Increasing Real Throughput with Compression

Can I really get more the 2Mb/s on a T1?

The laws of physics says that 1.544Mb/s is 1.544Mb/s. But physics doesn't consider that the data may be compressed, then transported, and then decompressed, creating more than just an illusion of greater speed. Throughput is a function of time, and if you indeed do decrease the time to transport any given piece of data, then the gain is real. You really have shrunk the time of acquiring the data, and thus increased the throughput beyond what the physical medium might dictate. So yes, you really can get more than 2Mb/s on a T1.

What is Compression?

Everyone already knows what compression is. Compression is an organized reduction in the volume of data, where "organized" implies that you can "decompress" it to get something similar to or exactly the same as the original. You use compressed data all of the time. You use "zip" files all of the time. Your system backups are compressed. You might even be unlucky enough to have used a compressed file system. But its more than just that. Virtually every graphic file is compressed in one form or another. Just about all multimedia is compressed. Without compression there is too much information to transport over non-dedicated networks.

There are 2 general "kinds" of compression: "lossy" and "lossless". "lossy" compression allows for decompression that yields something that is not exactly what was originally compressed. JPEG (.jpg) images are the most common example. With graphics, if a few dots here and there are missing or wrong, your eyes can't tell the difference generally. So it doesn't matter that much. With audio and video, a minced bit here and there won't be noticed, unless you are planning on saving it many times (as if you are mixing audio for a CD). For data, lossless compression is required. You can't have a bit in an IP packet lost, or the checksum will fail or the file probably won't work.

You might ask why anyone would ever use "lossy" compression as opposed to "lossless", since it seems a lot better to get exactly was was originally compressed rather than something slightly different. The answer is that lossless compression has limits as to what compression ratios can be obtained that are much less than lossy compression. You can cheat more with lossy compression, so you can get much better compression ratios. But you don't get exactly what you started with, so its no good for the purposes discussed here.

Compression and the ET/BWMGR

Products running the ET/BWMGR software (including all of Emerging Technologies' bandwidth management appliances) implement selective compression between ANY 2 POINTS in a data stream that have the a unit. It doesnt matter whats in between, only that the data that is compressed also passes through the second unit so it can be decompressed back to its original content before it reaches the target device. If it seems pretty simple, its because it is. If you have an ET/BWMGR "box" at your office and you have one at home, you can compress all of the traffic from your office network to your home PC transparently. So if you have a 256kbs connection, it will appear to be 2-3x faster, in most cases.

About Compression Ratios

OK, so you read the "in most cases" above and now you're wondering what's going on. Why is there a conditional attached to this? Well, you have to know something about internet content and compression to understand. When you read about compression products the first thing that most people check is the compression ratio. Vendors advertise "4:1 compression" or "8:1 compression", but you have to be very careful. "4:1" implies that the data is 1/4 the original size. In the context of an internet data compression unit like the ET/BWMGR, that means that you can get 4 or 8 times the throughput. But compression is highly dependent on what is being compressed and the size of what you are compressing. So there is a dependence on the traffic.

First and foremost, a general rule is that you can't compress things that are already compressed. Think of it like a trash compactor. Once you run the compactor, the trash is compressed to maybe half of what it was, but running it again isnt going to do anything more. Its already as small as its getting. The same goes for data. So once its compressed, thats it. As mentioned before, graphics are already compressed. So are videos. And if you're downloading a .tgz file, its already compressed, so you won't see any "gain" for such files.

Secondly, there is the matter of size. A product like the ET/BWMGR compresses 'on the fly", which means that it doesn't try to figure out what the data is, it just grabs packets and tries to compress them. Internet packets are generally not larger than 1500 bytes, so you're not going to get as good compression on a small chunk as you might get on an entire file if you were compressing it on your hard drive.

Compression Ratios and the Internet

Compression ratios obtained in the context of a network appliance require some understanding of internet traffic and the way the compression is utilized. Since the packet needs to continue to be "IP functional", that is, it needs to be a routable packet, the IP header cannot be compressed nor can the ethernet header. Also its worthy of note that most 60 byte packets don't have any IP "data", so compressing them won't do much good, as the "filler" to fulfill the minimum ethernet packet size isn't going to be transported over your internet connection anyway.

