RFC 1030
On testing the NETBLT Protocol over divers networks
RFC 1030: On Testing the NETBLT Protocol over Divers Networks 💖 It's from the Network Working Group, written by M. Lambert at the M.I.T. Laboratory for Computer Science, and dated November 1987 🎀 The header doesn't list a category, and it doesn't list any obsoletes or updates, so this is just a test-results write-up 💅
Status of This Memo ✨
okay so this is the vibe check 🫶 this RFC is about testing NETBLT over three networks with different bandwidths and round-trip delays. The results aren't finished yet, BUT what's in so far is seriously promising and backs up RFC-998's claim that NETBLT can deliver super high throughput over networks that are totally different from each other 🚀 Distribution of this memo is unlimited, so share away 💜
1. Introduction 🌸
so, NETBLT (NETwork BLock Transfer) is a transport level protocol made for zipping a huge pile of data between computers, fast ⚡ It's reliable and flow controlled, and it's designed to hit maximum throughput across a wide variety of networks, which is giving main character energy 👑 The protocol itself is specified in RFC-998, and this doc assumes you already know that spec, so go read it first bestie 📖
We tested over three different networks 🧚 The first was a 10 megabit-per-second Proteon Token Ring, which served as a "reference environment" to figure out NETBLT's best possible performance. The second was a 10 megabit-per-second Ethernet, which acted as an access path to the third network, the 3 megabit-per-second Wideband satellite network. Testing over the Ethernet let us account for Ethernet-caused behaviour in the NETBLT transfers that went over Wideband, which is honestly such a smart move 🙈 Each network's results get their own section. The last section has conclusions and further research directions, and the appendices list the test results in detail 🍓
2. Acknowledgements 🫶
Huge thanks to Bob Braden, Stephen Casner, and Annette DeSchon of ISI, who spent time analyzing and commenting on test results from the ISI end of the NETBLT Wideband network tests 💕 Bob Braden also ported the IBM PC/AT NETBLT implementation to a SUN-3 workstation running UNIX, so he's literally iconic 👑 Thanks also to Mike Brescia, Steven Storch, Claudio Topolcic, and others at BBN, who shared lots of useful info about the Wideband network and helped monitor it during testing 🦋
3. Implementations and Test Programs 💅
This section gives a quick rundown of the NETBLT implementations and test programs used in the testing 🎀 Right now NETBLT runs on three machine types: Symbolics LISP machines, IBM PC/ATs, and SUN-3s. The test results in this paper came from the IBM PC/AT and SUN-3 implementations. The IBM and SUN versions are very similar, and most of the differences are in timer and multi-tasking library implementations 🌈 The SUN implementation uses UNIX's user-accessible raw IP socket, so it is not implemented in the UNIX kernel 🔥
The test application does a simple memory-to-memory transfer of however much data you want 💖 All the data gets allocated by the application, handed to the protocol layer, and copied into NETBLT packets. So the results are pretty realistic, and with big enough buffering, disk-based applications could hit these numbers too ✨
The test app has a bunch of parameters you can tweak to change NETBLT's performance. The most important ones are:
- burst interval: the number of milliseconds from the start of one burst transmission to the start of the next burst transmission ⏱️
- burst size: the number of packets transmitted per burst 📦
- buffer size: the number of bytes in a NETBLT buffer. All buffers must be the same size, except the last one, which can be whatever size is needed to finish the transfer 🧁
- data packet size: the number of bytes in a NETBLT DATA packet's data segment 🍬
- number of outstanding buffers: the number of buffers that can be in transmission or error recovery at any given moment 🌸
Now throughput, which gets measured two ways 💜 First there's the "real throughput", which is what the user actually sees: the number of bits transferred divided by the time from program start to program finish. It's useful from the user's point of view, but for analyzing NETBLT's performance, another measurement is way more useful 🫧 That's the "steady-state throughput", which is the rate data gets transmitted as the transfer size approaches infinity. It ignores connection setup time, and (more importantly) it ignores the time spent recovering from packet-loss errors that happen after the last buffer in the transmission is sent out 😭 On networks with long round-trip delays (which means lots of outstanding buffers), that "late" recovery phase can add a huge chunk of time to the transmission, and that time doesn't reflect NETBLT's peak transmission rate. Every throughput listed in the test cases that follow is a steady-state throughput 🦋
4. Implementation Performance 🧚
