For 17 of the 30 lmbench development mi crobenchmarks

This preview shows page 6 - 8 out of 13 pages.

For 17 of the 30 lmbench DEVELOPMENT mi- crobenchmarks, the performance of PLK, Xeno- Linux, and stock Linux are nearly identical; i.e., within the margin of error. Table 1 shows the results of the latency microbenchmarks for which there is a performance discrepancy between stock Linux, PLK, and XenoLinux. The first row in Table 1 shows that latency to han- dle TCP select is a bit faster for XenoLinux when compared to stock Linux and PLK. This is the only case where XenoLinux has lower latency, which happens to be due to minor code path difference be- tween the vanilla 2.6.12 and FC4-1398 2.6.12 ker- nels. The next two rows show mmap latencies for 32MB and 64MB files. As can be seen the laten- cies scale wrt to the size of the file, rather than there being a constant overhead for XenoLinux. While this makes perfect sense, we were initially surprised because previous discussions with respect to this benchmark made it seem like it would be a constant overhead of a few tens of microseconds. The following three rows in Table 1 show the per- formance of fork process , exec process , and sh pro- cess across the systems. XenoLinux is slower when compared to stock Linux. Barham et al. [1] describe why this is the expected behavior for XenoLinux: hypercalls to the VMM update a large number of page tables. PLK is nearly identical for the fork sh benchmark, a bit slower for exec process, and only a bit faster than XenoLinux for sh process. Techni- cally, PLK should more closely match stock Linux for these latter two benchmarks and we are still in- vestigating what causes the additional delay. The next seven rows in Table 1 show context switch overhead between different numbers of pro- cesses with different working set sizes. As ex- plained by Barham et al. [1], the 1 μ s to 4 μ s over- head for these microbenchmarks are due to hyper- calls from XenoLinux into the VMM to change the page table base. In contrast, there is little overhead between the stock and paenevirtuailzed versions of Linux. Finally, for protection fault and page fault han- dling, XenoLinux is more than 1.4x slower than stock Linux, which is also due to Xen VMM hy- percalls. For PLK the cost for these operations are essentially identically to stock Linux. Table 2 shows the results of the bandwidth based microbenchmarks for which there is a per- formance discrepancy between stock Linux, PLK, and XenoLinux. The first three rows show an odd theme: XenoLinux drastically outperforms both stock Linux and PLK for strictly user-level bench- marks. The differences between these should be much smaller. We have run these benchmarks with sufficient warm up iterations to factor out cache effects and page fault effects and continue to get the same results. Not shown in the table are the 6
Image of page 6

Subscribe to view the full document.

Linux PLK Xeno Config FC4 Linux bcopy (libc) 597.70 583.08 729.12 bcopy (hand) 585.72 571.82 744.62 mem write 834.60 795.22 1038.00 mem read 2237.14 1910.02 1933.02 pipe 1993.80 1906.32 2294.30 AF UNIX 2838.82 2768.52 2668.70 file reread 1739.16 1656.12 1809.22 mmap reread 2241.82 1922.36 1942.62 Table 2: lmbench bandwidth OS benchmarks - in Mbytes/sec bandwidth numbers for the host domain (dom0) under Xen. For the bcopy (libc) benchmarks, it falls right in between PLK and XenoLinux—i.e., 660Mbytes/sec. The cause for this is not well un-
Image of page 7
Image of page 8
You've reached the end of this preview.
  • Spring '12
  • GwangS.Jung
  • virtual machine, VMs, Linux kernel, Xen, namespace isolation

{[ snackBarMessage ]}

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern