SSD hard drives have caused a storm of interest in recent months, what with its application of new technology to an old requirement, advertised improvements in speed and power consumption, and premium cachet due to its price and exclusivity. At the same time, this same application of technology results in new considerations that must be put to rest before an SSD can favorably compare to the old standard on most fronts. We explore these in this article with the help of the Patriot Warp 64GB SSD.

                                             The Patriot Warp SSD - a 2.5" bundle of robust memory chips.

Specifically, the Patriot Warp SATA 64GB SSD version 2, with product code PE64GS25SSD, denoting an MLC (Multi-Level Cell) NAND memory with 64GB of storage. SSD has been touted as the “next generation” storage solution for PCs due to its robust nature (no moving parts) theoretically no-lag random seek times (again no moving arm required for looking for a particular sector on a spinning disc) and power sipping nature (no mechanical components that need power to move). Recently though, internet conjecture and many, many forum posts has reported on certain anomalies with the performance of SSD drives that merit closer inspection.

                                             Standard SATA and SATA-Power connectors at the rear.

While it is true that sequential read and write speed (sequential here meaning reading and writing one contiguous block of data as opposed to random reads and writes all over a hard drive) and random access time (looking for a specific piece of data, block, or page on a hard drive) is very fast on an SSD due to its design of stand-alone memory chips instead of a magnetic storage system, some SSDs are encountering problems with random small writes to storage, meaning anything written in the single-digit and lower double-digit kilobyte range. This anomaly is notable due to reports from a number of reliable and knowledgeable sites online, and also from reports from individual SSD users.

To make this problem clearer, a simple example is in order. Say a user wants to transfer hundreds of small text files into the SSD. Or say a simple write operation initiated by a program like your web browser, saving multiple tabs before shutting down (a common enough occurrence with casual PC users) is performed. In other words, it can be any routine operation that’s happening on your PC at any moment that involves writing data to disk. When this occurs, the reported problem from a lot of SSD users is a “stuttering” phenomenon that renders your desktop nearly useless, as it causes your PC to freeze anywhere from half a second to a full 2 seconds every time a small write operation occurs.

                                             Take note of product numbers - this is v.2 of the Patriot Warp SSD.

The way we interpret the problem is the inability to cache to-be-written items to local memory, and thus an inability to handle multiple small files. A single large file is easily handled – after all, an SSD is practically one big section of “cache” memory and can be written to at will. The culprit in most of these cases is the combination of an MLC type of SSD with Jmicron’s controller – a controller with only 16KB of cache memory. With multiple small files, this small cache is quickly filled, and the SSD for all intents and purposes, chokes on the data and freezes. This has only been reported with the above combination of a Jmicron controller and MLC NAND SSDs. I already knew that the Patriot Warp v.2s are MLC drives, but I wasn’t sure whether it had the same problem as other SSD drives in its write operation.

                                             Chart from www.anandtech.com from their test of the Intel X25-M.

We set off to find out – emulating tests done by Anandtech (www.anandtech.com), we fired up the Iometer program and after much study simulated the exact situation said to make certain types of SSDs crawl. We were actually apprehensive about the result, as there are many conflicting testimonies and reports on the Patriot 64GB SSD version 2, and with no detailed spec sheet available, We could not know whether the Patriot held the dreaded Jmicron controller short of actually prying open the SSD casing (which we weren’t able – yet – to get permission to do).

                                             Important sliders to set for this particular test include the read/write distribution and the random vs. sequential settings.

We set up Iometer to write multiple small files over a 5 minute period in a random (rather than sequential) manner, to simulate multiple writes to disk from multiple programs running at the same time – typical for a multi-tasking user. This would probably translate to an antivirus running an update in the background, an IM client with a couple of conversation windows open, perhaps a Word file autosaving every few minutes, and a browser with some tabs open for email or research. We tested the Patriot SSD against our trusty Raptor 150GB mechanical hard drive (hopefully soon to be updated to a 300GB Velociraptor) to serve as a baseline. Comparisons between different SSD types would have been better, but for the moment, Patriot is the only one that has sent a sample to us.

                                             An above average result from our "conventional" mechanical hard drive - a Raptor 150GB.

