Protocols in traffic
In TCP/IP, the "TC" stands for "Transmission Control," and, unfortunately, some of the rules implemented in TCP/IP to make it a reliable networking protocol actually work against it when large amounts of data need to be transferred. For example, when packets are dropped on a WAN link, TCP/IP automatically tells servers to throttle back transmissions, which greatly decreases performance. In our testbed, HyperIP appliances were configured as IP gateways sandwiching a NIST Net WAN simulator.Our NIST Net WAN simulator was configured to simulate 40 milliseconds of latency, which is roughly equivalent to the latency of a 1,000-mile WAN link. We also configured NIST Net to simulate packet drops at a rate of 0.5 percent to make the link even more realistic. We configured the HyperIP appliances to be our default gateways. As TCP/IP traffic was sent from our servers to our WAN, the HyperIP gateways terminated the TCP/IP connection at the gateway and streamed the data across the WAN link using NetExs proprietary WAN accelerator protocols. Once data crosses over the WAN, it is received by a second HyperIP appliance, which re-establishes the TCP/IP link to the receiving server. The beauty of the HyperIP method is that it makes the whole WAN acceleration process transparent to the servers. HyperIP harnesses proprietary performance-enhancing protocols to keep TCP/IP traffic flowing. These protocols are similar in function to the Performance Enhancing Proxies documented in The Internet Societys RFC 3135 (at www.rfc-archive.org/getrfc.php?rfc=3135), released in 2001. From the servers point of view, with the HyperIP appliances in place, the WAN link is a stable high-speed connection and can send data quickly without having to throttle back data rates or resend numerous packets over the WAN. The HyperIP appliances have built-in cache, which allows them to reorder packets that come in out of sync and helps prevent the transmission control protocols from slowing down data replication. In tests, we found that traffic was significantly accelerated, with HyperIP appliances enhancing WAN transmissions. Using NSI Softwares Double-Take to replicate data, we were able to push roughly 46MB per minute over a simulated T-3 link at 1,000-mile latency with Double-Takes compression activated. Using the HyperIP to accelerate the WAN, we could push 348.5MB per minute over the link, which essentially increased our transmission speeds more than sevenfold. The HyperIPs benefits are not limited to software-based replicationthe appliance can also be used to enhance storage-array-based replication solutions, including EMC Corp.s SRDF (Symmetrix Remote Data Facility), and it can accelerate other IP-based protocols, such as FTP. In FTP tests using a 160MB movie file as payload, we saw a significant level of acceleration in our tests. With the HyperIP appliances activated, we could retrieve files using FTP at a rate of 4,964KB per second. Running the same test without the HyperIP appliances, performance dropped to 774KB per second, which shows that FTP was accelerated sixfold (6.41 times, to be exact). For IT managers who do not want to run the HyperIP in gateway mode, the appliance can also be run in proxy mode so that the HyperIP accelerates only specific applications instead of all outbound traffic. This will ensure that the HyperIP focuses on mission-critical traffic from replication applications. The HyperIP appliance has an LZO (Lemspel-Ziv-Oberhumer) compression engine that compresses blocks of data to maximize WAN acceleration. In our tests, we chose to disable the LZO compression engine. Next page: Evaluation Shortlist: Related Products.
To create replication traffic, we used NSI Software Inc.s Double-Take to replicate data from one Windows Server 2003 server to a second one on the other side of the simulated WAN.