This section covers the theoretical performance of the IBM PC/AT NETBLT implementation on both the sending and receiving sides 💅 The theoretical numbers were measured on two LANs: a 10 megabit-per-second Proteon Token Ring and a 10 megabit-per-second Ethernet. "Theoretical performance" means the performance you'd get if the sending NETBLT did nothing but transmit data packets, and the receiving NETBLT did nothing but receive them 🌈
Measuring the send side is pretty easy, honestly 💖 The sending NETBLT does very little beyond transmitting packets at a predetermined rate, so there are few, if any, factors that can change how fast it processes things ✨
With a Proteon P1300 interface on a Proteon Token Ring, the IBM PC/AT NETBLT implementation can copy a maximum-sized packet (1990 bytes, not counting protocol headers) from the NETBLT buffer into a NETBLT data packet, format the packet header, and transmit it onto the network in about 8 milliseconds ⚡ That works out to a maximum theoretical throughput of 1.99 megabits per second 🔥
With a 3COM 3C500 interface on an Ethernet LAN, the same implementation can transmit a maximum-sized packet (1438 bytes, not counting protocol headers) in 6.0 milliseconds, for a maximum theoretical throughput of 1.92 megabits per second 🎀
Receiving is way trickier, bestie 😭
Measuring the receive side is harder 🙈 All the timer management and message ACK overhead lands on the receiving NETBLT, so its processing speed can be slightly slower than the sender's (and that doesn't even count the demultiplexing overhead it takes to match packets with protocol handling functions and connections). How much slower depends on a few things, and the most important are: the length of the NETBLT buffer list, the number of data timers that may need to be set, and the number of control messages that get ACKed by the data packet. Almost all of this extra overhead is directly tied to the number of outstanding buffers allowed during the transfer 🫧 So the fewer outstanding buffers you have, the shorter the NETBLT buffer list, the faster the scan through that list, and the shorter the list of unacknowledged control messages 🍓
Assuming a single-outstanding-buffer transfer, the receiving NETBLT can DMA a maximum-sized data packet from the Proteon Token Ring into its network interface, copy it from the interface into a packet buffer, and finally copy the packet into the correct NETBLT buffer in 8 milliseconds 🌸 That's the same speed as the sender of data, which is so iconic 👑
Under the same conditions, the implementation can receive a maximum-sized packet from the Ethernet in 6.1 milliseconds, for a maximum theoretical throughput of 1.89 megabits per second 💜
5. Testing on a Proteon Token Ring 🦋
okay so the Proteon Token Ring they tested on was a 10 megabit-per-second LAN with about 40 hosts on it 🌸. The machines at either end of the transfer were IBM PC/ATs, each using a Proteon P1300 network interface. The Token Ring gives high bandwidth, low round-trip delay, and negligible packet loss ✨, which makes it a seriously chill place to debug. Packet loss, packets arriving out of order, and long round-trip times would all make debugging way harder 😭. The big network MTU (max 2046 bytes) is also a major part of the performance glow-up 🔥. Bigger packets take a bit longer to send (8 milliseconds per 2046 byte packet vs 6 milliseconds per 1500 byte packet), but there's less per-byte computational overhead, so throughput goes up somewhat 💅.
The fastest single-outstanding-buffer transmission rate was 1.49 megabits per second 🏆, achieved with a test case using these parameters:
transfer size 2-5 million bytes
data packet size
1990 bytes
buffer size 19900 bytes
burst size 5 packets
burst interval 40 milliseconds. The timer code on the IBM PC/AT
is accurate to within 1 millisecond, so a 40
millisecond burst can be timed very accurately.With only one outstanding buffer, the protocol ran in "lock-step" 🔒 (the receiver of data sends a GO, the sender sends data, the receiver sends an OK, followed by a GO for the next buffer). Every buffer ate up one round-trip delay's worth of overhead (between sending a buffer's last data packet and getting the OK for that buffer / GO for the next one) 😩. So a test with two outstanding buffers, which gives essentially constant packet transmission, should have come out with higher throughput 🌈.
A second test, this time with two outstanding buffers, was run with the same parameters except for a burst interval of 43 milliseconds 🎀. The highest throughput recorded was 1.75 megabits per second 🔥. That's 95% efficiency ✨, since 5 1990-byte packets every 43 milliseconds gives a maximum theoretical throughput of 1.85 megabits per second 🧮. The jump over the single-buffer run happens because, with two outstanding buffers, there's no round-trip-delay lag between buffer transmissions, so the sending NETBLT can transmit constantly 🚀. The P1300 interface can transmit and receive at the same time, so no packets were dropped due to collision on the interface 🙌.
