Gigabit Media Converter Kit

Overview of the second device type "Deviceuseless "from the world of fiber-optic communication. Like the previous time, there will be several digressions into the field of the accompanying theory. And I apologize in advance that in some places the review will depart from a serious statement. Review of p.18 - I have to praise it. As I can, so I praise.

The overlooked device has a rather narrowapplication area. And not because it is bad or expensive - it’s just really highly specialized. Therefore, at first there will be a small and superficial excursion on the topic of optical networks. I will try to minimally repeat two (1.1) of my previous reviews on a similar topic.
"A couple of words" about optical networksOptical communication lines were very simple - transmitter, optical core, receiver. If you need a bi-directional exchange - put the second set (receiver, core, transmitter), only in the opposite direction.
This scheme lasted long enough, the reasonsthere is plenty to find. Moreover, for short lines (up to several kilometers or inside a building) such a scheme is still alive today. It is cheaper / easier to lay excess fiber than to complicate the equipment.
The need to skip more and moreTraffic caused investment in development, which, in turn, provided an increase in bandwidth. This increase is achieved in two ways.
First, the change in production technologyLasers made it possible to increase the speed of data transmission. From the original hundreds of megabits (and for the first, semi-experimental lines - even less), they quickly came to gigabits per second (with a bit rate on a line of 1.25 GHz), and now the standard for the average backbone is 10 Gigabits / s. Also often used 40G, but in most implementations it is the same 4x10G, only collected in one device. There are already free-selling solutions for 100 Gigabits, but they are expensive. VERY EXPENSIVE.
Secondly, instead of two veins per channel, the main lines now use technologies that allow several channels to be transmitted through one core.
The first reason for such a seal immediatelycoming to the head - the “extra” vein of optical fiber costs money. In reality, this only affects really long lines, since the increase in value is non-linear. For example, a cable with 16 fibers (down the drain) is one and a half times more expensive than four-fiber one. If the cable is designed for laying into the ground, then its jacket is even more expensive and the rise in price for quadrupling of fibers will be only 20 percent. Relatively speaking, the cost of an additional optical core will be around $ 20 per kilometer. This is at a cost of relatively cheap shell in the region of $ 300- $ 700.
The second reason is the desire to expand the existing ones.communication lines. If the cable has already been laid, then it is impossible to “add” an extra vein. It’s like a subway - it’s almost impossible to make changes to an already constructed line that are not covered by the original project. Neither a new station should be built in the middle, nor the length of the existing station should be extended. Just build a new line. And the life of the subway, that a good cable is big. During this time, several generations of active equipment can be replaced (cars - for metro or media converters - for optics). I would even put forward a somewhat speculative idea that the development of optical passive sealing systems arose with some delay precisely because it was first needed when there were already quite a few communication lines built and the required throughput began to massively exceed the originally calculated one. However, there are many reasons for this, for example, the initial high cost of lasers at 1550nm and lasers with a narrow emission band.
It is clear that if many similar colors are mixed in one carrier, then it is difficult to divide the result into separate components:
Because of this, the first systems used only two colors, which are easy to separate:According to this principle, the WDM type seal is built.(wavelength-division multiplexing), which provides transmission in both directions on the same fiber at very different wavelengths. It is arranged like this:Here you can draw almost direct analogywith radio waves when the transmitter and receiver use the same antenna. At the same time, it is difficult to emit and receive signals of the same frequency (wavelength) - a powerful output signal must be directed to the input. Technically, for optics is not so simple, because the signal has a clearly defined transmission direction. There is a solution that allows you to work in two directions at the same wavelength - the so-called optical circulator.They use all sorts of clever tricks with rotation of polarization, allowing to reduce the reflection to acceptable values. Outwardly, this device looks very simple:Inside the case is flooded with a small steel tube: with a fairly complex optical system. Everything would be fine, but this design is twice as expensive as the considered media converters and it is used relatively rarely.
On one channel, one fiber does not progresshas stopped. With the development and introduction of technology (both optical and in terms of the production of lasers), it became possible to transmit several channels in each direction in a single fiber.
A simpler CWDM G.694 standard.2 (multiplexing with sparse spectral separation) in practice provides up to 8 channels over a single fiber. Theoretically, there are 18 wavelength ranges (9 bidirectional channels):but not all are commonly used.
A slightly more expensive DWDM G.694.1 (dense spectral multiplexing) provides almost 30 times more dense channel placement (after 0.8 nm, instead of 20 nm).
All these standards use a single principle - somewhat close, but still suitable for the separation of colors in one medium. If you continue the "cat" style:The “tighter” wavelengths are packed in a channel, the more expensive emitters and separation systems become.
