SSD with CompactFlash | for robust servers

Post date: Mar 5, 2011 8:35:45 AM

Solid State Disks (SSD) based PCs are more robust than their HDD (disks with moving parts) versions. It is important for servers. Small office, home office (SOHO) servers are typically hidden in closets. Closets expose computers to extreme heat (yes, some of your servers are direct neighbors to HVAC heaters) and continuous vibrations from other equipment. This environment kills HDDs quickly. SSD can provide immunity, since electronics handles heat and vibration much better than mechanical parts. Read here about other benefits (like speed-up etc.)

Though as of 2009, the complete HDD replacement for SSD isn''t a viable solutions for SOHO. It is too expansive for the majority of applications. Fortunately, many applications can live with a partial solution. For example backup servers can fail. The data is most probably available on the workstations by the time of a failure. So the main feature is quick and hassle-free HDD swap. That is the HDDs with work data can be lost, but the HDD with system data (operations system, settings, user accounts, etc.) have to persist. A small and thus inexpensive SSD would resolve the maintenance and reliability challenges if it harbors only system data.

Small SSD can be substituted by CompactFlash cards (CF). CF provides even cheaper alternative. Firstly, the cards have much more size steps (everything from 16MB to 128GB in small increments). So you can easily find the closest fit to your needs and only pay for what is required. Secondly, CF cards most parts use the same storage technology as SSD. There is no cut on core performance and longevity. What differs is packaging, which significantly varies from CF to SSD. Packaging is the technical challenge that prohibits CF from being a main stream. This how-to will help you to master the technical challenge and put some smarts in your purchases by exploiting {packaging sophistication = costs) phenomenon, so you can enjoy SSD benefits on smaller budget; or you can bring SSD in tiny spaces.

SSD combines storage and storage controller - the same concept used in HDD. The concept has been introduced in mid 198X and known as IDE (integrated drive electronics). There were many modifications of IDE concept since then. The most recent (2003) and significant (total prevalence since 2007) change is SATA (serial ATA interface). Though, SATA is alive implementation of IDE (IDE is a parent term for all ATA) it is commonly referred as different from IDE because of new cabling and transfer mode. IDE today is used in context of older, parallel ATA (aka PATA, ATA-1, ATA-2, ATA-4, EIDE, ATAPI). SSD exists in both formats SATA and IDE. CF exists only in IDE format (CFast SATA cards are expected in 2010, but excluded here). That constitutes a significant problem for new PCs not having older IDE ports. Such systems may require PCI-to-IDE extension cards (~$17 to $50) if used with CF. The extention card is also advisable if you want to upgrade transfer speeds (read about it later in step 2.)

(Watch that the PCI2IDE chip is supported by your system (for Linux you''ll want to compile kernel and integrate the module. Check out supported chipsets in menuconfig.) And you need a bootable card. That is that the card''s BIOS can interact with motherboard''s BIOS. Typically the card is recognised as SCSI. It''s BIOS setup prompt appears right after motherboard''s setup prompt.)

The reach legacy of IDE variations presents constrains on applicability of CF cards. CF-cards are designed for mass consumer markets, where price and brand lock-in are decisive factors. Therefore, IDE controllers in such cards support limited set of IDE protocols. So you typically can save money while buying the same kind of cards because others wouldn''t work.

Three flavors of IDE protocols (that are relevant for CF) are PIO, DMA (aka WDMA, MW DMA) and UDMA. 1. Check what protocols are supported by your motherboard or PCI-to-IDE card. You can find it in your BIOS settings if you choose an IDE port and set it manually up. The protocols have a number attached to the protocol name. The numbers are decisive. If there is no match in both protocol and number, the CF card will not work.

The numbers also say about maximum data transfer speed. There is no need to buy a speedy card if it can''t be utilized by the motherboard. The price difference for different card speeds is substantial. 2. Determine maximum speed supported by your hardware. Check the speeds for IDE protocols in the following tables.