As mentioned earlier, the minimum transmission time for a maximum-sized packet on the Proteon Ring is 8 milliseconds ⏱️. So you'd expect the best double-buffered throughput at a burst interval of 8 milliseconds times 5 packets per burst, or 40 milliseconds 🤔. That would let the sender transmit bursts with no "dead time" in between 💖. Unfortunately, the sender has to take time to process incoming control messages, which usually forces a 2-3 millisecond gap between bursts and drags throughput down 😤. With a 43 millisecond burst interval, the incoming packets get processed during that 3 millisecond-per-burst "dead time", which makes the protocol more efficient 💅✨.
6. Testing on an Ethernet 🌊
The network for this round of tests was a 10 megabit per second Ethernet with about 150 hosts 🌐. The machines at either end of the NETBLT connection were IBM PC/ATs using 3COM 3C500 network interfaces 💻. Like the Proteon Token Ring, the Ethernet gives high bandwidth with low delay 🌸. Unfortunately, the specific Ethernet used (MIT's infamous Subnet 26) is known for being pretty noisy 📢😬. On top of that, the 3COM 3C500 interfaces are kinda basic, with only a single hardware packet buffer for both sending and receiving 🥲. That gives them an annoying tendency to drop packets under heavy load 💀. Put all of that together and analysing protocol performance was a lot harder than on the Proteon Ring 🙈.
The fastest single-buffer transmission rate was 1.45 megabits per second 🍓, achieved with a test case using these parameters:
transfer size 2-5 million bytes
data packet size
1438 bytes (maximum size excluding protocol
headers).
buffer size 14380 bytes
burst size 5 packets
burst interval 30 milliseconds (6.0 milliseconds x 5 packets).A second test, identical to the first except for two outstanding buffers instead of one, gave substantially lower throughput (994 kilobits per second) 📉, with a large number of packets retransmitted (10%) 😱. The retransmissions happened because the 3COM 3C500 has only one hardware packet buffer and can't hold a transmitting packet and a receiving packet at the same time 🙃. With two outstanding buffers, the sender transmits constantly, so when the receiver tries to send a packet, its interface's receive hardware goes deaf to the network, and any packets the sender is transmitting at that moment get lost 🙉. The same problem happens in reverse with control messages from the receiver to the sender, but there are so few of them, and the retransmission algorithm is so redundant, that control message loss barely dents performance 🫶.
When the burst interval went from 30 milliseconds per 5 packet burst to 45 milliseconds per 5 packet burst, a third as many packets were dropped, and throughput climbed to 1.12 megabits per second 📈✨. The likely reason is that the longer burst interval left more dead time between bursts, so the receiver's interface was less likely to be deaf to the net while the sender was sending a packet 🧠. The most interesting part: when the same test ran on a special Ethernet with only the two NETBLT machines attached, no packets were dropped once the burst interval rose above 40 milliseconds per 5 packet burst 🎉. The improved performance was doubtless due to the absence of extra network traffic 👑.
7. Testing on the Wideband Network 🛰️
okay bestie, this section is about the Wideband network, a satellite-based network with ten stations all competing for a raw satellite channel bandwidth of 3 megabits per second 🛰️✨. The tests caused some big changes to the NETBLT spec and implementation, so we're gonna walk through the major ones, plus the results and problems that forced them 💅.
The Wideband network is literally the perfect place to test NETBLT 🦋. First, it has an extremely long round-trip delay of 1.8 seconds, which is a great test of NETBLT's rate control and multiple-buffering skills 🎀. Rate control lets the packet transmission rate be regulated separately from the maximum amount of outstanding data, so you get flow control plus very large "windows" 🌸. Multiple buffering means data can keep moving while earlier data is waiting on retransmission and later data is getting prepped to send 🧚. On a network with a long round-trip like this, the alternative "lock-step" approach would need a 1.8 second gap between each buffer transmission, which would tank performance 😭.
Throughput is another iconic thing about this network 💖. The raw bandwidth is 3 megabits per second, but at the time of these tests, fully 2/3 of that was eaten by low-level network overhead and hardware limitations 💀. (A detailed analysis of the overhead appears at the end of this document.) That leaves just over 1 megabit per second. NETBLT can run way faster than that, so testing over the Wideband net let us measure how well NETBLT uses a super high percentage of the bandwidth that's actually available 🔥.