For assembly and separation of channels are usedspecial passive optical multiplexers with filter mirrors inside, allowing, depending on the design, or in turn to allocate one signal from a specific wavelength to the output:several boxes of this type:or immediately divided into many veins:using a big box:
As a rule, such devices are used on operator communication channels - for the end user, they are expensive.
The previously listed devices enabletransfer one fiber more data. But there are cases when it is enough to get more individual channels, even without increasing the total bandwidth. For ordinary Internet users, there are cheaper devices that allow one to simply divide one core into many taps (like a TV antenna splitter), without filtering by wavelengths. They are many times cheaper than filter multiplexers. Such devices allow to diversify the connection options for end users, depending on the needs and the wallet.
The classic version A with providing fiber to every consumer is very expensive and is used rarely or for short distances.
Option B is very often used by ISPs, only in most cases the terminal connection is copper.
Option C is the classic connection option.Internet users using PON technology (passive optical networks). The system uses time band division (approximately as in GSM telephony), with a main transmitter (hiding in the CO house and called OLT, Optical Line Terminal) and a bunch of user access points (ONT, optical network terminal or ONU, optical network unit) emit at certain times. To separate the directions (to the provider / user), two different wavelengths are used. In this scheme, several passive multiplexers can be used. For example, in my house one fiber comes to the front door, where it is divided into 8 veins, which have one cable of this type:go on the riser. On every second floor there is a switching box with a divider of 8.
There may be tricky cases with asymmetrical division of the fiber:
Usually, up to 64 users are connected to one fiber from the server operator to the house (but technically it is possible up to 128). The dividers themselves are for an arbitrary number of taps, from two:up to 64:with a completely affordable price from $ 10 to $ 150.
Unfortunately, PON also has its drawbacks. There are two main ones:
First, the terminal installed atA user is no longer just a router, he must be able to interact with the OLT and transmit data blocks at strictly defined points in time. Of course, the operator usually buys such devices by wagons, which is why they cost less. But all the same, it is noticeably more expensive than the usual sfp-module and, especially, the sockets for a regular twisted pair.
Secondly, on all users of a single fibertypically there is a 2 Gbit / s downlink bandwidth. If suddenly all 64 users are connected and they all start downloading something, each will have less than 30 Mbps. That, however, is not so little.
Devices for pairing optics and activeequipmentUnlike copper networks, where the cost of interfacing transformers is relatively small, optical transceivers are much more expensive even now. Therefore, they are almost always made in the form of individual devices or modules. The most common options are:
1) module for switches / routers. The first modules were focused on a specific device or, at best, on a group of devices from the same manufacturer. Here, for example, is the module for Avaya Cajun P330R:
20 years ago, such a switch cost as my then-10-year salary for 10.
Later began to produce universal modules. But, unfortunately, standardization here is not quite absolute. Not only is there a huge number of structurally different modules (GBIC, SFP, SFP +, XFP, XENPAK, X2, QSFP, QSFP +, QSFP28, CFP, CFP2, CFP4, CSFP, ...) and many well-known manufacturers like to block the use of "alien "Modules. The most common and cheap now is a SFP-type module providing a connection at 100 Mb / s or 1 GB / s. If there is an appropriate connector in the device, the use of an embedded module usually costs the least. In addition, in the case of the use of modules with the monitoring function (DOM / DDM), on the "smart" devices you can usually see the transmit / receive power and the temperature of the module. The latter is also useful because Air conditioners tend to break, and here temperature monitoring is free, easily screwed to the automatic monitoring system via SNMP. Strangely enough, many very expensive Cisco switches do not have a built-in sensor, or it gives OK / FAIL readings.
2) a separate "box" that provides the jointcopper network with optical - media converter. This is the device that will be tested. On the one hand, such a solution is quite versatile and independent - you can connect it at least to a switch, at least to any computer. Sometimes this is useful if there are no free slots for expansion modules on the switch or a local network is connected through some sealed device like this:
or such:
On the other hand, the media converter is more expensive than the embedded module and does not provide access to the signal parameters. It also takes up more space and requires a separate outlet.
In addition to the monolithic media converter, now there is also an option for use with a plug-in module:
But such devices will not work out for sureclassify - in terms of stuffing, this is a two-port switch, no matter how crazy it sounds. In terms of cost, it is slightly cheaper than a multi-port switch - only due to the absence of unnecessary connectors, LEDs and transformers. In case you need to buy a switch, it’s easier and cheaper to immediately find a switch with an SFP slot.
Significantly less (and, mainly, for short connections inside the building or server) use network cards with optical output, such as:or this:The first cards were with tightly mounted optical modules. Now, adapter cards are more common, in which there are no optics, but there is a slot for modules, as for switches:

