Title

64-Bit Vista ReadyBoost and Paging


Summary

Benchmark tests were run to measure the impact of ReadyBoost with heavy paging using a 4 GB flash drive. The system used was a Core 2 Duo with 4 GB RAM and 64-Bit Vista. An earlier flash drive marginally failed on the Vista writing speed performance test. The new one just passed, yet other benchmarks showed that writing speeds were significantly faster than the first drive. It is not clear why Vista shows such low speeds.

Booting time to the opening display was no different with and without ReadyBoost but following disk activity was 1 to 2 minutes less than disk/flash drive usage with ReadyBoost.

With heavy paging, ReadyBoost improved performance significantly, reducing program running time by up to three times, with data transfer rates faster than had normal disk input/output programming functions been used to avoid paging.

ReadyBoost improved application loading times somewhat, and this was consistent over three sets of tests. The most noticeable improvement was on system responsiveness following heavy paging, where reactivating an application via Restore was up to eight times faster.


Introduction

The minimum performance requirements for ReadyBoost are 2.5 MB/second for 4 KB random reads, 1.75 MB/second for 512 KB random writes and an access time of 1ms or less. The Vista test rejected a flash drive as unsuitable, failing on writing speed at 1723 KB/second, although other benchmarks suggested that both requirements could be met. A second drive, enhanced for ReadyBoost, was reported as reading at 5486 KB/sec and writing at 1814 KB/sec, with the latter again slower than other benchmarks.

ReadyBoost tests were carried out by using my paging benchmark described in Paging.htm. A small number of applications were loaded and minimised before the tests and restored at the end, with response times being measured. The system and flash drives tested were:


  Core 2 Duo 2400 MHz, Asus P5B motherboard, 800 MHz DDR2 RAM,
  Seagate ST3400633AS SATA-300 disk, 16 MB buffer, 7200 RPM,
  GeForce 8600 GT graphics, 64-Bit Windows Vista Home Premium.

  Staples USB Flash Drive 4 GB
  SanDisk Cruzer Micro 4 GB, Enhanced for ReadyBoost

ReadyBoost default SFCache size of 3.76 GB was used.

To Start

Disk Benchmarks

DiskGraf benchmark writes and reads files at 1 KB to 1024 KB block sizes. It also measures random reading speeds of 1 KB, from 1 MB data up to the chosen file size (maximum 1 GB). Results are below using NTFS formatting. For FAT based results and more details see DiskGraf Results.htm.

The random reading test is for a 1 GB file where the SanDisk drive has reasonably consistent results for all data sizes. The shortest time of 0.52 milliseconds gives 1.92 MB/second, similar to the reading speed measurements at 1 KB. The results suggest that random access times can be largely ignored and measurements reflect a reading overhead. The reading speed of the SanDisk drive at 4 KB (5.4 MB/second or 0.74 milliseconds per block) is similar to the Vista measurement. Writing speed at 512 KB is 3.6 MB/second or 142 milliseconds per block, compared with Vista's 1.814 MB/second or 282 milliseconds. [Numbers are subject to slight adjustment depending whether K is 1000 or 1024 etc.].

CDDVDSpd benchmark measures writing and reading speeds of one long file and 520 small files, with selections 1 MB/2 KB, 2 MB/4 KB upwards. See CDDVDSpd Results.htm for further details. Results below include average file writing and reading time of the small files in milliseconds. This includes overheads for file opening and closing, causing much slower speeds for the smallest files. Even with these overheads, the SanDisk drive writing speed is faster than the Vista 512 KB result at 32 KB. Above 32 KB, adding a few milliseconds for random access would make little difference.

The SanDisk drive is clearly much faster on all the writing tests but Vista measurements indicate that they are nearly the same.

