A Network Performance Evaluation

[ Previous Tip ] [ Mac Tips ] [ Next Tip ]



I run mostly Apple Macintosh computers at my home and I use Apple's Airport series of networking devices along with a wired Ethernet network (mostly 100BaseT with one section running 1000BaseT). The Airport is a little more expensive than the PC flavor of network gear, but it's a lot easier to set up and integrates into an Apple environment better. Since Apple came out with the Airport Extreme wireless networking system, I've been slowly accumulating more Airport gear that supports Airport Extreme, or more generically, 802.11g.

As of February 2007, Apple has introduced another version of the Airport Extreme, this one supports the 802.11n draft standard. 802.11n runs faster than 802.11g through the use of newer modulation schemes and a dual band approach combining 2.4 and 5 GHz wireless links. I just acquired and 802.11n version of the Airport, but I don't have any computers that use 802.11n so you won't see any wireless evaluations of this new hardware yet.

As part of the process of equipment accumulation, I have been evaluating the performance of all of this gear. I generally find that the stuff meets its claims. However, the configuration of the network gear and the connections established have a great deal to do with wireless network performance too. It is these variations that I am interested in so that I can avoid the most disadvantaged configurations.

The original 802.11b WiFi standard supports data rates of up to 11 Mbits/sec under ideal (ideal=marketing) conditions. Conditions are rarely ideal. The rates usually obtained are about half the maximum rate. 802.11g supports rates of 54 Mbits/sec, or about 5x that of 802.11b and in real world usage, again the rates are about half of theoretical maximum. 802.11n is supposed to support rates of about 200 Mbits/sec.

The first section of this page tests particular Airport network configurations. The second part tests wired and wireless networking between a set of computers with varying capability.


Acronym/Name Description
WiFi Wireless Fidelity, the industry adopted zinger for a compatible wireless networking device
Airport Apple Computer's trade name for wireless networking (WiFi)
AEBS Airport Extreme Base Station (supports 802.11b and 802.11g, 2007 model supports 802.11n as well)
AXBS Airport Express Base Station (also supports 802.11b and 802.11g)
WDS Wireless Distribution System
A method for extending the range of an Airport network by adding remote base stations connected to the main network wirelessly
WAN Wide Area Network, or the path to outside world
10baseT A wired networking standard that supports traditional Ethernet rates of up to 10 Mbit/sec or about 1 Mbyte/sec
100baseT A wired networking standard that supports enhanced Ethernet rates of up to 100 Mbit/sec or about 10 Mbyte/sec
1000baseT A wired networking standard that supports enhanced Ethernet rates of up to 1000 Mbit/sec or about 100 Mbyte/sec. Also known as Gigabit Ethernet
FireWire 400 A wired serial connection supporting a data rate of 400 MBits/sec.
LAN Local Area Network. The local network configuration under the user's control.

Wireless Network Performance

Peer to Peer Performance

A peer to peer configuration is where two computers communicate directly via a wireless link, I find that the manufacturer's claims of performance are generally supported (factored by 0.5x for real world conditions).

This round of testing is fairly straightforward. I turn off all local base stations to clear the air of any interference and then use one computer to create a peer to peer network. In this case it is an iBook G4 with an Airport Extreme card. Encryption is turned off. Two client computers were used for comparison, a PowerBook G4 with Airport Extreme and an iBook G3 with a standard Airport card (802.11b only). I copy a large file (in this case, the OS 10.3.5 Combo updater disk image file) from the server to the client to see what happens. I use the Activity Monitor to view the peak data rate and I use a stopwatch to time the transfer time to calculate the average rate. I also observe the Activity Monitor to judge the steadiness of the data rate. When the data rate is flat, the average rate is usually 90% of the peak rate. When the data rate is choppy or had dropouts, the average rate is lower.

When I see a steady rate, I make only one timed run because experience has told me that this is the best condition that I will see. When the rate has dropouts or is highly variable, I make several runs to see if this condition repeats. It has turned out that only one test condition consistently produced choppy rates and this case was choppy in the extreme.

Connection Average Data Rate
Extreme to Extreme 3.3 MBytes/sec
Airport to Extreme 0.62 MBytes/sec

This makes a lot of sense, the rate differential is almost exactly 5 to 1. Assuming 8 bits/byte, then the rates are 26.4 MBits/sec and 4.9 MBits/sec, or about half of the theoretical rates. So far so good.

Airport Base Station Related Performance Issues

Another possible impact on network performance is throughput issues related to a wireless base station. The base station acts as a bridge between wired and wireless clients or between different wireless clients. The base station will have no impact on the communications rate between wired clients.

When both clients are wireless and are communicating through the same base station, the clients have to share the available bandwidth so one would expect that the rate will drop somewhat. I find that if the clients are matched, the rate drops by about half. This is probably because the base station has to share its attention between two clients. Further, if encryption is used, some combination of hardware and software has to encrypt and decrypt the data. This will take some time as well. The heavier the encryption, the more likely that it will take more time to do and therefore slow down the works even more. I wanted to see what devices were impacting encrypt/decrypt performance and by how much.