Summarizing, we can say that the current trend is the use of separate more or less universal optical modules. Devices with tightly mounted optical output gradually die off.
Now back to the surveyed devices.
HTB-4100A and HTB-4100B media converters kit
First, take a description from the product page and analyze it by items
Compatible: Intel, Huawei, Transition Networks, Alcatel-lucent, Planet, TP-link, Extreme, Supermicro, Moxa, Ericsson, Brocade, Cisco, Aruba, Foundry, Mikrotik, HP, Ubiquiti, Enterasys, Zyxel, Trendnet, Netgear, IBM, F5 Networks, H3C, Avago, Customized Brands, Arista, Mellanox, Avaya, Dell Force10, Dell, Hirschmann, Open Switch, Finisar, Ciena, Juniper, SMC, Linksys, Allied, D-Link
"I know karate, aikido, judo ... and many morescary words "©. Quite a strange set of manufacturers. Based on this list - do not connect to realtek network cards? I think that we can safely ignore this item.
Model Number: HTB-4100AB
More precisely - a set of HTB-4100A and HTB-4100B
Fiber Type: Single Mode
Designed for normal single mode fiber. This is understandable, there is no other 20km.
Transmission Distance: 20KM
Theoretically, it will “break through” twenty kilometersfiber. So it is, or not - depends on many reasons, we will test. However, in most cases, such devices are used on substantially shorter lines.
Fiber Connector: SC
Currently the most common type of connector.
Power: Built in power
This is immediately apparent.
Main Features:
Complies with IEEE 802.3, IEEE 802.3u, IEEE802.3z, IEEE802.3ab, IEEE 802.3x
IEEE 802.3 is the foundation of Ethernet standards. Why specify a group name is not entirely clear, because there is nothing else, from coax to 100Gb lines, from optics to POE.
IEEE 802.3u - one hundred megabit ethernet over twisted pair and optics.
IEEE 802.3z - gigabit, as a rule - on optics. Because of the 8b / 10b coding, the actual speed in the 1.25bitG / s line.
IEEE 802.3ab - gigabit over twisted pair
IEEE 802.3x - flow control in full duplex mode.
Actually, if the device is incompatible with the above standards, then it will be used ... difficult.
Auto negotiation function select 10m, 100M or 1000M
Auto speed selection on a twisted pair port. This, by the way, is important. The first gigabit media converters were sometimes designed only for gigabits, without the support of low speeds. Modern devices carry a small switchboard (two ports) on board, which allows independent coordination of speed and other exchange parameters.
Ethernet port supports MDI / MDI-X auto crossover
It seemed to me that gigabit without MDI-X is not to be found. This is an optional part of the standard, but in fact it is everywhere. Now even 100 without MDI-X is almost impossible to find.
Distances up to 20km (single mode fiber)
In fact - a repetition of the previous paragraphs, emphasizing the range.
Auto MDI / MDI-X support on RJ-45 port
Again double. Interestingly, on which other port (except for “RJ-45”) can there be Auto MDI / MDI-X?
Support Full Duplex and Half Duplex
Support for half duplex and full duplex sharing. Theoretically, it can be useful for older devices at low speeds.
Status TX, FXDK LINK / ACT, POWER, FDX to monitor network status
Not all LEDs are indicated.
1Mbit RAM for data buffer
Supports jumbo frame size 9K bytes
Copied from the description of the switch chip.
LED Indicators:
FX, TX, Link / Act (Left / Right), FDX, PWR
FX: Light on when fiber speed is 1000Mbps, else light off
TX: Light on when TP speed is 1000Mbps, else light off
Link / ACT: for the TP port, for light transmitting and receiving
FDX: TP on full-duplex mode is active, TP off-half-duplex mode is active
PWR: Light on when DC power on, else light off
A variety of light bulbs are listed.
Output voltage: AC 100-250V, 50-60Hz
Output ??? Places where it would give out 220V I did not find.
Power dissipation: ≤3 watts
Judging by the measurements - the truth.
Housing: Metal enclosure
Yes. Metal.
Dimensions: 93 x 70 x 26mm
Dimensions, about the same. Only you need to take into account the hefty plug of a standard electric cord.
Weight: 0.5kg / Pair
Sinful, not weighed.
Operating Temperature: 0 ° C to 60 ° C
Theoretically and, most likely, not for long. In fact, in the loaded mode, overheating (with the case open!) Is more than 20 degrees. With the case closed and at +60, the inside will be over 80, and on the laser and the switch chip even higher.
Storage Temperature: -20 ° C to 70 ° C
Another confirmation that the previous parameteroverly optimistic. Additional heating during operation is several times higher than the specified difference of 10 degrees. Most likely, the storage temperature is correct, and the real working ambient temperature should be no more than 45-50 degrees.
Summing up, I would call the description quite ordinary, not carrying any particularly valuable information, slightly controversial, but also not having particularly terrible mistakes or problems.
The upper part of the box is quite standard for such devices, gray cardboard with "type of care" about the environment:

