optimized. Otherwise, the array may not give the performance (throughput and IOPs) that it's capable of delivering. We discovered both problems the hard way!
Once the storage array is discovered, you have to enter the member and group information. Here member is the individual array or physical box whereas group is the actual SAN that you would access. One more important concept in EquaLogic storage arrays is pool. You can create a pool of members to further control resources. a Pool can be understood as a SAN within a SAN. The PS series arrays allow creation of 4 pools within a group. Once you have assigned IP addresses to members and groups and defined RAID levels of the group, your array is ready for use. Connect your browser to the group IP address and enter the login credentials to manage your SAN. In order to use your SAN, simply create volumes and then assign them to machines (to user machine, volume would look like a local hard drive). The steps to do this are similar to connecting any SAN target to an initiator. Simply create the volume using PS Group Manager's management interface, then start the iSCSCI initiator on the local server that's connected to SAN. Discover the targets and connect to them. On the local server, open Disk Management (Computer Management>Storage) and assign a drive letter to the volume and you're done. The volume would appear as a local hard drive to the server.
To make sure that there is good connectivity from a server to the storage array, the PS series comes with a feature called multipathing. Each controller on the PS6000 storage box comes with 4 network ports that can be used independently for a fault tolerant connection from the array to the switch. Plus, they also do load balancing and multi pathing, to ensure that no single network card gets overloaded.
So if your server has two or more network cards, then you can team them up for multipathing and load balancing as well (provided they're kept in the same subnet). This is also easily done using the 'Remote setup wizard' that comes with the storage array. You can simply select 'Configure MPIO settings for the computer' from the first page, define the subnet for multipathing along with other configuration parameters like type of load balancing, etc.
For our test setup, we used round robin load balancing policy. To test this feature, we enabled all network cards on the storage array and used four network cards from the server. To see if multipathing is working properly simply start the iSCSCI Initator and click on 'Dell EquaLogic MPIO' tab. If a heavy transfer is happening, you can see utilization of all network cards. In our evaluation, we used this feature and discovered thta there's considerable performance gain if multipathing is used.
For taking snapshot of your data, you often have to pay extra, but with EquaLogic storage array every feature including this one comes bundled at no additional charges. Taking snapshots is as easy as right clicking a volume and selecting snapshot. But what is unique with these storage arrays is the ability to integrate with commonly used solutions like VMware, MS Exchange, etc. You also can modify snapshots and then apply them if needed. Auto snapshot manager is a solution that come in two flavors as mentioned above--one (Microsoft Edition) is for protecting and recovering Windows application data, including protection for systems using Windows NT File System (NTFS), Windows Server 2008 Hyper-V, Microsoft SQL Server databases, and Microsoft Exchange Server storage groups. In fact using this solution, you could recover a single mailbox of a user within a short period of time and that too without affecting any other user. On the other hand, EqualLogic Auto-Snapshot Manager/VMware Edition (ASM/VE) is a data management tool designed to simply create and centrally manage online point-in-time copies of VMware virtual machines and VMFS datastores using PS Series SAN-based snapshots.
We tested the PS6000 storage array using the IOMeter benchmark, which is commonly used for testing all kinds of storage devices, right from simple NAS boxes to high-end SANs. While there are many other benchmarks for testing file server and database performance, we stuck to IOmeter, simply because it provides the flexibility to measure I/O Operations per second, and let's you modify the payload as per your requirement. So we used this functionality to check how the IP SAN performs at different payloads. Plus, since we received both SAS and SATA based storage arrays, we decided to check the performance difference between the two, considering that the former is supposed to be faster than the latter. Lastly, we also tested the performance difference with and without multi-pathing. The results we got were fairly interesting.
Difference Made by Multi-pathing
Multi-pathing is a key feature of the Dell IP SAN, which is supposed to give a significant boost to its performance. To test it, we ran IOMeter with 64 MB of payload on both SATA and SAS RAID arrays, both with and without multi-pathing enabled. As can be seen from the results, both storage arrays give better performance with multi-pathing enabled. In the SATA array, the performance really shoots up to more than 100% with multi-pathing for random and sequential writes. The performance difference in the random and sequential reads is not as significant, but is higher nevertheless.
The performance difference in the SAS array however, is much more significant. There's a boost of more than 100% when multi-pathing is used to do three of the tests--sequential read/write, and random writes. The random read test however, seems to be the toughest, with the performance shooting up by 62% (low as compared to others, but high nevertheless). In the SATA array, it was worse, with only a 9% jump in performance.
Difference between SATA and SAS Arrays
We next tried to find out the performance difference between the SAS and SATA based storage arrays. Considering that the SAS based arrays are more expensive, we were curious to know whether the performance difference was significant enough to justify the additional cost. So, we ran IOMeter with 64 MB payload and multi-pathing enabled on both types of arrays.
We found that overall, there was a boost in performance, but it varied depending upon the type of operation you perform. Sequential read operations for instance, saw a 100%+ jump, while random reads saw 50%+ jump. The random and sequential writes however saw moderate jumps in performance.
Performance difference at different payloads
We tested the SAS storage array with multi-pathing at payloads of 64, 128 & 256 MB to understand where the highest performance comes from. We noticed that the array gave the highest throughput at 64 MB of payload, and it gradually dropped as we increased the payload. This is expected, because larger data would take longer to get read and written to the storage device.