You also must consider that compression is not a freebie in terms of overhead. It takes time to compress and decompress a packet, and it also takes time to "try" to compress a packet. The theory is that its worth the CPU cycles if compression can be obtained, because the benefit in terms of gain in throughput outweighs the cost of the reduction in overall performance of the compression device, particularly at today's CPU speeds. But when you try to compress a packet and fail, either because the packet is uncompressable or because there is no "gain" in the compression (ie the resulting packet would not be smaller than the original), then you have lost CPU cycles with no benefit.

Almost half of the packets that pass through an internet router or bandwidth manager are small packets. Most of them are TCP ACKs or handshake packets for starting and terminating connections. None of these packets can be compressed. They are more than 50% overhead, and the data content is minimal.

Compression Tuning with ET/BWMGR

An important strategy when using the compression feature on the ET/BWMGR is to minimize wasted overhead. The ET/BWMGR allows you to set general strategies for packet size and realized compression ratios, and also allows you to override the settings for particular "connections"or for specific protocols. If system capacity is a concern, you can elect to only try to compress large packets, or only HTTP traffic, for example. You can select a minimum compression ratio so that you don't waste cpu cycles compressing data that is only marginally compressable.

Realistic Compression Ratios Obtainable by the ET/BWMGR

Typically, you should actually be able to double your throughput with normal use.

Real Life Use of ET/BWMGR Compression

The diagram below illustrates a variety of environments and how compression can be used between them. In the discussion below, you should consider that the "Lakeville POP" is a hosting site and the "Clarkestown POP" provides residential and commercial service via dedicated T1 lines. The "cable modem" can be any "home" environment, such as the home of an executive or a tech support person.

Internet Utilization of ET/BWMGR Compression

Each location above has 1 or more "Bandwidth Manager" system, which can either be a bandwidth management appliance or a PC running the ET/BWMGR software module. Without any explanation at all, it should be noted that any of the locations can selectively option to compress data sent do any of the other locations. In a perfect world everyone would use the same device, but realistically such scenarios would have to be planned.

Scenario 1: ISPs offering Dedicated Services over WAN

Perhaps the most valuable use of compression would be for the Clarkestown ISP, as they could specifically benefit by offering an "enhanced" T1 service to any of their customers by simply offering a bandwidth manager solution to the client. Since each bandwidth manager can handle compression to any number of locations, they would only need a single unit at their location to handle any number of subscribers. So their variable cost would be nothing after the initial purchase. Clients who opted for the service would have to buy a bandwidth manager box and then pay a monthly fee for the service, but the cost of a 50-70% likely increase in throughput would be a small fraction of the cost of getting a second T1 and load balancing the lines. The solution is particularly useful on a shared connection like DSL, because the reduction in congestion on the shared access line would give everyone better throughput as well as more reliable service.

Scenario 2: Increased Bandwidth between Locations of a Multi-POP ISP or Corporation

Its likely that the Clarkestown and Lakeville POPs share more that a bit of information, whether it be a centralized database or backups. Enabling compression between them will increase the speed of the transactions, as well as reduce the amount of bandwidth that internal activities use, making more available to sell.

Scenario 3: Inter-Corporation Compression

The Allied and Acme corporations may do a lot of business, and may move large documents back and forth. Cutting the download time of those documents increases productivity. As long as both companies have an ET/BWMGR bandwidth management device, they can compress traffic between them. Both companies have a T1, so isn't 3Mb/s better than 1.5Mb/s?

Scenario 4: Increased Throughput for Home Access by Company Personel

Suppose your company is on a cable modem network, and you are on the same or a different one at your home? Cable companies have lower bandwidth upstream, so the traffic coming out of your office will be 256k or 1Mb/s or something in between, depending on what system you are on. By putting a bandwidth manager at the office and the homes of any executives that regularly access the office network, you can significantly increase access speeds with compression.