From one side the list of possible devices:

None is noted, and the names do not converge.
On the other hand, a more correct label that matches the sticker on the case:

Devices are mirror-symmetrical, so marking is different. I suspect that the sequence of numbers is the release date. If so, it is very fresh.
Apparently, in the manufacture of thisthe label did not use unproductive human labor, because the use of modern intelligent scanning and text recognition systems allows you to significantly optimize costs and reduce time to market. Yes, and what a normal user will read this fine print?
Inside the box, the converter itself and the power cable with plug for fans of network splitters from Xiaomi:

Also in the box there is documentation for Chinese experts:


If you have already managed to read and translate it, then keep in mind - it is intended for advertising purposes or to take care of expanding your horizons. It is not from the supplied model.
Unlike the small font on the label, the inscription on the upper part of the case immediately makes it clear to the consumer that this is not some kind of fake from Japan, but a real Chinese quality, from the backwoods:

But I didn’t decide how to properlyto translate into Russian - is it an Internet media converter or a network media converter? However, for the review it is not important, it is possible to simply convert the Internet carrier to enter the network during the test.
The front panel pleases the user with its smooth curves:

Especially for users wishingto customize the bends to fit your aesthetic ideas, the manufacturer made a case of easily bending iron for cans of the highest quality, with thickness reaching 0.6 mm in some places (if you count the paint thickness and slight bends at the edges). The thickness of the pure metal is about 0.5 mm. The outside is painted quite normally, inside ... let's say that there are traces of paint.
To optimize cost and maintain a singleThe design solution of the case is unified for different models, including two-fiber ones. This should not make it difficult for the user, since in this single-fiber converter there is no slot above the tag physically RX, the cable can only be inserted into the TX socket.
On the left side of the housing seal
was
When I ordered devices with built-in power supplies, I counted on something similar:

This is a relatively old media converter for 100 megabits, abandoned by the WestCall during retreat after the completion of the contract.
We open the case and see the following:

There is clearly a modern modular solution. In confirmation of this, I can pay attention to the fasteners: both boards have fastening points designed not only for this particular solution. On the electronics board this is the top right hole - there is no mounting support under it. Similarly designed right bottom hole of the power board. The work of the technologist is also visible - he had time to make a rationalization proposal that using the third screw to fix the power board (top left, next to the date) does not make sense and will only increase the cost and weight of the device. Unfortunately, the cases have already been made, so the third mounting box on this place still remained.
The design of the input power connector is also of particular interest:

It is well known that two wires are enough for passing electricity. Therefore, the third wire (grounding) can not be used, while improving ventilation inside the case.
Then you can unscrew the two screws from each of the boards and remove the transceiver cover:

Lower part of the case:

Screw axle boxes for mounting boards are visible. For threaded holes for the case screws applied a more technological solution - stamping.
Buksa, theoretically should ground the ethernet connector on the case:

Since we already found out earlier that it is quite possible to do without grounding, there is no point in removing paint from the support.
Power Supply Board on top:

The basis of the board is one-way getinaksby foiling. Not sure if the work of a technologist can be called excellent or just good. He successfully crossed out the completely unnecessary RV1 varistor from the installation list, but for some reason left the capacitor CY1. The remaining unnecessary elements (such as input filters) were apparently eliminated in previous revisions of the board. On the board of the type of care about the user - they warn him about the need to replace the fuse with the same type. Otherwise there will be a terrible squeak of fire ("pisk of fire"). The role of the fuse FR1 assigned to the resistor. Before using the device, it is necessary to write off its face value, because when it burns out, the labeling may not be read.
Power Supply Board Bottom:

Circuitry is standard for low-powerconverters. The input uses a diode bridge MB10F, providing a current of 0.5-0.7A (depending on temperature) at voltages up to 700V. Further installed electrolytic capacitor C1 (4.7mF, 400V).
Converter Chip:

What is the DP3773 I did not find, but the scheme is completely transparent:
1 - Regulator from the second winding and divider R3 / R4
3 - Power supply (provided by capacitor C2 - 4.7mF, 50V)
5.6 - key output.
R2, C3, D2 - the extinguishing chain (snubber). It seems that even the polarity of the diode is not confused.
The output half-wave rectifier, two capacitors (ceramic and electrolyte) and a resistor.
Kirich Y-capacitor:

