30 Minute Interview

SSDs, Unified Storage, ILM...looking beyond the buzzwords

Shailesh Agarwal, Country Manager – Storage, IBM Systems & Technology Group, IBM Global Services India Pvt Ltd talks to Prashant L Rao about trends in enterprise storage

Shailesh Agarwal

There’s plenty of buzz around Solid state disks (SSD). Is this technology going to be used in enterprise storage arrays?

SSD is being used for caching. Large Internet companies use it to cache frequently accessed information. The technology remains prohibitively expensive in terabyte (TB) capacities. As far as reliability goes, it hasn’t been tested in a commercial environment. That said the technology is moving very fast. The biggest benefit of SSD is lower power consumption and space utilisation.

Who’s the winner in the battle between iSCSI, fibre channel (FC) and NAS?

Unification is the trend here with a choice of protocols on the same box. Why should the attach protocol be hard linked to the storage media, after all. All disks are capable of relocating. This is easily done using a router. We offer the N series (under the IBM NetApp alliance) unified storage boxes that support native SAN, iSCSI and NAS. Every company has workloads that are tuned to each of these protocols. For file sharing, NAS is best. Databases work best with a FC SAN. For e-mail its iSCSI. Unified storage supports all three simultaneously. You can dynamically allocate [disks to each of these protocols]. SMBs can’t afford to do only NAS or SAN. Even a low-end box lets you change or reallocate. The way this works is that there’s a multi protocol router between the fibre channel storage box and the user. It handles Ethernet, iSCSI or FC requests.

Networking companies are talking about wide-area file systems. What are your thoughts on that?

Unified storage has caching. That’s one aspect of a wide-area file system. In the US, LAN and WAN bandwidth is ubiquitous. In this scenario, where files are commonly shared you cache them. If you have two teams working on projects in Chennai and Pune, you can put them on a shared box with a cache. We’ve talked about WAFS to our customers but there’s not a lot of excitement.

There seems to be a trend of storage boxes running on server processors. Is this the case with IBM as well?

Our enterprise boxes run on the Power 5+. The mid-range boxes use ASICs. The appliances are based on Xeon processors and not counted as part of our storage line-up. These run Windows Storage Server.

We leveraged the Power5 and built storage around the processor. In 2001 we had the ESS F20 built around the Power3 and 3+. In 2002 the ESS 800 was launched, it was built around the Power4. In 2005 the DS8000 debuted built around the Power5 followed by the DS8000 Turbo that uses the Power5+.

We keep thinking about how we can harness the processing power of the Power5/5+. As of now the processors aren’t fully utilised. Over time we want to put some applications that are linked to storage in the storage box. Right now there’s replication and FlashCopy. Can you put backup and ILM in the box? For these are data based applications. To do so would make them server independent. The results of this strategy have been phenomenal in terms of SPC-1, Cache I/O and SAP benchmarks.

What’s your SMB focus?

70 percent of our sales are to SMBs. We have the N Series of products for this segment. While every company has block, file and iSCSI requirements, in a large enterprise a separate storage infrastructure exists for each of these protocols. SMBs, on the other hand, may not be in a position to pick and choose between SAN and NAS. All N series boxes come with fibre, iSCSI and NAS (CIFS, NFS). It’s a single box that satisfies the needs of an SMB with multiple workloads. It’s a software-based platform and can therefore be upgraded without changing the hardware (for SAN you may need to change the box). Compliance, DR and ILM are functions of the software. This is application-aware storage. Most storage boxes only differentiate between block and file. Here you have agents to recognise the application to which the data belongs starting with Microsoft Exchange and SQL Server. The agent should be able to recognise an e-mail from Exchange. Backup and restore are faster in this case as the process is granular. Snapshotting is undertaken at a suitable frequency. If you realised that your Exchange box has been the victim of a virus attack, you can rollback in a second. You will still lose some data but you’re up and running quickly with most of it.

x86 storage is an emerging category. will this result in storage becoming a commodity as is the case with entry-level and to some extent mid-range servers?

In the SMB segment or at the low-end you won’t get dedicated x86 storage. Servers are an example of replacement infrastructure; storage happens to be cumulative. Affordability is important here. We have a whole set of products for the x86 environment, the DS 300/400 etc. Pluggability is more important here than scalability. Start with two or four TB and plug in more boxes as you grow and keep things simple. If there’s a problem just ship the appliance back. The cost goes down as cost is a function of configurability.