Finally, the Wideband net has some interesting packet reorder and delay characteristics, which make it a great test of how NETBLT handles those problems 🙈.
Testing went in several phases 🍓. Phase one used source-routed packets along a path from an IBM PC/AT on MIT's Subnet 26, through a BBN Butterfly Gateway, over a T1 link to BBN, onto the Wideband network, back down into a BBN Voice Funnel, and onto ISI's Ethernet to another IBM PC/AT 🛤️. It went pretty slowly because of gateway software and source-routing bugs 😩. Once a connection was finally established, the best throughput we recorded was approximately 90K bits per second 🥲.
Several problems dragged the throughput down 📉. First, the gateways at both ends were forwarding packets onto their LANs faster than the IBM PC/AT's could accept them. The 3COM 3C500 interface wouldn't have time to re-enable input before another packet arrived from the gateway 🙃. Even with bursts of size 1, spaced 6 milliseconds apart, the gateways would aggregate groups of packets coming from the same satellite frame and send them faster than the PC could receive them. The obvious result was lots of dropped packets and degraded performance. Also, the half-duplex nature of the 3COM interface caused incoming packets to be dropped whenever packets were being sent 😤.
Drops on the sending NETBLT side, caused by the long interface re-enable time, were reduced by packing as many control messages as possible into a single control packet, instead of putting only one message in each. That cut control packets down to one per buffer transmission, which the PC could handle 🫶. Specifically, OK(n) messages got combined with GO(n + 1) messages, so two control packets couldn't arrive too close together to both get received 🧠.
Performance loss from dropped control packets got a glow-up too 🌈. The implementation switched to a highly redundant control packet transmission algorithm. Control messages now live in a single long-lived packet, where ACKed messages keep getting bumped off the head and new messages get added at the tail. Every time a new message needs to go out, any unACKed old messages get sent again as well. The sending NETBLT, which receives these control messages, is tuned to ignore duplicates with almost no overhead 💅. This redundancy puts little reliance on the NETBLT control timer, which further reduces performance degradation from lost control packets 💜.
The receiving NETBLT couldn't be completely protected from dropped packets, but some implementation changes helped a bit 🩵. Data packets from the sending NETBLT are guaranteed to be transmitted by buffer number, lowest number first 🔢. In some cases, that let the receiving NETBLT make retransmit-request decisions for a buffer N, if packets for N were expected but none had been received by the time packets for a buffer N+M arrived. This optimization was kinda complicated, but it improved NETBLT's performance in the face of missing packets 🙌. Unfortunately, the dropped-packet problem stuck around until the NETBLT implementation was ported to a SUN-3 workstation. The SUN handles incoming packets quite well, dropping only 0.5% of the data packets (as opposed to the PC's 15 - 20%) 🌞.
Reordering and delay were the next boss 🥊. The NETBLT implementation had been "optimized" to generate retransmit requests as soon as possible, if possible without relying on expiration of a data timer. For instance, when the receiving NETBLT got an LDATA packet for a buffer N and other packets in buffer N hadn't arrived, the receiver would immediately generate a RESEND for the missing packets. Similarly, under certain circumstances, the receiver would generate a RESEND for buffer N if packets for N were expected and hadn't arrived before packets for a buffer N+M. Obviously, packet reordering made these "optimizations" generate retransmit requests unnecessarily 🙃. In the first case, the implementation was changed so it no longer generates a retransmit request on receipt of an LDATA with other packets still missing in the buffer. In the second case, a data timer was set with an updated (and presumably more accurate) value, hopefully allowing any reordered packets to arrive before timing out and generating a retransmit request ⏰.
Delay is the tricky part 🐌. It's hard to accommodate Wideband packet delay in the NETBLT implementation. Packet delays tend to occur in multiples of 600 milliseconds, due to the Wideband network's datagram reservation scheme. A timer value calculation that used a fixed variance on the order of 600 milliseconds would cause performance degradation when packets were lost. On the other hand, short fixed variance values wouldn't react well to the long delays possible on the Wideband net 😬. The solution was an adaptive data timer value calculation algorithm. It maintains an average inter-packet arrival value and uses that to determine the data timer value. If the inter-packet arrival time increases, the data timer value lengthens 🧘.