The marking is formally present Y1.
The built-in power supply board is similar in design to the external power supply boards from other media converters:

But the built-in board is somewhat optimized - all unnecessary elements of the input filter type have been removed.
Main board:
Right below are the input filters andlinear stabilizer based on ASM1117. All components use 3.3V power. At the bottom left - the LAN connector and transformer. On the left, an open optical transceiver is visible. You can see two differential lines connecting it with the main chip.
Main components, large:

"Heart" of the media converter - microcircuitfour-port gigabit switch QCA8334 from Qualcomm / Atheros. The chip has many functions, most of which are in no way used in this product. It can be considered in detail for a long time, only in its main datasheet almost 300 pages. Theoretically, it costs less than $ 4. In such devices, it is often used, for example, in Mikrotik RBFTC11.
The small 14-foot microcircuit is also not just logic - it is an 8-bit, one-time programmable microcontroller. He is engaged in setting the switch and also controls the LEDs.
From the point of view of the network, the device is transparent, it is not controlled in any way and is not seen on the network.
The bottom of the board is of no interest.
To cool the transceiver, a cutout was made in the board:

Through the cut-out you can see two microcircuits - the amplifier of the receiver and the transmitter of the transmitter.
By the way, such a form factor (sometimes called 9 × 1)An integrated optical transceiver is quite common and no less ancient. In the earlier picture from the Avaya Cajun P330R module, you can see almost identical, only in a metal case and with a two-core connection.
Nutrition, consumption
Testing has shown that the board ischaracteristics very similar to conventional 5-volt external power supplies from similar devices. At the output of the power supply in idle mode 5.4V, under load, the voltage drops to 5.2. From the degree of loading of the device practically does not change. The quality requirements for this unit are minimal, since the main board has an additional 3.3V stabilizer. For normal operation, the AMS1117 has enough reserve at 1.3V, i.e. anything above 4.6V is permissible.
The test showed that the current when turned on is about 0.45A, but with unconnected ethernet, after a few seconds it drops to 0.28A. When connecting ethernet - 0.4A.
The presence / absence of optics is not affected. This is understandable, since the optical transceiver does not physically have a power control circuit, and it always shines the same.
Thus, in most cases, AMS1117 works with almost double margin for current limit.
Data Exchange Testing
For testing media converters, a stand was assembled from the materials at hand. It was not originally intended to achieve some kind of record indicators, on the contrary, I tried to bring it closer to life.

As servers for the exchange of information weretwo identical Supermicro 6016 typewriters with E5630 processors (2 pieces each) and an onboard Intel 82576 Gigabit Dual Port network card are involved. Servers are used in work, but lightly loaded in part of the network. In one, a free interface was involved, in the other, a low-load one. However, the extra traffic at the level of kilobits per second is not important for testing the gigabit interface.
Windows Server 2003 is installed on one server, the following settings are hammered into the network card:
IP address. . . . . . . . . . . . : 192.168.20.211
Subnet mask . . . . . . . . . . : 255.255.255.0

Also added a permanent route to the second server:
route add 192.168.0.0 mask 255.255.0.0 192.168.20.254 -p
The second server is running the old-fedor with the kernel Linux2.6.35.14-106.fc14.x86_64
eth1 Link encap: Ethernet HWaddr 00: 25: xx: xx: xx: xx
inet addr: 192.168.10.125 Bcast: 192.168.10.255 Mask: 255.255.255.0
Kernel IP routing table
Destination Gateway Genmask Flags MSS Window irtt Iface
192.168.10.0 0.0.0.0 255.255.255.0 U 0 0 0 eth1
192.0.0.0 192.168.10.1 255.0.0.0 UG 0 0 0 eth1

In addition to servers, a middle switch is involved.Cisco 3750G Layer and Eltex ESR-200 mid-level router. As is often the case in the real world (where the square is rolled and the round is dragged) routing is assigned to the switch, and a bridge is assembled in the router. It's more convenient for me, but iron allows it.
Cut from switch config with two VLANs and routing between them:
ip routing
vlan 20
name test
vlan 2
name native
interface GigabitEthernet1 / 0/5
description test_media_conv
switchport trunk encapsulation dot1q
switchport trunk allowed vlan 20
switchport mode trunk
spanning-tree portfast
!
interface GigabitEthernet1 / 0/14
description linux
switchport access vlan 2
switchport mode access
!
interface vlan2
ip address 192.168.10.1 255.255.255.0
!
interface Vlan20
ip address 192.168.20.254 255.255.255.0
!