To get this data, I set up three computers and three base stations. Only two of these computers and a single base station is on at any one time so that there would be no interference issues. The three base stations are a Graphite ABS, a Snow ABS and an Extreme ABS. The three computers were an iBook G4, a PowerBook G4 (aluminum) and an iBook G3. For all tests, the iBook G4 was designated as the "server." It was there to provide a source of data, in this case an 85.2 MByte file. The server is always connected wirelessly. To avoid any network signal strength issues, the computers and base stations being tested are all within a few feet of each other.

The two "clients" were used to download the file from the server. The Activity Monitor is used to monitor the peak data rates and to judge the steadiness of the rates. A stopwatch is used to time the download to get an average rate. In most cases, the peak rate was just a little higher than the average rate which itself is pretty steady. However, this isn't always the case. It was those odd cases that I was really looking for so that I could try to avoid those cases.

Graphite Airport Base Station

The Graphite Airport Base Station is the first model introduced in about 1999. It got its name from its gray color. The unit didn't have a WAN port to speak of, at that time a dial up connection was the WAN port for most users so it used a modem port for its WAN connection. I had DSL at the time so I used a Quadra 700 running a software NAT/DHCP router with the Graphite ABS acting only as a wireless access point. This Graphite ABS is still in use, but in its intended configuration with the modem serving as the WAN interface at a house that doesn't have high speed internet service.

The results of the tests with a Graphite base station are listed below.

Graphite Base Station Results
Client Encryption Client Connection Peak Data Rate
Average Data Rate
Average to Peak Ratio Data Rate Stability
iBook G3 40 Bit wireless 0.304 0.16 54% very choppy
wired 0.555 0.51 92% steady
none wireless 0.366 0.32 88% steady
wired 0.711 0.66 92% steady
PowerBook G4 40 Bit wireless 0.331 0.29 89% steady
wired 0.558 0.17 90% steady
none wireless 0.361 0.279 85% steady
wired 0.711 0.72 101% steady

With no encryption and a wired client, the peak rate achieved is the same. This tends to indicate that the CPU performance difference (which is significant) between these clients has minimal impact on the data rates. The iBook is slower by 10% on the average, but this is not a really significant difference. Also, the PB data has an average rate that is 1% higher than the peak rate. This is an artifact of the somewhat imprecise data collection methods.

Turning on encryption for the wired clients leads to identical results as a function of encryption. This is to be expected because the wired client doesn't have to do any encryption work. This is all done by the base station and the server which was the same for these test pairs.

With no encryption and a wireless connection, the results are again the same. This indicates that the computers are fast enough to keep up with the work and the wireless performance is the same.

The Graphite Base Station is reported to have insufficient internal processor power to handle encryption at full rate. The data does not indicate this. With the Powerbook, the peak rate with encryption is a little lower, but the average rate is a little better. This is probably due to data precision because the difference is small.

However, turn on encryption with a wireless connection, and the iBook drops nearly by half. This implies that whatever is doing the decryption (either the iBook or the Airport card, I don't know which) is hurting. The PowerBook has about the same performance with or without encryption. The iBook just dies when encryption is enabled.

Snow Airport Base Station

The next table is the same test done on the next generation Airport base station. This one is commonly called the "Snow" version as it came in a white case. This is the first one with separate WAN and LAN ports and a modem as well. This one replaced the Graphite ABS when the software router was finally retired (it crashed too much and was hard to get started again).

Snow Base Station Results
Client Encryption Client Connection Peak Data Rate
Average Data Rate
Average to Peak Ratio Data Rate Stability
iBook G3 40 Bit wireless 0.310 0.24 76% steady
wired 0.559 0.49 87% short drops
none wireless 0.387 0.35 91% steady
wired 0.715 0.51 71% very choppy
PowerBook G4 40 Bit wireless 0.339 0.31 92% steady
wired 0.560 0.52 92% steady
none wireless 0.389 0.36 93% steady
wired 0.717 0.66 91% short drops

The Snow works better than the Graphite and doesn't completely fall apart with encryption enabled. However, it seems to have a problem on the LAN side. The wired connections, especially to the iBook, tended to have either short dropouts in the data rate or in the highest data rate case, the result was very chopped up.

Airport Extreme Base Station

The results of the same tests with an Airport Extreme base station are listed below. During these tests, the AEBS was locked to 802.11b rates because I wanted to test the difference between the base stations internal capability so that they had to be set to the same maximum wireless capability. The Graphite, Snow and Extreme base stations all use different internal CPUs and software. However, I did make one change, the encryption method used for the AEBS was 128 bit WEP. This method is not supported by the Graphite base station, it ran at 40 bit WEP. I installed the Extreme base station in place of the Snow when the Snow became unstable after a couple of years. It still works but it tends to lock up every few days when used as a high speed router.

Airport Extreme Base Station Results
Client Encryption Client Connection Peak Data Rate
Average Data Rate
Average to Peak Ratio Data Rate Stability
iBook G3 128 Bit wireless 0.350 0.32 91% steady
wired 0.732 0.68 92% steady
none wireless 0.395 0.36 92% steady
wired 0.731 0.68 93% steady
PowerBook G4 128 Bit wireless 0.399 0.37 92% short drops
wired 0.735 0.66 90% short drops
none wireless 0.396 0.37 93% steady
wired 0.733 0.68 93% steady

Overall, the results with the AEBS are better than those of the Graphite base station. In particular, with encryption enabled and a wireless connection to the iBook, there isn't the dramatic slowdown. The data rate remains stable and consistent unlike the Graphite rate which was seriously choppy. Overall, there was not much difference in the data rates between the iBook and the Powerbook.