To Start


DiskGraf

MBytes/Second at KB Block Size 1 2 4 8 16 32 64 128 256 512 1024 Write Staples 0.21 0.37 0.47 0.54 0.59 2.50 2.50 2.70 2.70 2.60 2.60 SanDisk 0.61 0.95 1.40 2.30 3.40 4.20 4.80 4.00 3.90 3.60 3.30 Read Staples 1.90 3.90 5.20 6.80 7.80 8.70 9.10 9.20 9.10 9.20 9.20 SanDisk 1.90 3.10 5.40 8.90 12.30 14.60 16.50 16.60 16.50 16.60 16.50 Random reading milliseconds 1 KB from 1 MB to 1024 MB data Staples 0.65 0.54 1.31 1.70 1.27 1.17 0.84 0.71 0.59 2.18 2.22 SanDisk 0.52 0.52 0.56 0.55 0.52 0.53 0.54 0.58 0.60 0.62 0.66

CDDVDSpd

Large Files Small Files Per File MB Write Read KB Write Read Write Read MB/s MB/s MB/s MB/s msecs msecs Staples 1 2.58 8.94 2 0.06 0.93 35 2.1 8 2.52 9.16 4 0.14 1.69 28 2.4 16 2.61 9.09 32 0.76 5.28 42 6.1 32 2.60 9.14 64 1.28 6.71 50 9.5 SanDisk 1 3.26 16.15 2 0.19 1.02 10 2.0 8 3.63 16.54 4 0.36 1.91 11 2.1 16 4.04 16.40 32 1.95 7.74 16 4.1 32 4.05 16.51 64 2.83 11.03 23 5.8


To Start

ReadyBoost Tests

Tests were run without and with ReadyBoost, each starting with power on. Applications loaded and timed were, in order, PhotoShop Elements, then a photograph, MS Works, Spreadsheet application, and Internet Explorer. These were then minimised. Paging tests were run using the 64 bit program IntBurn64. See Paging.htm. It is not possible to obtain consistent speed measurements with paging, some of the larger variations being due to inconsistent delays between tests. To reduce this effect (and produce the worst performance) , the tests were run via commands in a BAT file (see below). The BAT file was also opened and minimised and the benchmark folder opened. After the paging tests, each application was restored and response time measured. This was considered important following observations of extremely slow responsiveness during earlier paging tests.

Booting times of around 49 seconds to Log On and 11 seconds to Welcome Centre were the same for both set ups. Following this there was 3 to 4 minutes of disk activity with no ReadyBoost and 5 to 6 minutes of disk and flash memory usage when the USB drive was in place.

Results of two runs of the paging tests are shown below. With ReadyBoost, the USB drive activity light was flashing frequently and the benefits are clearly demonstrated in the testing time recorded within the program. Without ReadyBoost, there was a noticeable delay before the BAT command started the following test. So, overall elapsed time is shown besides the total reported time. [Reduce totals by 17 seconds as there are multiple passes on first few tests for minimum running time of 5 seconds]. As shown, ReadyBooost can improve performance by more than three times with heavy paging activity.

During the paging tests, minimum measured data transfer rates were 26 and 29 MB/second for the two sets of tests without ReadyBoost, much slower that had normal disk input/output been used. Minimum speeds with ReadyBoost were 88 and 89 MB/second, faster than disk I/O.

ReadyBoost improved application loading times somewhat, and this was consistent over three sets of tests. The most noticeable improvement was on system responsiveness following heavy paging, where reactivating an application via Restore was up to eight times faster.

To Start


                        Seconds To Write and Read Data Array Bytes

             20M   30M    2G    3G  3.5G  3.6G  3.7G  3.8G  3.9G    4G Total  Elap

               5     5     5     5    33    12    34   146   237   320   803  1025
  ReadyBoost   5     5     6     7    31    18    23    70    91    93   348   399
               5     5     5     5    25    22    43   108   256   278   753   970
  ReadtBoost   5     5     5     6    20    17    28    73    54    89   301   369



                     Loading Seconds                 Restoring Seconds
    
             Pshop Photo Works    SS    IE      BAT    IE    SS Works Pshop

               8.8   2.3   2.8   1.8   2.8      8.0  15.0   5.5   7.2   4.7
  ReadyBoost   6.2   2.3   2.1   1.5   2.6      1.5   1.8   1.8   1.5   1.5



  BAT File Commands

  Start /wait IntBurn64 KB 20000,   Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 30000,   Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 2000000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 3000000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 3500000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 3600000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 3700000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 3800000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 3900000, Auto, Paging, Log pagingRBno.txt
  Start /wait IntBurn64 KB 4000000, Auto, Paging, Log pagingRBno.txt


To Start




Roy Longbottom October 2007

The new Internet Home for my PC Benchmarks is via the link
Roy Longbottom's PC Benchmark Collection