The ESR-200 has a bridge between Gi1 / 0/3 and Gi1 / 0/6.
Interface to the media converter / sfp:
interface gigabitethernet 1/0/3
description "mc_test" security-zone trusted
switchport mode trunk
switchport trunk allowed vlan add 20
exit

Interface to the W2003 server:
interface gigabitethernet 1/0/6
description "test_server" security-zone trusted
switchport access vlan 20
exit

Servers are connected to access ports of the switchand the router, in the direction of the media converters, there is already tagged traffic with VLAN 20. At the same time, we will check the operation of the switch integrated into the media converters in the mode of tagged traffic transmission. Sometimes there are problems with this.
If anyone is interested, you can see all versions of firmware and hardware devices used:
Cisco3750


ELTEX ESR-200


Further tests will be performed using the iperf utility. On Windows, it is running in server mode, on Linux - as a client. It is possible and vice versa - it does not affect the result.
The test was performed as follows
1) Connect the necessary wires / devices
2) iperf was run a couple of times at idle to achieve repeatable results
3) iperf was launched 3 times for 30 seconds with small pauses between launches
In the reports, first comes the protocol from the linux server, then from windows.
Before testing the media converters, I checked the stand on the usual 20-meter copper wire:
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40986 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.09 GBytes 886 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40987 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.08 GBytes 883 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40988 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.08 GBytes 882 Mbits / sec
[1828] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40986
[ID] Interval Transfer Bandwidth
[1828] 0.0-30.0 sec 3.09 GBytes 885 Mbits / sec
[1848] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40987
[ID] Interval Transfer Bandwidth
[1848] 0.0-30.0 sec 3.08 GBytes 883 Mbits / sec
[1832] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40988
[ID] Interval Transfer Bandwidth
[1832] 0.0-30.0 sec 3.08 GBytes 882 Mbits / sec

Exceptionally just in case, a test is made on a short patch cord (3 meters):
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40991 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.09 GBytes 884 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40993 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.06 GBytes 877 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40994 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.09 GBytes 884 Mbits / sec
[1840] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40991
[ID] Interval Transfer Bandwidth
[1840] 0.0-30.0 sec 3.09 GBytes 884 Mbits / sec
[1848] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40993
[ID] Interval Transfer Bandwidth
[1848] 0.0-30.0 sec 3.06 GBytes 877 Mbits / sec
[1832] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40994
[ID] Interval Transfer Bandwidth
[1832] 0.0-30.0 sec 3.09 GBytes 884 Mbits / sec

As you can see, reducing the length of the wire does not increase the speed.
Test on media converters:
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40997 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 2.98 GBytes 853 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40998 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 2.98 GBytes 854 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 40999 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 2.98 GBytes 852 Mbits / sec
[1848] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40997
[ID] Interval Transfer Bandwidth
[1848] 0.0-30.0 sec 2.98 GBytes 853 Mbits / sec
[1828] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40998
[ID] Interval Transfer Bandwidth
[1828] 0.0-30.0 sec 2.98 GBytes 854 Mbits / sec
[1840] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 40999
[ID] Interval Transfer Bandwidth
[1840] 0.0-30.0 sec 2.98 GBytes 851 Mbits / sec

Immediately visible drawdown of approximately 30 Mbits / sec. This is the result of introducing into the trunk two additional switches (in each of the media converters). Unfortunately, any, even the best switch will cause an additional delay. The reason for this is simple - the block begins to be transmitted only after the receipt and verification of checksums is complete.
To test compatibility with other equipment and test the effects of an integrated switch, a 10 km SFP module is inserted into the ESR-200, replacing one of two media converters:
Interface ‘gi1 / 0/3’:
SFP present: Yes
Connector Type: LC
Type: SFP / SFP +
Compliance code: BASE-BX 10
Laser wavelength: 1550 nm
Transfer distance: 10.00 km
Vendor OUI: -
Vendor name: Carelink
Vendor PN: CLSFPWDM1055DD
Vendor SN: GSxxxxxxxxxx
Vendor date: 04/25/14
Vendor revision: 1.0
DDM supported: Yes
Temperature: 35.430 C
Voltage: 3.3868 V
Current: 19.620 mA
RX Power: 0.1851 mW / -7.3259 dBm
TX Power: 0.3135 mW / -5.0376 dBm
RX LOS: No