Another thing that I wanted to learn is the impact of an 802.11b client on the data rate of an 802.11g client when both are connected to the same base station. The Airport Extreme Base Station supports a b/g compatibility mode so that both rates can coexist. However, the base station still has to support the lower rate client part of the time and this requires its attention at lower data rate. While the base station is servicing the lower rate client, it is not servicing the higher rate client. In this case, an encrypted link was established at the 802.11g rate and then the data rates were tested with an 802.11b client connected or not connected to that base station. The 802.11b client wasn't actually doing anything, it was just there.

Impact of an 802.11b Client on an 802.11g link
Client Encryption Client Connection Peak Data Rate
Average Data Rate
Average to Peak Ratio Data Rate Stability
no 802.11b client 128 Bit wireless 1.97 1.85 94% steady
wired 3.9 3.7 95% steady
with 802.11b client 128 Bit wireless 1.11 1.00 90% steady
wired 2.03 1.94 95% steady

The table tells the story, when an 802.11b client comes in range, the 802.11g users take almost a 50% hit in data rate even if the 802.11b client isn't requesting any bandwidth. This test doesn't show the additional impact of an 802.11b client actually doing anything, but I am guessing that the high data rate users will hurt even more. If enough 802.11b clients are present and working the whole base station will eventually run at 802.11b rates.

Another test was to see what impact 256 bit WPA encryption has on 802.11g data rates vs 128 bit WEP encryption. This test is done only with the PowerBook because the iBook doesn't support 802.11g.

Impact of Encryption on an 802.11g link
Client Encryption Client Connection Peak Data Rate
Average Data Rate
Average to Peak Ratio Data Rate Stability
PB G4 WPA wireless 1.94 1.85 95% steady
wired 3.75 3.55 95% steady
128 Bit wireless 1.99 1.85 93% steady
wired 3.99 3.87 97% steady

WPA is a 256 bit encryption scheme so it is reasonable to expect that it has more processing overhead. The data seems to indicate that the data rate with WPA is about 93% to 97% of the 128 bit WEP data rate.

Another thing that I want to see is the overall impact of encryption on data rate vs the processor performance of the client. In this case, the base station rate was locked at 802.11b rates because the iBook could not support 802.11g rates. Then the encryption was varied from none, to 40 bit WEP, to 128 bit WEP to 256 bit WPA.

Impact on Data Rate of Encryption Method Vs CPU Performance
Client Encryption Client Connection Peak Data Rate
Average Data Rate
Average to Peak Ratio Data Rate Stability
iBook G3 None wireless 0.407 0.36 92% steady
40 bit 0.352 0.32 91% steady
128 bit 0.352 0.32 91% steady
WPA 0.319 0.29 90% steady
PowerBook G4 None 0.396 0.37 92% steady
40 bit 0.392 0.36 92% steady
128 bit 0.396 0.37 92% steady
WPA 0.390 0.36 92% steady

rate vs encryptionThe result is a little easier to see in graphical form. The graph has data for all three types of base station. As expected, the change in encryption had little impact on the faster computer, but the harder encryptions drag the slower computer down significantly more. Further, the older base stations hurt more with encryption, probably because they really don't have the processing power to keep up.

Intel based iMac Issues

Airport Data Rates

Update 29Jan06

When the iMac Core Duo (Intel) first came out, it appeared to have real difficulties with Airport related throughput. There is a web site that tests network speed. SpeakEasy is a high speed network provider. They provide a tool that will show you how pitiful your network is so that you might be inclined to buy their 6 Mbit/sec service. I've got a 6 MBit/sec cable connection anyway. Their tool is quite good and easy to use. It reports average speed for a short test. The tool also meters peak rates. If you have the Activity Monitor open in its Network pane at the same time, you can get a graphical indication of the peak rate and also a reading on the absolute peak. I ran this tool on 6 different computers, 4 wireless and all 6 wired. All but the iMac behave in an expected way. The iMac is clearly having difficulties.

The results are in the table below. I tested all 4 wireless computers with 6 different base station configurations, 802.11g and 802.11b each with WPA, WEP and no encryption. The base station was the same for all tests, an Airport Extreme configured as a wireless access point. It is connected to another AXBS (which is the router connected to the cable modem) via CAT5 ethernet. The router and the cable modem were not touched during these tests.

Several passes were made at each test. If all the results were about the same, the highest values were used. If the results varied a lot, a range of results is presented. If just one test was out of family, it was discarded because there might have been other network traffic interfering. In all cases the upload rate started slow and built up to the peak rate of 384 kBits/sec, averaging around 350. In the download case, the rate usually popped up to a fairly steady rate EXCEPT for the iMac Core Duo. It would wander all over the place when it was unstable.