3. Find out maximum CF card speed. CF card speeds are given in three basic forms: a) directly in MB/s, b) in multiples of a standard speed of a CD-ROM, i.e. x80, x133, x300 etc., c) hidden behind a trade mark. In the later case you''ll need to find out specs for the given mark. The specs are typically presented in form a or b. Note, that CF card speeds are rounded to the next larger nice looking number. For example, 15MB/s aren''t 15 but rather 13.6, 12.7 or even 10.2 MB/s. You may want to find on the net real test data rather than rely on manufacturer''s marketing collateral. In addition to that there are two different speeds: reading and writing. You want both of them closer to higher level. There are some older technologies out there with difference in the speeds up to factor 3. Actual technologies give less than 30% speed drop for writing.

HDD speed is also similarly impacted. HDD''s and SSD''s never ever live up to the standards they carry on the hood. Recent consumer grade UDMA/100 disks give 40 to 70 MB/s UDMA/133 may have 60 to 90 MB/s. If you use disks attached via USB 2.X, the max speed is limited around 25 MB/s. A random data access/storage on HDD can slow down the process to as low as 1 MB/s to 2 MB/s. SSD''s and CF -on the contrary- can proceed overall much faster despite slower max speed ability. So if your hardware supports speedy protocols, it doesn''t mean you should max out the ability. 4. Consider practical max speed limitation for your application. Slower chips may be cheaper and more mature in terms of reliability. Additional info on CF advantages and applicability read here. 5. Take the lower value of steps 3 and 4.

6. Determine required CF Size. For modern operations systems the size is typically considered in GB increments. An actual-build PC-Linux kernel with overhang can be trimmed without much hassle to the size of approx 250 MB. A no-brain server installation would fit into 2GB. A really comfortable environment with source code, man and space for system add-ons and work will be completely satisfied by a 4GB CompactFlash. Windows Server 2008 can be trimmed to work on about 1 GB, provided there is 2 GB for operations somewhere else. A no-brainer would be 3.5 GB + 3 GB somewhere else. A comfortable environment starts depending on the size of installed applications with 8 GB. Anyway, check your properties.

7. Find and buy proper CF card. Recommend: start search with size, drill down to speeds and finally check supported protocols that they match with your hardware. If there is no data on that, DON''T buy, unless you can return the item at no costs. Many cards don''t fit. Once you''ve pointed out your cards and brands, find out the best delivery and/or price you can get.

In general use equipment there is no CF card slot. Though CF cards support IDE, you can''t attach them directly. You need an IDE-to-CF adapter. 8. Purchase an IDE-2-CF card. There are variations with several layouts and features. Among most important: supports one or two CF cards; attaches directly to IDE port or requires cable; connector for 2.5 IDE (laptops) or 3.5 IDE (desktops); straight-through or some electronics on board restricts card to certain IDE protocols. Typical costs $10 to $30 a piece.

IDE 2 CF card example for 3.5 IDE

If you use a PCI-to-IDE card than check if you have enough room in the PC for directly attached IDE2CF adapters + CF card atop of it. It can be too high. Otherwise buy a cable version. Check that the adapter supports your selected speeds and protocols. It is important that the IDE2CF card supports boot-up process. Two CF cards typically aren''t cheaper than one of doubled size. So there is no point for a double slot adapter, unless it is the same costs and you want to upgrade overall SSD capacity later. One slot cards are easier to screw down. Two slot cards are better to use directly plugged into IDE slot.

9. Install hardware and operations system. Once you have all right equipment, the process should be pretty straight forward. Your motherboard should recognize the disk. You handle it as if it were a hard drive. Put it forward in the boot sequence. Period.

Some systems may require manual setup. It typically constitutes Cylinders, Heads and Sectors (CHS) or LBA + disk size as well as transfer protocol setup for your card. For example Transcend TS4GCF133 has CHS 7769/16/63, PIO 6 or DMA 4, or UDMA 4 to offer.

10. Optimize the system for CF operations (optional). More about this read here.