It seems that every time I configure an EqualLogic iSCSI SAN array in the company of folks who've never seen the process, they ask the same question: "Really? You're already done?" The answer, always, is Yes.
The simplicity of EqualLogic's management platform extends from the CLI to the GUI interface, and is also exhibited physically with the simple chassis and controller design. Form certainly follows function, but the PS3800XV isn't lacking in either. Suffice it to say, I really like this array.
Scalability and performance
There are many ways to build and scale an iSCSI SAN, but the general rule is that as you add capacity, you also add bandwidth. With EqualLogic, this is part and parcel of the overall solution: You can't add capacity without adding bandwidth.
The PS3800XV sports 3Gb NICs per controller, for a total of six interfaces. The controllers function as active/passive partners, with three of these NICs waiting in the wings to take over and the other three as independent entities, each with a unique IP and MAC (media access control) address that's shared with its passive counterpart. There's no EtherChannel, bonding, or other port aggregation methods in use, since load balancing is handled with native iSCSI referral commands.
This setup performed perfectly in the lab with a selection of iSCSI software initiators. Theoretical throughput is in the 300MBps range; in practice, I was able to get close to 300MBps with three clients, each pushing 100MBps reads from the array. Even more interesting is the fact that I was able consistently to reach 100MBps throughput on CIFS filesharing tests using an HP DL360 as a fileserver with an iSCSI LUN (logical unit number) mounted via the Microsoft iSCSI software initiator, and 12 simulated clients running from nbench. That's better performance than many native CIFS filers are able to produce. In fact, it's better than most fileservers writing to local disk.
Part of the reason for the PS3800XV's high performance is its SAS (serial attached SCSI) drives. My evaluation unit came with 16 147GB Maxtor 15K RPM SAS drives, providing a raw capacity of 2.3TB (usable capacity is 1.53TB with RAID50). The rest of the high performance comes from the enormous 2GB battery-backed cache integrated into each controller.
These batteries are designed to provide power to the cache for more than 72 hours with a full charge, so worries about lost writes are minimal. Also, with RAM caching of this size, it's nearly impossible to eclipse the cache space during normal operations. The end result is screaming throughput.
I ran a slew of performance tests on this array, including the aforementioned nbench tests. I also ran my not-really-patented four-day stress test, which involves creating a 1TB LUN (logical unit number), mounting that LUN on a Linux server and writing 1TB in 2GB files from /dev/zero, deleting all files, then starting again. Rinse and repeat for four days, and see what breaks. In this case, nothing broke, and the average write speed was 95MBps across all iterations, with total data transfers well over 10TB.