Testing then moved along between NETBLT implementations on a SUN-3 workstation and an IBM PC/AT 🖥️. The arrival of a Butterfly Gateway at ISI eliminated the need for source-routed packets, and some performance improvement was expected too, since the Butterfly Gateway is optimized for IP datagram traffic 🚀.
To put the best Wideband network results in context, here's a short analysis showing the best throughput you could expect on a fully loaded channel 📊. Again, a detailed analysis of the numbers that follow appears at the end of this document.
The best possible datagram rate over the current Wideband configuration is 24,054 bits per channel frame, or 3006 bytes every 21.22 milliseconds. Since the transmission route begins and ends on an Ethernet, the largest amount of data transmissible (after accounting for packet header overhead) is 1438 bytes per packet. This translates to approximately 2 packets per frame. Since we want to avoid overflowing the channel, we should transmit slightly slower than the channel frame rate of 21.2 milliseconds. We therefore came up with a best possible throughput of 2 1438-byte packets every 22 milliseconds, or 1.05 megabits per second 🧮.
Heads up: because of possible software bugs in either the Butterfly Gateway or the BSAT (gateway-to-earth-station interface), 1438-byte packets got fragmented before transmission over the Wideband network, which caused packet delay and poor performance 🫠.
The best throughput was achieved with the following values 🏆:
transfer size 500,000 - 750,000 bytes
data packet size
1432 bytes
buffer size 14320 bytes
burst size 5 packets
burst interval 55 millisecondsSteady-state throughputs ranged from 926 kilobits per second to 942 kilobits per second, which is approximately 90% channel utilization 🤩. The amount of data transmitted should have been an order of magnitude higher, to get a longer steady-state period. Unfortunately, at the time we were testing, the Ethernet interface of ISI's Butterfly Gateway would lock up fairly quickly (in 40-60 seconds) at packet rates of approximately 90 per second, forcing a gateway reset 💀. Transmissions therefore had to take less than that amount of time. This problem has reportedly been fixed since the tests were conducted 🙌.
To test the Wideband network under overload conditions, we tried several tests at rates of 5 1432-byte packets every 50 milliseconds 🌪️. At this rate, the Wideband network ground to a halt as four of the ten network BSATs immediately crashed and reset their channel processor nodes 😱. Apparently, the BSATs crash because the ESI (Earth Station Interface), which sends data from the BSAT to the satellite, stops its transmit clock to the BSAT if it runs out of buffer space. The BIO interface connecting BSAT and ESI doesn't tolerate this clock-stopping and typically locks up, forcing the channel processor node to reset. A more sophisticated interface, allowing faster transmissions, is being installed in the near future ✨.
8. Future Directions 🔮
okay so we still need more testing over the Wideband Network to get the full picture of NETBLT's performance 🛰️. Once the Butterfly Gateway Ethernet interface lockup problem (the one we talked about earlier, iykyk) is fixed, we want to send 10 to 50 million bytes to get accurate steady-state throughput numbers 📈. We also want to run several NETBLT processes in parallel, each tuned to take a fraction of the Wideband Network's available bandwidth. Hopefully this will show whether or not burst synchronization across different NETBLT processes causes network congestion or failure 👀. And once the BIO BSAT-ESI interface gets upgraded, we'll want to go for higher throughputs and more hardware stability under overload conditions 💪.
As for future research, one big area needs exploring 🧠. A series of algorithms needs to be developed so NETBLT can dynamically select and control its transmission parameters (burst size, burst interval, and number of outstanding buffers). Ideally this dynamic control won't need any info from outside sources like gateways. Instead, NETBLT processes will use end-to-end information to make transmission rate decisions when channels are noisy and the network is congested 🫶. Some research on dynamic rate control is happening right now, but much more work needs doing before the results can be folded into NETBLT 🌸.
I. Wideband Bandwidth Analysis ✨
okay this part is the math tea ☕. Even though the raw bandwidth of the Wideband Network is 3 megabits per second, only about 1 megabit per second of it is actually available for sending data 😭. The big overhead comes from the channel control strategy (which uses a fixed-width control subframe based on the maximum number of stations sharing the channel) and the low-performance BIO interface between BBN's BSAT (Butterfly Satellite Interface) and Linkabit's ESI (Earth Station Interface). Higher-performance BSMI interfaces are coming soon to all Wideband sites, which should improve the available bandwidth 🎀.
