iSCSI sorts out performance issues

BANGALORE, INDIA: Internet Small Computer System Interface (iSCSI) failed miserably to live-up to the hype it generated back in 2003 and today, even after seven years, it is still considered a low-end technology.  

With issues, such as 'being the cheaper alternative to FCoE' (Fibre Channel over Ethernet), and 'SMB type', already been dealt with in earlier write-ups, let us here analyze a few others, which were and are still associated with  iSCSI.

Also Read:Time to get over the fibre fixation

Those issues, which prevented data center managers from taking up high-end networks over iSCSI networks, till sometime back.

Issue No 1: TCP/IP Overhead
There is a take that since iSCSI runs over TCP/IP (Transmission Control Protocol/Internet Protocol) it is susceptible to TCP "overhead".

K Raghavan, principal technical consultant, Cisco, says: “iSCSI runs directly on top of TCP/IP. Thus, there is CPU overhead in operating a TCP stack. For some applications with minimal I/O there is a minimal difference between iSCSI & native FC. However, for many there can be a huge CPU difference between the two.”

Hence, the performance of iSCSI has to deal with OS latency and CPU utilization on the iSCSI target rather than on the media, he added.

Harmeet S. Malhotra, enterprise solutions marketing director, Asia Pacific & Japan, Storage & Unified Fabric Solutions Marketing, Dell, explains that earlier, if a 1gigabit LAN card is used with iSCSI it eats away 1gigahertz of CPU performance. That means, if a server has 4giga hertz of total CPU capacity and I gigabit card is used,  it eats away 25 percent of the CPU performance.

However, not so anymore because today with  multi-core CPUs, one can achieve 20-40 gigahertz of server performance, he notes.

“Moreover, TCP/IP offload engine reduces the load of IGbps NIC from I gigahertz to a negligible amount, i.e 50 hertz, or 50,000 hertz, which is a mere .5 percent. So it probably has the same load on CPU as that of a fibre channel LAN card,” Malhotra adds.

{#PageBreak#}

Issue No 2: High OS Latency

This was another issue that kept the less tolerant large enterprises away from iSCSI for a long time because iSCSI also brought with it high OS latency and packet loss.

“OS latency is something that both iSCSI and fibre channel has to deal with because both of them have the same drive. iSCSI used to come as pure network card earlier. However, today you get iSCSI HBA, which takes care of the entire issue of latency,” Malhotra adds.

Sanjay Lulla, director, Technology Solutions, EMC India & SAARC, says, “Some of the initial inhibitors such as shared bandwidth, and lower throughput are now addressed by DCB. With time, the security on Ethernet & iSCSI will also evolve to the next level.”

{#PageBreak#}

Issue No 3: iSCSI is a low performer
Even after all these years, iSCSI still finds it difficult to shake off the tag of being a low performer when compared to fibre channel.

“iSCSI was originally developed over 10 years ago as an inexpensive option to transport storage data over unreliable Ethernet. However, iSCSI didn’t get too far as it could not penetrate data centers due to its relatively poor performance (1 Gbps) and security concerns, when compared to fiber channel,” Lulla adds.

However, today with the advent of 10 GbE, iSCSI has gained considerable ground, especially when FC is still at 8Gbit/s.

Having said that, I would also like to add that although FC networks are today restricted at two, four, and 8Gbit/s, a 16Gbit/s FC standard is on the cards.

The ANSI T11 committee is expected to ratify the new standard by the end of this year, with product is likely in 2011. Adding to it will be a 32 Gbit/s FC standard, which is also under consideration.

Similarly, on the Ethernet front, the IEEE 802.3ba standard for 40Gbit/s and 100Gbit/s Ethernet systems were ratified last month. Thus, the competition is only becoming tougher.

“Protocols will tend to co-exist, and depending on the need of the day, customers will choose FCoE or iSCSI. We do foresee a trend towards Ethernet oriented protocols owing to the fact that we can leverage the Ethernet to meet better performance parameters as we move forward,” notes Lulla.

This brings into mind a similar kind of dilemma over WiMAX and LTE that the Indian market is currently going through. Whatsoever, importance of Ethernet is picking up like never before and let's leave it to the market dynamics to decide upon which way to take with respect to FCoE and iSCSI.

As of now it seems, “It is not about one protocol versus the other. Customers today are increasingly asking for convergence of data centre fabric and flexibility to use both (iSCSI-FCoE) as required,” sums up Lulla.