Software and hardware annotations 2008 November

This document contains only my personal opinions and calls of judgement, and where any comment is made as to the quality of anybody's work, the comment is an opinion, in my judgement.

[file this blog page at: digg del.icio.us Technorati]

081127 Thu New line of 10Gb/s PCIe cards with SFP+
I have been setting up a fairly significant network infrastructure with 10Gb/s fibre links between routers, and also at long last some storage servers with 10Gb/s cards, and soon workstations with the same. This has been driven both by requirements for data read or write higher than 100MB/s, and their feasibility thanks largely to PCIe with its transfer rates much higher than the 133MiB/s of basic PCI (and the higher transfer rates of SATA and of course of disks themselves). Slots with 4 or 8 PCIe lanes are available on many motherboards, even consumer level ones thanks to the use of PCIe by graphics cards. All this makes possible to create storage systems capable of delivering at a cost affordable even to individuals storage with transfer rates in the several hundred MiB/s range. But not over the network, as the cost of 10Gb/s cards has been too high.
Myricom have been selling cluster oriented components for quite a while, and like the SAN people they have converged on Ethernet based links, and after selling for while some of the nicer 10Gb/s cards with XFP transceiver sockets, they are now moving massively to the new SFP+ socket with some quite interesting new cards and transceivers. The card design is clearly derived from that for FCoE SAN cards, with dual sockets (paths to a SAN array are typically redundant) for $700, and 850nm transceivers for $300 and 1510nm ones for $600, and drawing 6-10W fully loaded. At these prices it is much easier to justify 10Gb/s for servers, and even for data analysis workstations.
081124 Mon Quick loss of battery capacity
I have been quite happy with the Toshiba (manufactured by Compal) Satellite U300 laptop, but I have recently noticed that the life of the battery (model ST-PA3594U-1BRS) has gone down tremendouly: when new I could get around 3.5 hours of use out of it, currently only about 2 hours:
     PowerTOP version 1.9       (C) 2007 Intel Corporation

Cn                Avg residency       P-states (frequencies)
C0 (cpu running)        ( 1.4%)         1500 Mhz     0.0%
C1                0.0ms ( 0.0%)         1000 Mhz   100.0%
C2                0.0ms ( 0.0%)
C3               13.7ms (98.6%)


Wakeups-from-idle per second : 72.0     interval: 5.0s
Power usage (ACPI estimate): 13.0W (02:06 remaining)

Top causes for wakeups:
  36.8% ( 26.4)            kicker : __mod_timer (process_timeout)
  24.5% ( 17.6)                 X : __mod_timer (process_timeout)
  10.9% (  7.8)            xemacs : __mod_timer (process_timeout)
   7.0% (  5.0)   <kernel module> : queue_delayed_work (delayed_work_timer_fn)
That's impressively quick decay, considering that I bought it 9 months ago, and it was pretty good until at least July. Perhaps the colder weather reduces its capacity, but the lower useful capacity is the same after I have kept the laptop indoors. Curiously the battery in my Nokia has also become much shorter recently. I think that there are very few commodities and the fairly rapid deterioration of the capacity of batteries is yet another differentiator that is important but not that obvious.
As to practical consequences, I bought the 6 cell battery because its 3h30m useful charge was rather more useful to me than the 2 hours charge of the 4 cell battery bundled with the laptop, as I often do travel between 1 and 2 hours, and 2 hours of battery capacity is a bit too short to give a margin of safety. It looks like that the useful life of the battery is going to be a bit short of 1 year, and that to me means it must be considered as a consumable, a bit like the toner or the OPC drum of laser printers. Costing around £100 with taxes, the battery ends up costing £8-10 per month, which is about the same as my phone bill.