SpeakEasy Test Results
Computer Connection Method Average Data Rate (kBits/sec)
Method Security Download Upload
Rate Stability
400 MHz G4 PowerMac Wired n/a 4639 stable 350
533 MHz G4 PowerMac Wired n/a 4645 stable 350
20" iMac Core Duo Wired n/a 4735 stable 355
802.11b none 2803-4817 unstable 351
WEP 2774-4737 unstable 352
WPA 2779-4742 unstable 353
802.11g/b none 2754-3300 unstable 355
WEP 2517-4671 unstable 349
WPA 2771-4732 unstable 351
802.11g none 3161-4314 unstable 351
wired through AXBS n/a 4747 stable 358
1.25 GHz G4 15" Aluminum PowerBook Wired n/a 4767 stable 323
802.11b none 4725 stable 355
WEP 4686 stable 356
WPA 4727 stable 356
802.11g/b none 4730 stable 354
WEP 4709 stable 355
WPA 4686 stable 356
1.07 GHz G4 iBook Wired n/a 4615 stable 353
802.11b none 4549 stable 356
WEP 4582 stable 347
WPA 4582 stable 347
802.11g/b none 4705 stable 356
WEP 4560 stable 350
WPA 4578 stable 347
500 MHz G3 iBook
802.1b capable only
Wired n/a 4643 stable 354
802.11b none 3700 stable 354
WEP 3362 stable 354
WPA 3056 stable 353
802.11g/b none 3789 stable 356
WEP 2844 stable 353
WPA 2814 stable 353

All of the wired rates are about the same, this is the capability of the router/cable modem combination. The 500 MHz iBook is clearly hurting for wireless performance no matter what the Airport was doing.

The data that really stands out is with the iMac Core Duo running against an 802.11g/b access point. These rates were all over the place. Sometimes it would start weak and then pick up to a good rate at the end. Sometimes it would start strong and then drift downward during the download. A very few times it ran very strong for the entire download. The security setting seemed to have no impact at all.

Update 6 Feb 06

The more I use the iMac Core Duo, the more convinced that it has a problem. The problem is probably that it is a completely new system and it has some remaining bugs, although not very many. The wireless data rate problems seem be totally centered around the wireless implementation in the Core Duo itself. The problems may be related to this particular computer or my particular environment, but somehow I doubt that.

The table below shows the data rate vs time comparing the performance of a PowerBook G4 and the iMac Core Duo. I believe that the G4 is working properly so I used it for a comparison. An iBook G4 behaves pretty much the same as the PowerBook so the vote is 2 against 1 indicting the iMac.

In these tests, I use the Activity Monitor with the Network pane showing to graph the data rate vs time for various data transfer scenarios.

iMac Core Duo Vs. PowerBook G4 Data Transfer Rates
Test Condition Data Source Data Destination Transfer Rate Notes
Wireless download of a large file from apple.com downloads site apple.com iMac Core Duo graph This is typical of the choppy downloads I get from the internet on the Core Duo. The wireless link gets 4 bars on the Airport indicator. Note that the period of this "chop" is 10 seconds.
PowerBook G4 graph This is the same file downloaded to the PowerBook sitting next to the iMac. It is a little more ragged than usual, but significantly better than the Core Duo. The data rate stays above 400 KB/sec for the whole download. The Airport signal strength is also 4 bars.
Wired download of a large file from apple.com downloads site apple.com iMac Core Duo graph This is what the wireless downloads should look like. The iMac is connected directly to the LAN and gets the full data rate of the cable modem.
Wireless download of a large file from apple.com downloads site with a stronger AEBS signal apple.com iMac Core Duo graph One poster on the Apple Discussion boards had noted that his choppy downloads were substantially improved when the iMac and the AEBS were very close together. I moved the iMac to about 10' from the AEBS and I see virtually no change.
Wireless download of a large file from apple.com downloads site with interference robustness enabled apple.com iMac Core Duo graph Interference Robustness has essentially no impact. This is the same result from the SpeakEasy test.
Wireless download of a large file from apple.com downloads site with no AEBS encryption apple.com iMac Core Duo graph The AEBS was set for no encryption, no impact, again similar to the SpeakEasy test.
Wireless download of a large file from apple.com downloads site using an Airlink AR315W wireless router instead of an AEBS apple.com iMac Core Duo graph The Airlink was set up to replace the AEBS that was the router. Downloads on the PowerBook and an iBook are normal but the iMac is still having trouble. Another test using an AXBS as a wireless access point yielded the same result.
Wireless File Transfer through the same Airport Base Station iMac Core Duo PowerBook G4 graph The data rate when transferring a 54 MB file via an Airport Extreme Base Station is choppy no matter which way the data was sent.
PowerBook G4 iMac Core Duo graph
Wireless File Transfer through different Airport Base Stations on the same LAN iMac Core Duo PowerBook G4 graph A similar situation occurs when each computer is connected to a DIFFERENT AEBS which are interconnected by Ethernet. The data rate is doubled because the base stations don't have to share their RF bandwidth between two computers.
PowerBook G4 iMac Core Duo graph
Peer to peer wireless file transfer, no base station involved iMac Core Duo PowerBook G4 graph The story changes considerably when the AEBS is removed from the transfer entirely. The data rate is rock solid for the whole transfer.
PowerBook G4 iMac Core Duo graph
updatedWired connection to AXBS acting as a WDS remote base station apple.com iMac Core Duo graph I set up an Airport Express Base Station at my desk as a WDS remote base with wireless access disabled and connected it to the iMac via Ethernet. All is now well.

