I’ve been looking for an eSATA enclosure to put on an Atom board for a home NAS. My primary goal is to have 4TB of disk space that’s in a RAID. The secondary goal is to be cheap, which means low power. Assuming I keep this thing for four years, and that PG&E continues to charge $0.12/kwh, every extra watt of power consumption costs me a little over $4.
So, I’m looking for low power. This means an Atom processor and motherboard, which barely sip power. Unfortunately, none of them have both a reasonably recent Atom and more than two SATA ports onboard, so the goal was to use an eSATA enclosure to connect three 2TB drives in a RAID5. The first enclosure I came across was the AMS Venus ES5 (AMS DS-315SES), $170 on sale at Central Computers. I picked it up (after inquiring about restocking policy), along with three Hitachi 2TB drives (HITACHI 0F10311). I have been trying to get it to work with a PCI SATA board in my old desktop. That didn’t work: the SIL3512 chipset doesn’t support port mulitpliers. I borrowed another card from my friend Andy, a VIA chip this time. Also didn’t support port multipliers.
Andy was then kind enough to meet me at a Peet’s in downtown SF, where we plugged it in to his Dell laptop, with an Intel GS45 (ICH9M) chipset. This worked like a charm, and we proceeded to do some tests on the array. At a coffee shop. It drew surprisingly little attention from passersby: this is San Francisco, after all.
On the positive side, it just worked with linux. It came up a little curiously the first time, but that’s typical for older RAID enclosures (scanning disk by disk). Hotswap also seemed to mostly work, though it did cause the enclosure to reset its entire link every time you unplugged or replugged a drive. This makes hotswap a lot less useful, but it is still, technically, hotswap.
We tried reads and writes of 1GB with ‘dd’, varying the number of simultaneous spindles. We also tried buffered and unbuffered modes for these operations, to try and get raw performance of the array’s drives and chipset.
For writes, the enclosure does okay: it looks like it scales out nicely, though doesn’t nearly saturate a 3Gbps link. At peak, we got just under 210MBps, or 1.68Gbps, or 56% of the theoretical peak. More typical would be 1.2Gbps, or 40% of peak.
Reads were far more disappointing: buffered reading off three drives gave us 76MBps, or 20% of the theoretical peak. Note that a single drive reads at 130MBps, so it’s apparently just bus contention slowing this down. Since this is roughly my use case, I decided that I wasn’t keen on using this enclosure.
| Total bandwidth | 1 | 2 | 3 | |
| Read | Buffered | 130 | 99.4 | 76 |
| Unbuffered | 131.2 | |||
| Write | Buffered | 178.6 | ||
| Unbuffered | 126 | 208.5 | 175 |
A very large thank-you to Andy, for sharing both his hardware and expertise. He’s been testing spindle performance a lot lately; he may post results somewhere, and I’ll be sure to link to it here.
Also, thanks to the staff of that Peet’s, for not kicking us out, even though we unboxed a weird device and plugged it into the wall in your shop.
NB: We didn’t get to do mixed reads and writes, nor did we get to complete the table below. Please recall that these tests were performed in a coffee shop; we ran out of latte, and didn’t feel like ordering another, given the way these turned out.
Tags: AMS Venus ES5, nas, performance, sata