Bandwidth on the Wideband network is split into frames, and each frame has multiple subframes 🦋. A frame is 32768 channel symbols, at 2 bits per symbol. One frame is available for transmission every 21.22 milliseconds, so the raw bandwidth works out to 65536 bits / 21.22 ms, or 3.081 megabits per second.
Each frame has two subframes: a control subframe and a data subframe. The control subframe is split into ten slots, one per earth station, and control information takes up 200 symbols per station. Since the communications interface between BSAT and ESI only runs at 2 megabits per second, there has to be a padding interval of 1263 symbols between each slot of information. That brings the total control subframe size up to 1463 symbols x 10 stations, or 14630 symbols. The data subframe then has 18138 symbols available. The maximum datagram size is currently expressed as a 14-bit quantity, which drops the maximum amount of data in a frame further to 16384 symbols. After header information is taken into account, this value drops to 16,036 symbols. At 2 bits per symbol, using a 3/4 coding rate, the actual amount of usable data in a frame is 24,054 bits, or about 3006 bytes. So the theoretical usable bandwidth is 24,054 bits every 21.22 milliseconds, or 1.13 megabits per second 🌈.
Since the NETBLT implementations are running on Ethernet LANs gatewayed to the Wideband network, those 3006 bytes per channel frame of usable bandwidth turn into two maximum-sized (1500 bytes) Ethernet packets per channel frame, or 1.045 megabits per second 💅.
II. Detailed Proteon Ring LAN Test Results 💍
okay here are some of the test results from running NETBLT between two IBM PC/ATs on a Proteon Token Ring LAN 🪙. Here's what the table headers mean:
- BS/BI: burst size in packets and burst interval in milliseconds
- PSZ: number of bytes in DATA/LDATA packet data segment
- BFSZ: number of bytes in NETBLT buffer
- XFSZ: number of kilobytes in transfer
- NBUFS: number of outstanding buffers
- #LOSS: number of data packets lost
- #RXM: number of data packets retransmitted
- DTMOS: number of data timeouts on receiving end
- SPEED: steady-state throughput in megabits per second
BS/BI PSZ BFSZ XFSZ NBUFS #LOSS #RXM DTMOS SPEED
5/25 1438 14380 1438 1 0 0 0 1.45
5/25 1438 14380 1438 1 0 0 0 1.45
5/30 1438 14380 1438 1 0 0 0 1.45
5/30 1438 14380 1438 1 0 0 0 1.45
5/35 1438 14380 1438 1 0 0 0 1.40
5/35 1438 14380 1438 1 0 0 0 1.41
5/40 1438 14380 1438 1 0 0 0 1.33
5/40 1438 14380 1438 1 0 0 0 1.33
5/25 1438 14380 1438 2 0 0 0 1.62
5/25 1438 14380 1438 2 0 0 0 1.61
5/30 1438 14380 1438 2 0 0 0 1.60
5/30 1438 14380 1438 2 0 0 0 1.61
5/34 1438 14380 1438 2 0 0 0 1.59
5/35 1438 14380 1438 2 0 0 0 1.58
5/25 1990 19900 1990 1 0 0 0 1.48
5/25 1990 19900 1990 1 0 0 0 1.49
5/30 1990 19900 1990 1 0 0 0 1.48
5/30 1990 19900 1990 1 0 0 0 1.48
5/35 1990 19900 1990 1 0 0 0 1.49
5/35 1990 19900 1990 1 0 0 0 1.48
5/40 1990 19900 1990 1 0 0 0 1.49
5/40 1990 19900 1990 1 0 0 0 1.49
5/45 1990 19900 1990 1 0 0 0 1.45
5/45 1990 19900 1990 1 0 0 0 1.46
5/25 1990 19900 1990 2 0 0 0 1.75
5/25 1990 19900 1990 2 0 0 0 1.75
5/30 1990 19900 1990 2 0 0 0 1.74
5/30 1990 19900 1990 2 0 0 0 1.75
5/35 1990 19900 1990 2 0 0 0 1.74
5/35 1990 19900 1990 2 0 0 0 1.74
5/40 1990 19900 1990 2 0 0 0 1.75
5/40 1990 19900 1990 2 0 0 0 1.74
5/43 1990 19900 1990 2 0 0 0 1.75
5/43 1990 19900 1990 2 0 0 0 1.74
5/43 1990 19900 1990 2 0 0 0 1.75
5/44 1990 19900 1990 2 0 0 0 1.73
5/44 1990 19900 1990 2 0 0 0 1.72
5/45 1990 19900 1990 2 0 0 0 1.70
5/45 1990 19900 1990 2 0 0 0 1.72III. Detailed Ethernet LAN Testing Results 🌸
okay same vibe, but now it's NETBLT between two IBM PC/ATs on an Ethernet LAN 🖥️. See the previous appendix for the table header definitions, bestie.