Starting with the Raptor, we put Iometer to work and got some spiffy results that looked to be very much in line with online tests. Mimicking the tests done on Anandtech, we measured Input/Output operations per second, average write latency, and the maximum write latency. Our Raptor slotted in neatly above both a standard 7200 rpm notebook drive, the Seagate Momentus 7200.2, and an SLC (Single Layer Cell) SSD drive, an OCZ branded drive. The Raptor got 286 I/O Operations /sec, while the Momentus and the OCZ SLC SSD got 107 and 109, respectively. Remember, the test being performed is a purely random write test, not a direct reflection of a complete desktop environment, but for the purpose of isolating a potential problem with the Patriot, this test is perfect. And here’s the clincher: Average and maximum write latency – for our Raptor 10,000rpm drive, we got roughly 7ms and 32ms respectively, while the Momentus got 9.4ms and 76.5ms and the SLC SSD – 9.17ms and 83.2ms.

                                             Patriot SSD results. There is a huge disparity in I/O per second as compared to the Raptor.

So far so good. Also listed on Anandtech’s test charts is an OCZ MLC SSD with the Jmicron controller, and we’re about to find out whether our sample Patriot SSD has the same hang-up as the OCZ. With the Raptor as our own, local baseline, and the Anandtech sample units to further serve as bolsters to the test, we proceeded and got these results: I/O per second – 5.17, Average (write) Latency – 386ms, and Maximum (write) Latency – a horrible 901ms. That’s a whole second there folks. Before we go any further, it must be said that tests such as these are highly selective, meaning you may or may not encounter any kind of delay when using this drive in a normal user environment. That said, it is not too surprising to see that the OCZ Core – the MLC, Jmicron controller being tested by Anadtech, got results similar to the Patriot – I/O per second – 4.06, Avg Latency at 244ms, and a Maximum latency even worse than the Patriot, at 991ms.

                                             Advertised sequential read speed is actually on the conservative side, as testing reveals a 179.5 MB/s speed versus the advertised 175.

Mitigating these results somewhat are the numbers garnered from our HDTach test, which gave an amazing read result much, much faster than our test-bench Raptor. Advertised sequential read speed of 175MB/s was easily surpassed by the HDTach reading of 179.5 MB/s, and unlike spinning mechanical hard drives, read speeds over the whole size of the drive is very regular, compared to the Raptor’s steadily decreasing read speed as the mechanical arm moved across the whole disc surface. The Raptor gave up a respectable 136.2 MB/s, quite good for a mechanical hard drive, but far from the 179.5 MB/s of the Patriot.

                                             Patriot SSD tested with sequential reads - the Patriot is now much faster when compared to writing randomly.

The sequential write results as measured by Iometer – 504 I/O per second clarifies things a bit. The score is about twice that of the Raptor writing randomly, and much, much faster than the 5 operations per second achieved by the same drive while writing randomly. An average write latency of 3.9ms further clears up the evaluation of this drive. If, following online test results and our own comparison to our test bench setup has proven anything, it is that there is an anomaly with certain SSD drives, but not in all (the Intel MLC drive in Anandtech’s test showed very good, all-around performance). Most reports state that it is the combination of an MLC SSD and the Jmicron controller, but we can’t prove that (yet) with our own sample, although testing the Patriot shows pretty much the same behavior as other afflicted drives.

                                             HDTach sequential read resultsfor the Raptor hard drive.

Since the Jmicron controller has such a small cache, random writes must be stored before each piece of data can be written. With sequential writes, as we saw with the high numbers posted on Iometer, this is no problem. All of our other tests (including casual desktop testing and real-world applications which we will show in the future) shows a better and faster drive in the Patriot SSD than conventional drives, until you encounter the dreaded stuttering, which happened only a few times during our testing. For the moment, the Patriot Warp SSD has a very promising advantage over conventional hard drives – in addition to its speed, and the SSD’s natural robustness and low power consumption, we have quickly lowering prices (although still too high compared to conventional hard drives). All that’s left is further real-world testing to see whether all of the numbers are actually reflected during actual use. Watch this space for further updates.