FastTrak S150 SX4 Aces Low-Level Needs

 
 
By Henry Baltazar  |  Posted 2003-08-18
 
 
 

Serial ATA RAID continues its march into the workgroup-class market with the release last month of the FastTrak S150 SX4 RAID controller from Promise Technology Inc. At $199, the FastTrak S150 SX4 is a relative bargain compared with SCSI RAID controllers, which are typically found in the low-end to midrange server market. However, the trade-offs made to keep the FastTrak S150 SX4s cost down—it has only four drives and relies on the server CPU to operate—will likely keep the controller out of many midrange servers.

EXECUTIVE SUMMARY
FastTrak S150 SX4
Although Promise Technologys latest RAID controller might not be the most impressive storage device weve tested, it is a clear sign that Serial ATA RAID products have a good shot at replacing SCSI-based offerings in the low-end server range. The FastTrak S150 SX4 isnt very fast, nor is it flashy, but at $199 per card, it is ideal for low-end servers and workstations.

KEY PERFORMANCE INDICATORS
USABILITY GOOD
CAPABILITY GOOD
PERFORMANCE GOOD
INTEROPERABILITY GOOD
MANAGEABILITY GOOD
SCALABILITY FAIR
SECURITY GOOD
  • PRO: Inexpensive; good performance for price.

  • CON: No on-board processor means more work for CPU; supports only four drives.

  • EVALUATION SHORT LIST
    IBMs Lotus Software divisions Team Workplace Adaptec Inc.s SCSI RAID 2015S • 3Ware Inc.s Escalade 4 Port Serial ATA 150 RAID Card

    The FastTrak S150 SX4s test performance was satisfactory in eWEEK Labs tests, and we think the controller could be a major force in the workgroup server market, running general-purpose applications such as file serving and e-mail.

    Although the FastTrak S150 SX4s four Serial ATA ports should be enough for most low-end servers, we wouldnt recommend the FastTrak for use on a database server because databases run faster with lots of hard drive spindles. In a database environment, it is beneficial to have the database spread over several low-capacity hard drives instead of cramming the whole database onto a few large-capacity drives.

    The FastTrak controller would not be a good fit in a fast-growing server environment, where increasing storage demands could quickly consume all of FastTraks available capacity.

    Unlike more expensive RAID controllers, the Fast- Trak S150 SX4 does not have a built-in CPU; an integrated XOR engine handles parity calculations, but a full CPU is not on the board. Instead, the FastTrak S150 SX4 relies on server CPU cycles to function. Given the impact of CPU utilization, we dont recommend using this controller in older servers with slow processors. We used four Western Digital Corp. WD Raptor drives, the companys fastest enterprise-class Serial ATA drives, to test the FastTrak controllers performance.

    With current Serial ATA capacity as high as 250GB per disk, IT managers can still build relatively large RAID volumes using the FastTrak S150 SX4 controller.

    The FastTrak S150 SX4 can have as much as 256MB of local cache memory. (It uses standard 168-pin, dual in-line memory modules.)

    When we tested dual-CPU servers built on Advanced Micro Devices Inc.s Athlon 1.2GHz chips with 1GB of RAM per server, we found that CPU utilization ranged from 6 to 17 percent, depending on the type of tests.

    To ensure maximum performance, we used a separate drive (not connected to the Promise controller) to hold the operating system and applications so that we could use the FastTraks Level 5 RAID as a separate storage volume unencumbered by the operating system.

    For our tests, we ran the open-source Iometer 2003. 05.10 benchmark tool (available for free download at sourceforge.net/projects/ iometer), which measures storage hardware performance in terms of throughput (in megabytes per second) and transaction performance (in I/O operations per second).

    We found that the FastTrak S150 SX4 had a maximum throughput of 100MB per second on sequential read tests using large request sizes (512KB and 64KB). RAID performance topped out at 24,898 I/O per second (when processing small-request, sequential-read tests).

    When we ran more realistic loads with a randomized mix of read/write requests (at a ratio of 66 percent read to 33 percent write), the FastTrak S150 SX4 performed roughly 468 transactions per second (using a 2KB request size). This is good performance for a low-end controller. In tests, we found that the Promise Array Management software was a useful tool for checking on the health of our arrays and disks.

    Senior Analyst Henry Baltazar can be reached at henry_ baltazar@ziffdavis.com.

    Rocket Fuel