BS/BI PSZ BFSZ XFSZ NBUFS #LOSS #RXM DTMOS SPEED
5/30 1438 14380 1438 1 9 9 6 1.42
5/30 1438 14380 1438 1 2 2 2 1.45
5/30 1438 14380 1438 1 5 5 4 1.44
5/35 1438 14380 1438 1 7 7 7 1.38
5/35 1438 14380 1438 1 6 6 5 1.38
5/40 1438 14380 1438 1 48 48 44 1.15
5/40 1438 14380 1438 1 50 50 38 1.17
5/40 1438 14380 1438 1 13 13 11 1.28
5/40 1438 14380 1438 1 7 7 5 1.30
5/30 1438 14380 1438 2 206 206 198 0.995
5/30 1438 14380 1438 2 213 213 198 0.994
5/40 1438 14380 1438 2 117 121 129 1.05
5/40 1438 14380 1438 2 178 181 166 0.892
5/40 1438 14380 1438 2 135 138 130 1.03
5/45 1438 14380 1438 2 57 57 52 1.12
5/45 1438 14380 1438 2 97 97 99 1.02
5/45 1438 14380 1438 2 62 62 51 1.09
5/15 512 10240 2048 1 6 6 4 0.909
5/15 512 10240 2048 1 10 11 7 0.907
5/18 512 10240 2048 1 11 11 8 0.891
5/18 512 10240 2048 1 5 5 9 0.906
5/19 512 10240 2048 1 3 3 3 0.905
5/19 512 10240 2048 1 8 8 7 0.898
5/20 512 10240 2048 1 7 7 4 0.876
5/20 512 10240 2048 1 11 12 5 0.871
5/20 512 10240 2048 1 8 9 5 0.874
5/30 512 10240 2048 2 113 116 84 0.599
5/30 512 10240 2048 2 20 20 14 0.661
5/30 512 10240 2048 2 49 50 40 0.638IV. Detailed Wideband Network Testing Results 🛰️
last table, and it's the Wideband one 🎉. These are some of the results from NETBLT between an IBM PC/AT and a SUN-3 over the Wideband satellite network. See the previous appendix for the table header definitions, periodt 💖.
BS/BI PSZ BFSZ XFSZ NBUFS #LOSS #RXM SPEED
5/90 1400 14000 500 22 9 10 0.584
5/90 1400 14000 500 22 12 12 0.576
5/90 1400 14000 500 22 3 3 0.591
5/90 1420 14200 500 22 12 12 0.591
5/90 1420 14200 500 22 6 6 0.600
5/90 1430 14300 500 22 9 10 0.600
5/90 1430 14300 500 22 15 15 0.591
5/90 1430 14300 500 22 12 12 0.590
5/90 1432 14320 716 22 13 16 0.591
5/90 1434 14340 717 22 33 147 0.483
5/90 1436 14360 718 22 25 122 0.500
5/90 1436 14360 718 22 25 109 0.512
5/90 1436 14360 718 22 28 153 0.476
5/90 1438 14380 719 22 6 109 0.533
5/80 1432 14320 716 22 56 68 0.673
5/80 1432 14320 716 22 14 14 0.666
5/80 1432 14320 716 22 15 16 0.661
5/60 1432 14320 716 22 19 22 0.856
5/60 1432 14320 716 22 84 95 0.699
5/60 1432 14320 716 22 18 21 0.871
5/60 1432 14320 716 30 38 40 0.837
5/60 1432 14320 716 30 25 26 0.869
5/55 1432 14320 716 22 13 13 0.935
5/55 1432 14320 716 22 25 25 0.926
5/55 1432 14320 716 22 25 25 0.926
5/55 1432 14320 716 22 20 20 0.932
5/55 1432 14320 716 22 17 19 0.934
5/55 1432 14320 716 22 13 14 0.942📦 where this page comes from
- ietf: rfc1030 1987-11
- 💖 girlified by claude-haiku-5-5 on : a language model's rewrite of the original, under its terms