When you move up a level there’s the DS 4000 range that starts with SATA. Today a SATA box has 7200 RPM drives and a 4 Gbps fabric. Earlier SATA boxes came with a one year warranty. Now even the entry-level SATA box comes with a three year warranty. Moreover, these are all plug-and-play.

Are companies adopting a new direction when it comes to DR?

Predictability, orchestration, monitoring are the watch words—can I know what my RTO/RPO is in real time? We are working with Sanovi on this front. Unified DR—one DR across protocols—is a concept that is gaining ground.

Business continuity has entered the mainstream of enterprise computing. I don’t see any deal without BC being part of the sale.

We’ve heard a lot about ILM. What’s the latest on that front?

ILM is where BC was a year back. Everybody’s talking about it, there are some pilots. It hasn’t reached the inflection point yet. One approach to ILM is to let a consultant categorise all your data application-wise. In the short-term this will be the way.

E-mail is one of the first data stores that are compliance-related. In the case of ERP or core banking systems the data generation isn’t quite as rapid so the ‘pain’ hasn’t been felt yet. Most ERP or database applications come with support for selective archiving. Storage vendors provide cheap storage that works well with the application. We have the DR 550 (large cheap storage, software, tape library). E-mail is almost like a corporate virus. A 1.5 MB message going from person to person soon becomes 5 GB. One way out is to establish quotas forcing users to either delete large messages or to store them locally. Unfortunately under SOX or any similar local legislation, the responsibility of storing e-mail data lies with the company and not the individual. This is a requirement for e-mail that goes outside the company to customers, press, analysts etc. Companies have to archive all external communications that have material importance which leads to the need for e-mail archival.

20 percent of e-mail contains attachments and uses 80 percent of available space. There’s a huge amount of duplication. De-duplication lets you store 1 GB of messages in 10 percent of that. The value proposition is from the compliance and cost perspective. The technology exists in the N Series. With software called Snaplock. You can’t delete or modify. You have an expiry mechanism that lets you lock data for seven years and then delete it. You can’t delete or modify the data during that period. As of now this stuff is not there in SAN arrays. WORM tape can be used to archive SAN data. Hospitals use DVD media for archival.

There’s a big consulting service around ILM. Right now it’s application-specific, e-mail to start with and then flat files and finally databases.

Is multimedia storage a big area for you?

There are many segments here, each with its own characteristics. In animation you have DQ Entertainment and Crest. Most file systems are not designed to handle large files. We have the General Purpose File System (GPFS) that introduces a concept called bitlocking. Three people can simultaneously edit different portions of the same file.

TV Channels are going in for Digital Asset Management (DAM). They need to repurpose content. We work with Ardendo and DataForge. Here the final archival is onto tape. All our tape products are fully certified for DAM.

Studios make FX laden movies. We do not play a big role here right now. They use special equipment with its own storage. As they hand over the IPR to their clients, they tend to use cheap storage. A typical FX-heavy movie needs about 20 TB of storage. In this case the movie makers would use 20 1 TB USB boxes. The companies that sell the editing equipment also sell the storage.

In the case of Web portals, telcos and the Net in general these are interactive mediums. Video on demand (VOD) from VSNL, IPTV, DTH and delivery through IP are the trends.

Digital Surveillance (DS) is an emerging area. With analogue tape it takes ages to find what you want. Airports are large consumers of DS. Refineries, banks, ATMs, large manufacturing plants... DS will become the norm in all these segments.

The first two—animation and TV—are active. In the case of studios, we don’t play much and VOD and DS are upcoming segments.

With the cost per megabyte plummeting would it be correct to say that storage is effectively free?

All TB are not equal. In a PC you have a 5400 rpm disk that may not even be SATA. It may cost a cent per MB. For a 15000 rpm FC 150 GB disk the cost per MB is double that. A storage customer isn’t buying raw TB. In an enterprise storage system, the HDD is but one component. For a 50 percent populated system, the HDD cost isn’t more than 25 percent. The interconnect, cache and IPR account for the rest. In an enterprise system that can take 1,000 disks of 300 GB each you can have a maximum of 300 TB. If instead of a thousand 300 GB disks if you have a thousand 72 GB disks, the difference in cost is negligible. The difference in cost of the system between these two configurations is about 10 to 12 percent but the cost per TB has gone up 4x. Fully populated with the maximum disk capacity, enterprise systems cost about $20,000 per TB.