More Data

When I ping the Airport Extreme Base Station I get interesting results.

In Terminal,

This produces a list of ping times in milliseconds on one second intervals. Pings should be less than 1 mS for a local base station.

pb pingThe PowerBook ping is pretty normal, most of them are 1 mS. There are other computers using this base station so that every so often a ping is delayed. No big deal.

p2p pingWhen the PowerBook is set to create a network and the iMac connects to it, the result is about the same but the ping is about 0.5 mS.

imac pingHowever, when the iMac pings the AEBS, something is clearly amiss. I'll get 6 good pings and then the thing goes to hell for the next 4 pings. This pattern repeats every 10 seconds.

imac pingI finally connected up an Airport Express Base Station as a WDS remote at my desk and wired it to the iMac via Ethernet, then turned the Airport in the iMac off. All is now well. The ping is a little longer than the best direct ping, but the really bad ones are gone.

imac pingWith the iMac Airport active, the story becomes a little clearer when the ping rate is increased by 100x. The data is no longer undersampled. The ping appears to stop for almost 300 mS. Then, all of the previously delayed pings come back all at once. Since each subsequent ping started 10 mS later, the delay appears to be linearly decreasing. Once the burst of pings returns, the cycle starts over for a total of about 4 seconds. Then it is good for 6 seconds and this pattern starts again.

Update, 5 Apr 06, OS X 10.4.6

imac ping with 10.4.6The iMac Airport problem has been resolved as of 10.4.6. There was apparently a software problem after all. The SpeakEasy result is now a stable 4770 down, 356 up. Pings are good and the download rates are as stable as any other Mac in the house at the same time. The computer also connects to a new Airport network quickly and reliably.

imac download with 10.4.6The thing that started it all is fixed too. My download rates via the cable modem run steadily at the maximum rate that the cable modem can provide.

Airport Dropouts

Update, Dec 22 2006

Since the Airport Client Monitor won't run on an Intel Mac yet I had not been able to monitor the Airport performance vs time. I had noticed some odd characteristics at times (including complete dropouts) but couldn't get a good look at what was going on. AP Grapher received an update to Universal binary and it became pretty clear that the iMac Airport implementation is still having difficulties. At times, often when the strength of the AP signal changes abruptly, the iMac just looses connection to the AP base station. This all started on Dec 19, 2006, just after I installed Security Update 2006-008.

30 minute plot

This eyechart is a plot of 30 minutes of AP Grapher data. The white line is the data rate, 54 Mb/sec is at the top. The green line is the analog signal strength of the Airport connection and the red line is the noise level. If a laptop is held right up against an Airport Base Station, the green line will exceed 3/4 scale. Small disturbances in the green line usually result in a larger disturbance of data rate in the next few seconds. Near the 90% point on the timeline above, the link died completely and had to be manually restored.

ap diedThis is a blow up of the death event. The small dip in the green signal strength trace (about 1/3 the way across) was my son walking into the kitchen which is the shortest path between the AP base station and the iMac. The path is about 30' long through a wall and bookshelves. When I got up from my desk, the link died. I assume that my body created another reflection which confused the Airport in the iMac and it gave up.

powerbook referenceThis plot is from a PowerBook G4 that was running at the same time in nearly the same location as the Intel iMac. Through all of these disturbances, the signal strength was fairly constant and the data rate was also constant. The PowerBook rarely negotiates a 54 Mb/sec link with the Airport, but at least it is stable.

It is pretty clear that the Airport implementations in these two computers is different and that they respond quite differently to similar conditions. The iMac seems to renegotiate the data rate much more often than the PowerBook. It also appears to give up more easily. What is not clear is that either one of them is "broken", they are different.

microwave_drop_out.gifOne mystery has been solved. We did get a new microwave oven because the old one blew up, quite literally. The old one had absolutely no impact at all on the Airport connection, the new one does have impact. This is a plot during a period that the new microwave was on. The data rate crashed but the link hung on for awhile before the airport modems lost it and the connection died completely. Through this whole event, the PowerBook sailed on through with nary a sign of difficulty.

died_again.gifAfter I reconnected to the Airport, the link then died again abruptly WITHOUT the microwave oven on. Clearly there is more going on than just some interference from a microwave oven.

Later, I thought that I had found the problem. For some reason, the iMac is refusing to reconnect automatically to a network once it loses it's connection. All seems to be set up correctly in the Network Preference Pane, it just isn't working. If I force a base station off line via a reset, the iMac never reconnects but the PowerBook does. I did try deleting all of the networks in the list and then adding them again... no joy. The PowerBook reconnects as soon as the base station comes back on line, the iMac just sits there with no bars.

