[RCU] Speed-ups for a slow computer?
stan at hardwarefreak.com
Fri Aug 2 08:17:32 CEST 2013
On 8/2/2013 1:04 AM, Grant wrote:
>>>>>> Can anything be done to speed up Roundcube on a Celeron 700 with 512MB
>>>>>> RAM (maxed out) running Gentoo Linux? We're using the Chrome browser
>>>>>> and the CPU is 95% idle when no one is logged into X. I'm getting
>>>>>> reports of 30+ seconds to open an email.
>>>>> Your problem is likely the amount of RAM; when you start X and Chrome,
>>>>> your client almost surely starts swapping, because there isn't enough
>>>>> RAM for the running processes. Roundcubemail uses AJAX which requires a
>>>>> processes (including Chrome) are swapping.
>>>>> On such low-end systems you should use a very lightweight browser
>>>>> instead of Chrome, even if this can give some limitations.
>>>> This is simple to check. With all apps running simply run "free -m" and
>>>> post the output here.
>>> Here's what I get while someone is seated at the system and using
>>> Roundcube in Chrome:
>>> # free
>> Using the "-m" switch displays output in megabytes--easier to read.
>>> total used free shared buffers cached
>>> Mem: 513072 447812 65260 0 16896 209672
>>> -/+ buffers/cache: 221244 291828
>>> Swap: 2097148 0 2097148
>> This clearly shows the host isn't using swap, has 64MB free, and is
>> using ~200MB for buffer cache. So it seems to have plenty of memory.
>> JVMs and JAVA are notoriously inefficient, but that should cause a 30s
>> delay just opening a new email. Using HTTPS and JAVA together may be
>> enough to bring a Celery 700 to its knees. Try using straight HTTP
>> instead to eliminate the encryption. This will make a big difference
>> with binary attachments such as photos and the like. With small text
>> only emails it shouldn't cause a big delay.
> I'd rather stick with HTTPS if possible.
Of course. I'm simply recommending that you test with HTTP to see if
HTTPS is part of your delay issue.
> Are there any other good
> options for speeding things up besides switching browsers and
> disabling HTTPS?
There may be but we don't have a crystal ball. You're going to have to
do some testing and process of elimination to figure out the cause of
the problem. Only then can you identify a solution.
More information about the users