STORAGE SWITZERLAND HOW TO GET 2TB MORE STORAGE IN EVERY 2U SERVER Eric Slack, Senior Analyst Cloud storage infrastructures today are often made up of a grid of modules or nodes, each with storage and compute scenario, adding 100TB of storage would require 10U of additional rack space. For some cloud storage companies power in a single chassis. This ‘scale-out’ architecture can finding this much room in the data center could be a enable data centers to better meet capacity and performance needs as these cloud environments expand, challenge. compared with more monolithic ‘scale-up’ storage systems implemented in a traditional SAN environment. At a recent storage industry event, the CIO from a major corporation was quoted as saying essentially: “the next TB This type of architecture can also be implemented with of storage I buy will cost me $1 Million”. He was of course commodity or off-the-shelf hardware, instead of purpose built solutions, which helps to keep costs down. referring to the hypothetical cost of a data center expansion, or building a new data center altogether (which When these scale-out storage nodes (often server would probably cost much more than that). While it’s hard to imagine that there was really no room for a few more hardware with internal disk drives) are filled to capacity, server-nodes or hard drives in his environment, the whether implemented that way or after hard disk drives (HDDs) have been added, then expansion requires more message is clear. At some point the out-of-pocket cost of adding even a relatively small amount of capacity has to nodes. What this means is having to find more rack space in the data center, just to add more storage capacity. include the expansion of the facility as well as the storage system. These scale-out server-nodes are typically configured with one smaller, faster boot drive, and the rest larger, What if instead of trying to shoehorn another five or ten storage server-nodes into the data center, the IT manager ‘capacity’ drives. As an example, assume a 2U node could provide 20TB of raw capacity (10 x 2TB drives, in a 12- was able to simply add a few TB to each existing node? The question would then become: “How does one add drive chassis with one boot drive and a hot spare). In this more drives to already full scale-out storage nodes?” [1] The SSD DIMM order to meet this range of configurations with fewer hardware models. If a user needs more memory, they can One way is to use a solid state drive on a DIMM module, order the server configured that way. If their requirements like Viking’s SATADIMM. These SSDs plug into any empty DRAM DIMM socket and connect via a cable to a SATA aren’t on the high end, they simply leave these DRAM DIMM slots and even CPU sockets empty. In fact, servers header on the motherboard or RAID card HBA. With this simple implementation an empty DIMM socket can be with DIMM sockets less than 50% full are not uncommon. This would mean that many servers have 8-12 empty viewed by the data center manager as an empty disk drive DIMM slots, representing an equivalent of 8-12 empty drive bay. Power is supplied by the DIMM socket and the I/O is connected just as an HDD or SSD would be, via a SATA/ bays per server. One can confirm this by checking their current inventory of servers and see how many empty SAS cable. In this use case one SSD DIMM would replace DIMM sockets they have. an existing boot drive. Then the current boot drive could be removed to make room for a high capacity SATA HDD, and Putting an SSD into existing storage nodes can bring some with it, another 2TB (or more) of available storage. dramatic performance improvements too, such as speeding up the page swapping process that used to rely Adding 100 TB to the data center by putting a 2TB drive into each of 50 existing storage server-nodes might just on a slow, hard disk boot drive, as is described in another Storage Switzerland article. In other cases, where more meet the capacity expansion requirements that the IT capacity is needed, more DIMM sockets may also be manager is facing. If that 100 TB was the tipping point the CIO above was referring to then this strategy could available for additional SSDs, again, without having to make the tradeoff between using up a hard drive bay for postpone the capital expenditure (CAPEX) of a major facilities project. Adding a solid state drive to existing that performance boost. scale-out storage server-nodes can bring some other In scale-out storage architectures, SSDs may further benefits as well. But are there really empty DIMM slots in a typical commodity server chassis? improve storage densities as well. These systems don’t have a dedicated controller module, and instead spread this storage processing workload out across each node. When per-node performance is increased, as with the Empty DIMM slots addition of SSDs, the aggregate performance required may Server manufacturers build their systems for maximum be provided with fewer nodes, further improving the rackspace problem described above. flexibility - flexibility for them as well as their customers. There are a number of workload scenarios these server vendors design around, some with higher memory capacity A ‘doable’ solution requirements, such as virtualization hosts or hypervisors. Others require less memory, a good example of which is Optimizing one’s existing infrastructure is a sound policy; the use case described above, where commodity server products are used as scale-out storage nodes. But they’re using what you have before buying more is a responsible practice. Also, incremental solutions, such as adding an used in traditional application environments with moderate SSD on a DIMM module are often easier to get approved memory requirements as well. since it can be implemented in a piecemeal fashion, when the funds are available. It can also be less disruptive to In response to these varying scenarios, vendors build upgrade one node at a time rather than rolling in several motherboards to accommodate large amounts of DRAM, in new nodes at once. [2] Summary A good alternative may be to add an SSD DIMM, to replace the boot drive in each node and free up a drive bay for a Scale-out architectures are excellent solutions for high capacity disk drive. For cloud storage data centers expanding the capacity of cloud storage infrastructures. As complete nodes that include storage, network connectivity this strategy can increase available storage by tens of TBs without consuming any additional rack space. For and processing power, they allow capacity to grow almost without limit and performance to be tailored to that growth. companies getting close to capacity in the data center, putting an SSD DIMM, like Viking’s SATADIMM, in each They also support the use of commodity hardware to keep server may provide enough added capacity to postpone a costs down. data center expansion project by perhaps as much as a year. But scale-out storage expansion typically means adding more servers-nodes, not just more drives, which means a larger purchase. Eventually, more rack space can create a need for more data center floor space and, in a worst case scenario, adding a relatively small amount of capacity could end up triggering a much larger capital expansion project. About Storage Switzerland Storage Switzerland is an analyst firm focused on the virtualization and storage marketplaces. For more information please visit our web site: http://www.storage-switzerland.com Copyright © 2012 Storage Switzerland, Inc. - All rights reserved [3]