The base stations were set to WEP so that my son could connect with his Nintendo DS. I reset the stations to WPA2, deleted the WEP entries from the Preferred Networks list and added the stations again as WPA2. It seemed to be better for a short while and then went back to it's bad ways. I've simply turned the internal Airport off and configured an Airport Express Base Station as a WDS remote and then connected it via an Ethernet cable to the iMac. At least the connection is stable now even with the microwave oven running. I've concluded that, unlikely as it seems, the Security Update 2006-008 messed with the Airport and I am awaiting a software update from Apple.

Update, January 13, 2007

It's been a little over 3 weeks since the Airport in the iMac began to act up and it hasn't become any better using the internal Airport hardware. The external AXBS has worked fine, not a single dropped connection, however the AXBS does suffer from interference from the new microwave oven, it stays connected with reduced data rates.

I simply bit the bullet and pulled a CAT6 cable into my office. I found a path for 125' of CAT6 cable that is unobtrusive and didn't require any holes punched in any walls. I also looped it by the living room TV (with a CAT6 coupler in place) in case I ever decide to install an ethernet connected TV device such as the new Apple TV. I expect no trouble now. Wires are good.

Update, January 27, 2007

Apple has released the Airport 2007-001 update for Intel Macs. The description indicated some bug fixes and there were early reports that it fixed the Airport dropouts in some cases so I tried it. I can't say that the dropouts are completely gone as I have had one. However, the microwave oven does not force a disconnect as it used to every time. The data rate goes in the dumpster, but at least it stays connected.

These kinds of disconnect are most common on the Core 2 Duo machines, however my Core Duo seemed to act in much the same way. It is certainly better now, but maybe not totally solid. And now, I don't use the Airport anyway. Wires are still good.

Update, Mar 9, 2007

The Airport 2007-001 update didn't quite work completely well. However, the Airport 2007-002 update seems to have made an improvement. The microwave oven still kills the data rate pretty much dead, but there are no more disconnects. Further, when downloading from a site that is faster than my cable modem, I can get steady rates of 748 kBytes/sec as my cable modem provider has bumped my data rates to just under 6 MBits/sec. I have not seen such steady wireless data rates since the Security Update 2006-008 came out and hosed things up.

A note about signal strength indications

MacStumbler is a freeware tool that reports strengths of all of the non-closed networks that are in range. It reports in numbers that vary significantly between the iBook and the powerbook and so it is a relative indicator that is valid for comparison only on one particular computer.

AP Grapher is a freeware tool for plotting the strength of a network as a function of time. AP Grapher has received a recent (Dec 06) update to Universal Binary along with some performance enhancements. It now does all the stuff that the Airport Client Monitor does as well as what MacStumbler does in one tool. The most valuable part is the time domain plot of Airport link characteristics. The plot will record data for as much as half an hour.

The Airport Management Tools are available from a link on this page. One of these tools is the Airport Client Monitor which allows you to monitor the signal strength and noise level of your Airport link in real time. The Airport Management Utility allows you to monitor the strength or noise level of ALL of the users of a given base station. This tool also reports the received strength between a WDS main and remote station. As far as I know, this is the ONLY way to evaluate the proper placement of WDS stations besides the old cut and try method.

The Airport Management Tools normally only ship with the Airport configured for DC power over the Ethernet line, but they may be useful to individuals as well. As of Dec 06, the Airport Management Tools do not run on Intel processor equipped Macs.

bars_vs_data_rate.gifSomewhere along the line, Apple changed the way that the Airport strength indicators worked. Before some version of Tiger, the strength indicator was really determined by the analog signal strength of the RF signal received by the Airport card. After the change, the strength indicator then indicted the net data rate, which is really what matters anyway. This plot shows the strength indicators (bars) versus the data rate as reported by AP Grapher 1.0. Up to 4 bars are shown on the Airport icon in the menu bar, up to 15 bars are shown in the Internet Connect window. These strength indicators exactly track the data rate, however, there is some filtering going on as the indicators take about 20 seconds of a constant data rate to settle to a stable value after a change in the data rate.

The PowerBook G4 curve covers the full range of data rate as I could move around and find locations where the rate was stable at all values. I could only get a limited range of rates on the iMac because it is somewhat less than portable and I had to wait for the each rate to just happen and be stable. The lower rates would happen on the iMac, but there were not stable long enough to get the indicators to settle down. Over the ranges characterized, the relationship between the data rate and the indicators is identical on the iMac and the PowerBook.

pg_client_monitor_plot.jpgThis plot shows the data rate (bottom) curve and the strength and noise indicators (top graph) as reported by Apple's Airport Management Tools running on the PowerBook. I can't run this tool on the iMac because it doesn't work on the Intel hardware.

pb_apgrapher_plot.gifThis is the same data as displayed by AP Grapher. The same things happened at the same time so that it would appear that AP Grapher can be used to replace the non-functional Airport Client Monitor on the iMac.

Wired Network Performance

Wireless networks are good for portability and flexibility, but not so good for reliability and speed. The paragraphs above are some examples of the pitfalls of wireless networks. I also ran some tests of wired network performance as compared to a few wireless network configurations to highlight the differences. The wired connection tests include all three popular flavors of Ethernet and FireWire 400.

