With the shared nothing approach (more correctly called the shared some approach), LUNs are created in smaller RAID groups throughout the SAN. Small numbers of physical disks are created in RAID groups of various configurations and sizes and the LUNs are placed within these RAID groups. The good thing about this configuration is that if a specific LUN or disk starts to cause a slowdown, the only LUNs that are affected are the ones within that LUNs RAID group. The downside to this system is that it requires more hands-on configuration and more knowledge of the I/O levels that you need to support. The upside is that if a LUN needs more I/O, you can move it to another RAID group or even have it span more than one RAID group to give it more I/O throughput.
SAN vs. DAS
Today, the SAN is king in the world of storage. The SAN is very fast and usually made up of hundreds of fiber channel drives, all of which are connected to the server via high-end fiber cables. However, SAN solutions are very expensive, often costing hundreds of thousands of dollars just to get started. This brings us to DAS. DAS is a great solution if you’re company doesn’t need a large-scale SAN solution. DAS offers you pretty much all the same capabilities as a SAN, just at a smaller level, and usually with less cache. DAS devices are made up of anywhere from 2 to 45 SCSI, SAS, or SATA drives, with anywhere from a few megabytes of cache to a gigabyte of cache.
All the same configuration options are usually available with DAS as with SAN, including read cache and write cache balancing. Just like SAN drives, DAS drives still need to be aligned (as do all RAID arrays). Unlike a SAN, a DAS usually won’t let you control cache settings on a disk-by-disk level. Also, DAS systems are connected to a specific server, so you can’t have more than one server connected to a DAS, as you can with a SAN.
The biggest drawbacks to using DAS for your databases are the lack of growth potential, because there’s a physical limit on the number of disks you can put into a DAS, and the fact that you can’t pool your resources between multiple servers with DAS devices. Another problem you might run into with DAS devices is that you can’t spread a single RAID array beyond the number of disks in a single drive shelf. Most DAS drive shelves hold 15 drives, and each drive shelf is connected to a single RAID controller. Because the RAID controllers don’t talk to each other, you can’t RAID between drive shelves. However, these drawbacks are offset by the very attractive entry cost. DAS storage is inexpensive when compared to the costs of purchasing a SAN solution.
iSCSI and the Database
A newer storage technology that’s just breaking into the database world is iSCSI. iSCSI uses your existing Ethernet to connect the database to the storage. Many SAN vendors now include iSCSI as part of SAN systems. The upside to iSCSI is that you get SAN quality storage without the additional cost of the fiber channel equipment. The downside to iSCSI is that all your storage traffic now has to compete for bandwidth with your regular Ethernet traffic. If you have a slow or extremely busy Ethernet network, iSCSI probably isn’t the solution for you. Another potential problem with iSCSI is that because your storage traffic is going over the TCP network, you have to deal with the TCP timeout settings. If your storage array fails for some reason, fiber channel will time out very quickly and retry along another path. However, the timeout for TCP packets is much higher, which will be represented within your database as blocking or processes timing out.
To address network bandwidth issues, network engineers use a technique call VLANing to separate and isolate network traffic. VLANing ensures that traffic on your iSCSI network is isolated from other parts of your network. This isolation is a logical separation of network segments into Virtual LANs (VLANs) at layer 2 of the Open Systems Interconnection (OSI) Reference Model. A VLAN is a logical domain within the network switch that lets only traffic that has been tagged for that VLAN to access that network switch. Traffic within a VLAN, including broadcast traffic, is kept isolated within that VLAN. Any traffic destined for another VLAN will have to be routed. This helps prevent non-iSCSI network traffic from affecting your iSCSI traffic and helps prevent your iSCSI traffic from affecting other non-iSCSI traffic. VLANing works at layer 2 and separates traffic only at the layer. To further isolate traffic for applications that require layer 3 switching or routing, VLANs can be associated with a routing subnet that will let traffic cross logical domains (VLANs) to communicate with other VLANs. Furthermore, layer 3 routing can also be configured with an ACL to allow or prevent specific traffic within specific subnets from communicating with each other. Other techniques at layer 4 and layer 5 are also available, but that’s beyond the scope of this article.
Get the Best Performance Out of Your SQL Server
Properly setting up and configuring the physical disks is essential to getting the best possible performance out of your server. This is especially true with SQL Server (or any database server) because database servers read and write data to and from the disk in much higher quantities and at much higher speeds than other servers in the enterprise. Therefore, you want to make sure that every issue is addressed when configuring your storage, especially when you consider how much time it takes to make these changes after implementation.