Blogger Profile
XBox360 Blog
MyPage Links
- Mossy Cup Brewing
- MAME Cabinet
- Flightsim Cockpit
- My PVR
- My Lego sets
- Computer Systems Page
- Family Photos
- TheMcClains.net
Blog Archives
Contact me
Jeff McClain's Home Page
Saturday, February 18, 2006
Drive and Network Bandwidth
Spent some time bench marking some of my network file transfers and disk speeds. I'm really a little disappointed that Coolmax decided to only include an ATA-33 speed drive controller in that NAS enclosure. It really makes it almost useless for good speed transfers. It is only roughly 1/20th the speed of a local SATA-150 drive (granted, the 100Base TX network would limit it far before it got to that, but still, you could have over tripled the performance with an ATA-100 controller over a 100Base TX network, and the USB 2.0 480Mb/sec is basically wasted on this unit), and only about 1/13th the speed of a typical PATA-133 local drive. When you start talking video, that is getting pretty slow (only marginally faster than typical 54Mb 802.11g WiFi connections).
I may have to think about getting one of the Thecus 2100 Yes NAS with twin gigabit ethernet and dual SATA-150 drive support for some blinding fast network file support. But it costs $350, just for the bare enclosure...
I also took some time to create a RAM Drive and test that bandwidth. I chose a Windows XP compatible driver from SuperSpeed, and it is only a demo version.
Benchmarked Data Bandwidth Comparisons
| Device | File System | Size (GB) | Interface | Cache (MB) | Speed (RPM) | Access Time (ms) | Sustained Read (MB/s) | Sustained Write (MB/s) |
|---|---|---|---|---|---|---|---|---|
| PC3200 CL3.0-3-3-8 DDR1 DRAM | RAM | 2GB | Dual Channel | - | - | - | 2584 | 2584 |
| SuperSpeed RAM Drive Plus | RAM | 1.2GB | Dual Channel PC3200 | - | - | - | 966 | 975 |
| 1000Base T | 1000Base T | RAM | TCP/IP | - | - | 0.3 | 54 | 54 |
| 100Base TX | 100Base TX | RAM | TCP/IP | - | - | 0.3 | 10 | 10 |
| 802.11g WiFi | 54Mb/sec 802.11g | RAM | TCP/IP | - | - | 1.7 | 2.3 | 2.3 |
| Maxtor 6Y250M0 | Local NTFS | 250 | SATA-150 RAID-0 | 8 | 7200 | 7 | 116 | 89 |
| Maxtor 6Y200M0 | Local NTFS | 200 | SATA-150 RAID-1 | 8 | 7200 | 7 | 90 | 84 |
| Maxtor 6Y200M0 | Local NTFS | 200 | SATA-150 | 8 | 7200 | 7 | 57 | 57 |
| Maxtor 6Y200P0 | Local NTFS | 200 | PATA-133 | 8 | 7200 | 7 | 41 | 38 |
| Maxtor 6Y200M0 | 1000Base T NTFS | 200 | SATA-150 | 8 | 7200 | 7 | 37 | 34 |
| Toshiba 20GB 2.5" | Local NTFS | 20 | PATA-33 | 0.5 | 5400 | 18 | 16 | 6 |
| Maxtor 6Y200M0 | 100Base TX NTFS | 200 | SATA-150 | 8 | 7200 | 7 | 9 | 6 |
| Coolmax CN-550 Seagate 9Y704G Barracuda | 100Base TX FAT32 NAS | 250 | PATA-33 | 8 | 7200 | 7 | 3.1 | 3.0 |
| Coolmax CN-550 IBM DeskStar DTLA-307030 | 100Base TX FAT32 NAS | 30 | PATA-33 | 2 | 7200 | 7 | 3.1 | 2.9 |
Comments:
A couple things to note on this (looking back, I realize how little info I gave on how this was all taken).
1) The first memory score that is showing 2.6GB/sec was a Sandra raw memory transfer speed for around 1MB chunks.
2) The second memory score that is showing just under 1GB/sec, is treating a RAM drive like a disk and read/write testing it (like all the other drives were) in Sandra.
3) The network stuff is looking at Sandra's raw data read/write bandwidth across that network. Keep in mind, this is throughput it real values, meaning you have packet header over head and packet loss built into those numbers. I'm not sure why the 1000Base T is maxing out at 54, but that is going through 2 Gigabit hubs and a pretty long send route with multiple patch cables in between (all Cat5, so maybe the cables are limiting some of that).
4) I was surprised by how well the SATA-150 RAID-1 (mirroring) did. So much so, that I tested it several times and triple checked it other ways. It really is better than a single SATA-150 drive in the same setup. Not sure if that is because it takes advantage of the buffer and ping pongs the raw writes primary between the drives and then syncs them later or what. Seems strange though. It is almost approaching the RAID-1 (stripped) performance.
5) You can see how the overhead of putting even a gigabit ethernet in the middle of a file transfer and doing a disk to disk transfer. In this case, for the read speed, I'm writing TO my RAID-0 drive config, so that shouldn't have limited it. And vice-versa for the write speed.
6) File transfers over a 100Base network, essentially approach the raw full 100Base saturation of 10MB/sec (writes probably are worse, because of resends).
7) The Coolmax drive enclosure just plain sucks. It isn't even 1/3 of what it should be capped by from the 100Base network. And even my laptop drive can get 4x those speeds on the same PATA-33 interface.
Post a Comment
1) The first memory score that is showing 2.6GB/sec was a Sandra raw memory transfer speed for around 1MB chunks.
2) The second memory score that is showing just under 1GB/sec, is treating a RAM drive like a disk and read/write testing it (like all the other drives were) in Sandra.
3) The network stuff is looking at Sandra's raw data read/write bandwidth across that network. Keep in mind, this is throughput it real values, meaning you have packet header over head and packet loss built into those numbers. I'm not sure why the 1000Base T is maxing out at 54, but that is going through 2 Gigabit hubs and a pretty long send route with multiple patch cables in between (all Cat5, so maybe the cables are limiting some of that).
4) I was surprised by how well the SATA-150 RAID-1 (mirroring) did. So much so, that I tested it several times and triple checked it other ways. It really is better than a single SATA-150 drive in the same setup. Not sure if that is because it takes advantage of the buffer and ping pongs the raw writes primary between the drives and then syncs them later or what. Seems strange though. It is almost approaching the RAID-1 (stripped) performance.
5) You can see how the overhead of putting even a gigabit ethernet in the middle of a file transfer and doing a disk to disk transfer. In this case, for the read speed, I'm writing TO my RAID-0 drive config, so that shouldn't have limited it. And vice-versa for the write speed.
6) File transfers over a 100Base network, essentially approach the raw full 100Base saturation of 10MB/sec (writes probably are worse, because of resends).
7) The Coolmax drive enclosure just plain sucks. It isn't even 1/3 of what it should be capped by from the 100Base network. And even my laptop drive can get 4x those speeds on the same PATA-33 interface.