The tool that I used this time is the AJA Kona Speed Test utility. This tool is supplied by AJA to test for disk access speeds to support their high end video capture hardware. It basically writes a big file and measures the average data rate and then reads it back and measures the data rate again. It can be used to measure network drives if a preference is set to allow it.

I used it to test the speeds from an iMac Core Duo to:

I tested the read and write speeds of these disks via: (where appropriate)

Internal Disk Speed Tests

The first pass was to measure the speed of each computer's internal disk. All of these computers use the ATA disk protocol. As expected, the 7200 rpm disks are faster than the 4200 rpm disks. One would expect that the read and write speeds would be identical. They were not and this tendency existed over several tests. In all cases, the AJA Kona Speed test tool was using a 128 MB test file.

Internal Disk Speeds
Computer Disk Speed
Write Speed
Read Speed
iMac Core Duo 7200 37.7 37.9
15" 1.25 GHz G4 AlBook 4200 18 18.6
12" 1.07 GHz iBook G4 4200 18.2 13.7
12" 500 MHz iBook G3 4200 18.5 15.4

This makes sense, the computer with the faster disk has better disk performance. The rate between the three slower disks is about the same over quite a wide range of CPU performance so that CPU limitations are not indicated. The slower disks did not perform as well in proportion to their speed. This is probably due to the large capacity of the iMac's disk due to it's physical size. It is a 3.5" drive, the others are 2.5" drives. There is more data per track on the larger disk so that it has to seek less often and therefore can produce a higher sustained rate that just it's rotational speed would indicate.

FireWire Tests

The same computers were used to evaluate FireWire performance. In each case, the iMac was used to run the test software and the other computers were mounted in Target Disk Mode using FireWire 400.

Speeds In Target Disk Mode
Computer Disk Speed
Write Speed
Read Speed
15" 1.25 GHz G4 AlBook 4200 8.6 14.7
12" 1.07 GHz iBook G4 4200 8.7 12.7
12" 500 MHz iBook G3 4200 7.5 8

In these cases, the rates are slower than the path from a computer's own CPU and its own disk. They were also more asymmetrical. The iBook is noticeably slower than the faster computers so that CPU performance in Target Disk Mode may be a factor.

Tests to the External FireWire Disk

Each computer was then tested with a OWC Neptune External FireWire Disk connected via FireWire. In this case, the speed of the internal disk doesn't matter as it is not being tested. The evaluation is of the relative FireWire performance of each computer.

FireWire Speed to a 7200 rpm External FireWire Disk
Computer Write Speed
Read Speed
iMac Core Duo 32.5 37.2
15" 1.25 GHz G4 AlBook 36.9 35
12" 1.07 GHz iBook G4 35.5 36.2
12" 500 MHz iBook G3 30.9 32.8

This data indicates that the performance of each computer to an external FireWire disk is pretty much of a wash. To an infinitely fast disk, FireWire 400 ought to top out just past 40 MBytes/sec, but I've never seen that because my hardware just isn't fast enough.

Wireless Tests

For this test, just two runs were made, one with 802.11g and one with 802.11b. The G3 iBook was selected for this test because it is 802.11b native. The iBook G4 was also tested to the network which is enabled for 802.11b and 802.11g service but the iBook G3 was put to sleep so that the network would not see it and slow down while it was servicing the 802.11g client.

Speeds via a Wireless Network
Computer Connection Protocol Write Speed
Read Speed
iBook G3 802.11b 0.3 0.3
iBook G4 802.11g 1.2 1.2

These are average rates with the typical wireless network dropouts and slowdowns. The peak rates were a little higher, but not by much. 11g is about 4x faster during this test, in the best case, 5x could be expected. In any event, this is FAR lower than the 54 Mbits/sec advertised for 802.11g which would be about 4x faster yet. Further, this test is about half of the rate that was characterized in the wireless tests at the top of this page.

Ethernet Speed Tests

These tests were done two different ways. In all cases, the iMac was used to mount the target computer's disk and the test was run. The first method was with the iMac connected to the other computer via a NetGear GS605 GigaBit Ethernet switch. Each of the three laptops has a different Ethernet capability which forced the maximum Ethernet speed. The second method was to use the iMac/PowerBook paring each time, but through three different Ethernet connections, the NetGear switch, an Airlink 100BaseT switch and a no-name 10BaseT hub.

Ethernet Speed Tests to Various Laptops
Target Computer Ethernet Type Write Speed
Read Speed
15" 1.25 GHz G4 AlBook 1000BaseT 38 37.8
12" 1.07 GHz iBook G4 100BaseT 9.8 10.4
12" 500 MHz iBook G3 100BaseT 5.2 6.1

1000BaseT should be capable of about 100 MBytes/sec sustained transfer with no other Ethernet traffic on the wire. I couldn't see that because my disks aren't fast enough, the rates to the PowerBook were similar to the FireWire rates. The rate to the iBook G4 is about right for 100BaseT. The rate to the iBook G3 is less than expected, probably due to some CPU based limitation in dealing with Ethernet.

In order to force the Ethernet rate, the switch or hub connecting the computers was swapped out and a single pair of computers was used, the iMac and the Powerbook.

