Virtual tape libraries are on the crest of an adoption wave that is steadily gathering pace. The rush to develop disk-to-disk backup and the pushing back of tape to further away from the front line in the data protection cycle is prompting some of the consolidation in the tape automation market, such as Quantum buying ADIC. Network Appliance entered the data protection market with virtual tape products in February this year and has just announce three new virtual tape libraries.

VTLs have two main markets, according to Krish Padmanabhan, NetApp's general manager for heterogeneous data protection. Departmental-level users mostly want to replace tape while enterprise users are more intent on speeding backup performance.

The three new VTLs

These products present themselves to backup software products as tape libraries but are in fact totally disk-based and so offer disk-class backup and restoration speeds. They can automatically copy data directly to physical tape under backup application control to protect data for any open systems storage environment, including those that use EMC, Hitachi, and IBM primary storage.

NetApp's new VTLs are:-

- NearStore VTL300, an entry-level system with a capacity of 53TB and a sustained write speed of 500MB/sec or 1.8TB/hour.
- NearStore VTL700 with a capacity of 128TB and a sustained write speed of 850MB/sec or 3.06TB/hour.
- NearStore VTL1400, a dual-head system with a capacity of 256TB and a sustained write speed of 1,700MB/sec (6.1TB/hour).

Users of the first two models can upgrade to higher models. They all run a Linux-based NearStore VTL OS v5.0 - not Data ONTAP.

The performance and capacity increases are significant. Back in February the earlier models, without hardware compression, looked like this:-

- VTL600 with one controller and scaling from 4.5TB to 84TB in capacity. Its write through speed is 1.8TB/hour.
- The two-controller VTL1200 can hold up to 168TB of data and its write through speed is 3.6TB/hour.

With these earlier NetApp VTLs if compression was turned on then the host server would deliver as little as half the stated write through performance. That meant 0.6TB/hour per controller.

NetApp claims that the new models increase VTL write performance by more than 50 percent over these preceding models when compressed data is written.

Compression

Compression has always been a feature of backup to tape and typically has used server CPU cycles to achieve its generally 2:1 compression ratio. The actual ratio depends upon the types of files being compressed. Padmanabhan said JPEG files typically don't compress at all, while Microsoft Office documents can be compressed to a tenth of their size.

By carrying out the compression in embedded hardware these new NearStore products offload compression from the server running the backup software. NetApp stated these 'systems provide customers with the ability to substantially increase backup and recovery service levels by doubling to tripling the amount of backup data that can be stored on disk ... When data is compressed at the industry-accepted standard of 2:1, customers achieve 50 percent savings. When data is compressed at 3:1, customers achieve 67percent savings at even higher performance levels.'

If you double the amount of data that can be stored in a VTL then the cost per TB is halved.

Performance

NetApp says these new VTLs outperform the competition. Patrick Rogers, NetApp's VP of products and alliances, said: “Before today’s announcement, customers had to choose between reduced performance with software compression, or opting for no software compression and spending more for disk-to-disk storage capacity.”

For example, EMC's Clariion Disk Library, v2.0, will see its write performance drop by up to half if compression is turned on. The Red Book for IBM's VTS (IBM Virtualisation Engine TS7510) states 'Backup compression decreases the performance speed of the virtualisation engine. For this reason we recommend that you do not use the Backup compression option.'

NetApp commissioned independent tester Veritest to compare its new VTL600 with EMC's Clariion DL710. Both test systems used 320GB SATA drives and Fibre Channel connection to a server.

Veritest conclusions

The verbatim test conclusions were:-

1. In our test configurations, we found that the Network Appliance NearStore VTL600 best case throughput was 512 MB/sec, which is 102 percent of the published throughput rate stated on the Network Appliance website.

2. In our test configurations, we found that the EMC CLARiiON DL710 best case throughput was 517 MB/sec, which is 86 percent of the published throughput rate of 600 MB/sec stated on the EMC website.

3. The Network Appliance NearStore VTL600 was able to self tune when the number of disk trays increased from one to four while the data stream count increased from ten to forty, resulting in a 240 percent increase in throughput, from 205 MB/sec to 492 MB/sec.

4. The CLARiiON DL710 throughput increased by 150 percent in the same incremented test bed as (3) above, from 205 MB/sec to 311 MB/sec.

5. The Network Appliance NearStore VTL600 was able to consistently perform at or above 95 percent of its published throughput rate in twenty separate test runs in which backup data was randomized to simulate the variability of real world workloads.

6. The CLARiiON DL710 performance ranged from 50 to 80 percent of its published throughput rate in twenty separate test runs in which backup data was randomized to simulate the variability of real world workloads.

7. In our tests, we found that the Network Appliance NearStore VTL600 was able to automatically adjust the size of a virtual tape in order to optimize the utilization of physical tape media in a SAN attached physical tape library.

8. In our tests, we found that the EMC CLARiiON DL710 did not automatically adjust virtual tapes to optimize the utilization of physical tape media in a SAN attached physical tape library which resulted in twice the number of physical tapes being used.

9. During our rebuild test of a partially filled LUN, the NearStore VTL600 took 40 minutes and 42 seconds to rebuild the RAID group.

10. The same rebuild test of a partially filled LUN using the EMC CLARiiON DL710 19 hours and 23 minutes to rebuild the RAID group.

11. A test to measure the rebuild time for a RAID group fully filled with data on the NearStore VTL600 took 4 hours and 30 minutes.

12. The NearStore VTL600 was able to automatically switch a workload to an alternative RAID group when the initial RAID group was in degraded mode, providing 97 percent of the throughput measured before the simulated disk failure.

13. In the same test on the EMC CLARiiON DL710, all write activity remained directed to the initial RAID group that was in degraded mode due to the rebuild operation, providing 86 percent of the throughput measured before the simulated disk failure.

Deduplication

Deduplication can deliver greatly increased effective capacity with dedupe ratios of 20:1 being achievable. However the write speed of a deduping array can be quite low. For example, Diligent and Data Domain deduping VTLs can run at under 100MB/sec. The VTL700 can run at 850MB/sec. A backup session taking 1 hour with the VTL700 could take more than 8 hours with a Data Domain product and thus severely stretch backup window time.

Padmanabhan said NetApp was working on developing de-duplication technology that did not reduce backup performance.

Price and availability

The NearStore VTL300 starts at $99,000 with one disk shelf and 10TB of storage. The NearStore VTL700 starts at $154,000 with one disk shelf and 10TB of storage. The NearStore VTL1400, which has a dual head, starts at $238,000 with 20TB of storage. All three are available now.

Existing NearStore VTL models will continue to be sold. In addition, NetApp offers an in-place upgrade to the new models.