This module has a DOM / DDM function,allows to obtain additional information about power levels, consumption, etc. The reception level at the short connector has a relatively large margin - with a typical sensitivity of 20 dB, it is permissible to lose another ten decibels on the cable. However, for a line of 20 kilometers it will be on the verge - depending on the quality and number of welds and connectors.
The port Gi1 / 0/3 is combo, so you can switch hot and even without changing the configuration. Just pull out the copper cable and insert the module.
Communication test between sfp module and media converter:
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 41001 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.05 GBytes 874 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 41002 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.04 GBytes 871 Mbits / sec
[[email protected] asterisk] # iperf -c 192.168.20.211 -t 30
Client connecting to 192.168.20.211, TCP port 5001
TCP window size: 16.0 KByte (default)
[3] local 192.168.10.125 port 41003 connected with 192.168.20.211 port 5001
[ID] Interval Transfer Bandwidth
[3] 0.0-30.0 sec 3.05 GBytes 875 Mbits / sec
[1836] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 41001
[ID] Interval Transfer Bandwidth
[1836] 0.0-30.0 sec 3.05 GBytes 874 Mbits / sec
[1832] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 41002
[ID] Interval Transfer Bandwidth
[1832] 0.0-30.0 sec 3.04 GBytes 871 Mbits / sec
[1848] local 192.168.20.211 port 5001 connected with 192.168.10.125 port 41003
[ID] Interval Transfer Bandwidth
[1848] 0.0-30.0 sec 3.05 GBytes 875 Mbits / sec

As you can see, the result falls almostmidway between the previous two. This is understandable - only one (of two) additional switch remains in the circuit. The SFP module provides a direct connection between the output of the ESR-200 network controller and an optical cable; it has neither intermediate buffers nor intelligence.
During the tests, I took readings of the CPU load on the ESR-200 router (he, poor thing, worked as a switchboard):
show cpu utilization
CPU Last Last Last
5 sec 1 min 5 min
0 5.29% 4.17% 4.26%
1 68.62% 18.36% 17.85%
2 4.26% 4.66% 12.37%
3 21.48% 5.50% 7.85%
CPU Last Last Last
5 sec 1 min 5 min
0 12.41% 5.63% 4.59%
1 4.50% 17.74% 18.54%
2 69.50% 11.92% 13.94%
3 22.13% 7.51% 8.52%

In conditional idle mode (it slightly processes NAT from other interfaces), the readings are as follows:
show cpu utilization
CPU Last Last Last
5 sec 1 min 5 min
0 3.78% 4.31% 4.29%
1 4.06% 21.71% 18.52%
2 4.27% 4.61% 12.40%
3 0.40% 6.49% 8.05%

As you can see, one loaded gigabitthe bridge between ports eats about a quarter of the device's performance. In the documentation of the performance values ​​in this mode, I did not find, but 2.3Gbps is promised in the NAT mode.
The load on the Cisco 3750G processor is load independent:
In idle mode:
CPU utilization for five seconds: 9% / 2%; one minute: 9%; five minutes: 9%
During the tests:
CPU utilization for five seconds: 8% / 2%; one minute: 9%; five minutes: 9%
This is natural, since The 3750 processor does not participate in packet routing; there are hardware blocks for this.
Power comparisonUnfortunately, unnecessaryAt the time of testing, I did not have a 20-kilometer optics on hand, and testing on two to five kilometer lines does not make sense, even if you turn on the loop (it will be less than 10 km).
Therefore, I will limit the measurement of power and comparison with other devices.
Laser power at 1310nm:

Laser power at 1550nm:

As you can see, the power is close and is slightly less than -5 dB. This is normal power for 20 km devices. Typical interval from -3 to -8 dB.
For comparison, I measured several other single-core sfp-modules at 10 and 20 km.