Ethernet Speed Tests With Various Network Speeds
Ethernet Type Write Speed
Read Speed
1000BaseT 38 37.8
100BaseT 9.7 10.3
10BaseT 0.8 0.8

1000BaseT rate is the same data as before, it cannot reach the network limit due to the slower disks. The 100BaseT rate is similar to the rate using the iBook G4 which is naturally limited to 100BaseT. At 10BaseT, things really slow down by roughly 10x. This is expected. This rate is barely higher than my cable modem's rate so that even if I still had a 10BaseT network, I would probably not notice it while surfing unless there was some other network traffic that wasn't going through the modem.

Other Network Issues

There are other issues that can impact network speeds, some relate to wireless access only, some are more general. These things can kill your network performance.

Bit Torrent

For those of us with highly asymmetrical internet connections, which is most of us, usage of a Bit Torrent client anywhere on your network can materially degrade the internet for everybody on your network. An asymmetrical connection is one where the upload and download rates are materially different. The ratio of my connection is current quite extreme, about 13:1. I get just under 6 Mbits/sec download, but only 0.45 MBit/sec upload rates. This asymmetry is usually not a problem as most of the bit traffic for a typical user is downstream from web based servers. The ISPs have figured out what little upload bandwidth that they can provide and not seriously impact the "average" user. However, if users in your house use services such as P2P file sharing using Bit Torrent, or any of the other similar protocols, then there can be steady upload traffic as well. This traffic can choke off the upload packets needed to keep download activities going as well. The typical symptom is that web pages will start to load and then just stop. If this happens often and pressing the reload button a couple of times tends to get pages going again, suspect that your upload pipe is clogged.

Other services can impact the internet experience as well:

Of these, Bit Torrent seems to be the worst offender because it can result in nearly constant traffic for hours or days on end. The others tend to be intermittent, they may go on for minutes at a time and then let up. Most Bit Torrent clients allow their upload rates to be throttled to control this mess, look in your client for something that indicates upload rates and set it at the minimum, less than 5kBits/sec if possible. If that doesn't cut it, then eliminate that service or restrict its hours of operation to times when other users in your household are not using the internet.

Wireless Interference

Wireless network interference is an entire subject unto itself, hence I have devoted an entire page to Airport Link Tips elsewhere on this site.

AirDisk Performance

At Macworld 2007, Apple introduced the latest version of the AirPort base station. This is the one is the more or less square case as opposed to the "UFO" style Airport Extreme Base Station. Besides supporting the faster 802.11n wireless protocol, it also allows a USB disk to be connected and shared network wide. Apple calls this feature an "AirDisk."

No one could expect that the data rate supported to a remotely connected disk would be as good as with the same disk connected locally, but when I first tried it, I was seriously dissapointed in the performance. I got only 1 MByte/sec transfer rates, about 5% of the capability of the same disk connected locally. Something had to be wrong.

As it turns out, something was wrong but it wasn't obvious. I had formatted that particular disk in HFS Extended Journaled because that was the default in Disk Utility when I initially formatted the disk. It turns out that journaling is not a good plan for remote disks. After the disks I intended to use with the Airport were reformatted without journaling, the performance improved to 6 MBytes/sec. This is approaching the maximum rate that one could expect for a 100BaseT wired connection to an AirDisk. User's coming in via 802.11n might get about the same net rate as 802.11n supports a theoretical rate of about 200 Mbits/sec. By the time that wireless overhead is figured in the real life rates will be about half of that and the actual data should be similar to 100BaseT.

I tried the AirDisk feature with two disks, a 750 GB Seagate 7200 RPM USB drive and a 500 GB 7200 RPM LaCie Porche Design drive. As can be seen from the table below, journaling had a very strong impact on the performance of the AirDisk.

AirDisk Performance Vs. Disk Format
Disk Connection Journaled Data Rate
Non-Journaled Data Rate
LaCie Direct 18 18
AirDisk 3 6
Seagate Direct 18 18
AirDisk 1 6

The AirDisk feature of the 802.11n base station also has some warts. FIrst, it never spins the disk down so that the spinning disk will eat electrical power all day and night. Most spinning disks consume on the order of 15 watts or so, which isn't a lot, but it does add up. Depending on your electrical rates, it could cost up to $0.05 a day to run an AirDisk.

Access permission to an AirDisk is via a password, there are several options for setting passwords, but as soon as a user has access to the AirDisk via the base station password, he has access to the whole thing. When passwords are set up by account, then each user gets a part of the disk and the other users can't even see the extance of that user. The base station will not recognize disks formatted with multiple partitions.

I have had considerable trouble keeping the disks mounted with the AirDisk feature using firmware version 7.2.1. Although I am not done testing yet, it appears that firmware version 7.1 tends to work better than the newer versions.

The printer sharing feature of the new base station is also not as robust as the older versions. If the printer is turned off (again, to save power) or disconnected, the base station may have to be restarted to recognize the printer. The older base stations would register the printer as soon as it was plugged in or turned on.

[ Previous Tip ] [ Mac Tips ] [ Next Tip ]

This page has been accessed hit counter times since Aug 22, 2004

© 2004-2007 George Schreyer
Created 22 Aug 2004
Last Updated September 13, 2007