Competitive Performance Evaluation versus Cisco Catalyst 4948-10GE Switch

T H E
TOLLYG R O U PNo. 208348 October 2008
BLADE Network Technologies, Inc. TestRackSwitch G8000Competitive Performance Evaluation versus SummaryCisco Catalyst 4948-10GE Switch
Premise: As demands for Web 2.0, high-performance computing (HPC), and vir- Test Highlightstualized data centers continue to grow, enterprise data centers that have already
Demonstrates an average of 3X less latency than a Cisco 4948-consolidated their server resources using blade server systems are finding that top- 10GE switch for various frame sizes using two different port-of-rack switches can provide unified and pairing schemesaffordable rack-level virtualization of
Sustains line-rate throughput for all frame sizes tested up to servers, blade server systems and storage resources, while reducing power con- 9,216 bytes without losing a single packet, while the Cisco 4948-sumption, cooling and floor space. At the 10GE achieves 98.8% of theoretical maximum throughputsame time, network performance and cost
Consumes 30% less energy than the Cisco 4948-10GE when remain the most important factors when choosing an Enterprise data center handling 50% traf?c load with all ports transmitting, which is solution. equivalent to $42.84 savings per year per switch
Exhibits 2.1X better price/performance compared to the Cisco LADE Network Technologies, Inc. (BLADE) commissioned The Tolly 4948-10GEBGroup to evaluate the Layer 2 per-formance and power consumption of its RackSwitch G8000 1/10G aggregation Layer 2 Latency Comparison of RackSwitch G8000 vs. switch with 48 GbE ports and four 10GbE Cisco 4948-10GE Using Two Port-pairing Schemes*uplink ports - designed for top of rack as reported by Spirent TestCenter 2.22aggregation for large data centers - (Lower bars are better))against a comparable solution from Cisco s 120dSystems, Inc. (Cisco 4948-10GE). The 64 Bytes 512nCisco 4948 is a 1RU switch with 48 GbE 100o 1,518 9,216cports and two 10GbE uplink ports. Since e 80sthe Cisco 4948-10GE supports two 10GbE Up to 3X lower latency than Up to 3X lower latency than orports only, two of the four 10GbE ports in 60c Cisco 4948-10GE Cisco 4948-10GEithe front of the RackSwitch G8000 switch m 40( were used. Avg 5.1 ?s Avg 11.3 ?s Avg 4.6 ?s Avg 9.9 ?syc 20nTolly Group engineers measured the Layer et 0a2 throughput, latency and multicast per- L BLRAaDckES wGi8tc0h0 0 Cisco Catalyst BLADE G8000 Cisco Catalystformance of the switches in a variety of Cisco RackSwitch Cisco G8000 44984489--1100GGEE G8000 44984489--1100GGEEcommon network topologies, and the Full Mesh Backboneenergy usage for the switch operations at idle and under load. Engineers also com- Notes: Port-pairing scheme 1 ("Hybrid" full mesh) - 48 GbE ports in full mesh and two 10GbE ports in pared the price/performance of the switches pair. Port-pairing scheme 2 (Backbone) - First 10 GbE ports transmit/receive traffic with first 10GbE port from both companies. and last 10 GbE ports with the second 10GbE port. Engineers measured the store-and-forward latency at the traffic load of 95% of theoretical maximum. Latency average does not include Jumbo Frames.Tests were conducted in September 2008. Source: The Tolly Group, September 2008 Figure 1
© 2008 The Tolly Group Page 1 BLADE Network Technologies RackSwitch G8000
Executive Layer 2 Throughput Comparison of RackSwitch G8000 vs. Cisco 4948-10GE-Using Summary Two Port-pairing Schemes*RackSwitch G8000 as reported by Spirent TestCenter 2.22consistently outper-forms the Cisco Cata- 100 lalyst 4948-10GE by 100%98.8%100%98.4%ci 80t )edemonstrating zero- r %(o loss, wire-speed 60e mht uthroughput, 66% less f 40mo i latency, 30% less t xanpower consumption 20e mcrand 100% better price/ e 0Pperformance than the RaBcLkASwDiEtch CCiissccoo RaBcLkASwDiEtch CCiissccoo Cisco 4948-10GE. GG88000000 4C8a4t9a-1ly0sGtE GG88000000 4C84a9t-a1l0yGsEt4948-10GE 4948-10GEApplication-driven network architectures are very common Full Mesh Backbonein the large Enterprise data cen-ters. As the application com- Port-pairing schemes and DUTplexity increases, different net- Notes: Port-pairing scheme 1 ("Hybrid" Full Mesh) - 48 GbE ports in full mesh and two work design and operations are 10GbE ports in pair. Port-pairing scheme 2 (Backbone) - First 10 GbE ports transmit/receive required. However, a non- traffic with first 10GbE port and last 10 GbE ports ... [download for more]