I will not insert measurement pictures, I will provide only text.
SFP present: Yes
Connector Type: LC
Type: SFP / SFP +
Compliance code: BASE-BX 10
Laser wavelength: 1550 nm
Transfer distance: 10.00 km
Vendor OUI: -
Vendor name: Carelink
Vendor PN: CLSFPWDM1055DD
Vendor SN: GSxxxxxxxxxxxxx
Vendor date: 04/25/14
Vendor revision: 1.0
DDM supported: Yes
Temperature: 35.844 C
Voltage: 3.3856 V
Current: 19.620 mA
RX Power: 0.1846 mW / -7.3377 dBm
TX Power: 0.3135 mW / -5.0376 dBm
RX LOS: No
Measured output power (dB): -7.4 db
SFP present: Yes
Connector Type: SC
Type: SFP / SFP +
Compliance code: 1000BASE-LX
Laser wavelength: 1310 nm
Transfer distance: 10.00 km
Vendor OUI: 00: 90: 2E
Vendor name: OptiCin
Vendor PN: SFP-WDM.DDM.3.10
Vendor SN: PE9Gxxxxxx
Vendor date: 09/23/14
Vendor revision: 1.0
DDM supported: Yes
Temperature: 35.656 C
Voltage: 3.3104 V
Current: 21.136 mA
RX Power: 0.3016 mW / -5.2057 dBm
TX Power: 0.2742 mW / -5.6193 dBm
RX LOS: No
TX Fault: No
Measured output power (dB): -6.16 dbm
Relatively branded D-Link:
Interface ‘gi1 / 0/3’:
SFP present: Yes
Connector Type: LC
Type: SFP / SFP +
Compliance code: 1000BASE-LX
Laser wavelength: 1310 nm
Transfer distance: 10.00 km
Vendor OUI: -
Vendor name: D-Link
Vendor PN: DEM-330R
Vendor SN: BHCxxxxxxx
Vendor date: 09/20/07
Vendor revision: 0000
DDM supported: No
Measured Output Power (dB): -7.4
Interface ‘gi1 / 0/3’:
SFP present: Yes
Connector Type: SC
Type: SFP / SFP +
Compliance code: 1000BASE-LX
Laser wavelength: 1550 nm
Transfer distance: 20.00 km
Vendor OUI: -
Vendor name: OEM
Vendor PN: SFP-T-20-WDM
Vendor SN: SGxxxxxxxxxx
Vendor date: 04.11.20 09
Vendor revision: 1.00
DDM supported: No
Measured output power (dB): -6.1
Interface ‘gi1 / 0/3’:
SFP present: Yes
Connector Type: LC
Type: SFP / SFP +
Compliance code: 1G SFP some fiber transceiver
Laser wavelength: 1550 nm
Transfer distance: 20.00 km
Vendor OUI: 00:90:65
Vendor name: OEM
Vendor PN: SFPWDM1.25G-1550
Vendor SN: PBxxxxxxxxx
Vendor date: 06/07/11
Vendor revision: A0
DDM supported: No
Measured output power (dB): -4.1

As you can see, for 10 km models the power is no more than -6 dB, for 20 km -4 ..- 6. The considered 20-kilometer media converters fall almost in the center of the range.
For comparison, here is the state of the working line (several kilometers):
Interface ‘gi1 / 0/4’:
SFP present: Yes
Connector Type: SC
Type: SFP / SFP +
Compliance code: 1000BASE-LX
Laser wavelength: 1310 nm
Transfer distance: 20.00 km
Vendor OUI: 00: FFFFFF90: 65
Vendor name: OEM
Vendor PN: SFP-1SM-13-20SC
Vendor SN: U11072301304001
Vendor date: 11/07/20
Vendor revision: A0
DDM supported: Yes
Temperature: 42.328 C
Voltage: 3.2682 V
Current: 27.600 mA
RX Power: 0.2369 mW / -6.2543 dBm
TX Power: 0.3976 mW / -4.0055 dBm
RX LOS: No
TX Fault: No
TX Disable: No
Soft TX Disable: No
show int gi 2/0/10 transceiver
Temperature Voltage Current Tx Power Rx Power
Port (Celsius) (Volts) (mA) (dBm) (dBm)
———
Gi2 / 0/10 34.1 3.19 27.2 -3.7 -9.4

By the way, if the diagnostics is not lying, then the difference in attenuation between 1310 and 1550 is already noticeable. On the same fiber, 1310 nm was attenuated by 5.4 dB, and 1550nm by only 3.5 dB.
Results Examined glands leaveconflicting impressions. On the one hand, the technical characteristics relating specifically to data transmission are quite adequate both in terms of power (range) and in terms of compatibility with other devices. No problems here have been identified. On the other hand, the scope of individual media converters in general, and specifically this device, is small and decreases every year. The reason for this is not an additional delay in transmission. Simply, any individual box costs money, and now it is much easier to buy a terminal device with an integrated SFP port (a gigabit microtic RB260GS with one SFP and 5 copper ports can be bought for $ 40). Practically, only the cheapest switches, "soap boxes" do not have SFP. And any "serious" switchboard without expansion slots is simply not to be found.
With the surveyed device is also not all clear:
- The kit can not be called cheap. The price is justified only if you need exactly gigabit, the distance is more than 10 km and there are no free SFP slots in the switches. If the connected switches have an empty SFP slot, then it is better (and slightly cheaper) to use SFP modules. A set of media converters for 100 Mbit will be half the price. Gigabit shorter distances can be found cheaper.
- I doubt that it can be certified for electrical safety
Here you can also add practicalthe inability to buy by bank transfer and with the necessary certificates. As a result, we have a device that is not useful to a private user (or does someone have an optical line for a dozen or more kilometers for personal use?), And it’s easier for an enterprise to buy more expensive, but with all the necessary paperwork and by bank transfer. Personally, I plan to use only for emergency replacement, and then only for a short time. In terms of electrical safety, ordinary Chinese media converters with separate power supplies in plastic cases are no better.
Legs, wings .. The main thing is the tail


Adding a store to your favorite opens additional discount coupons.